Intelligent transformer for power system

Abstract

The invention relates to the technical field of transformers, and discloses an intelligent transformer for a power system, and the intelligent transformer comprises a data receiving module which is used for collecting the transformer data of each transformer; the parameter analysis module is used for calculating comprehensive health indexes of each transformer; the collaborative analysis module is used for correcting the comprehensive health index of each transformer to obtain a health index; the abnormity positioning module is used for judging an abnormal transformer and the abnormity degree thereof; the early warning maintenance module is used for generating early warning information and a maintenance strategy; according to the invention, various operation parameters are comprehensively analyzed, a health index evaluation system is established, and the monitoring precision and reliability are improved; the equipment health state under the multi-transformer parallel operation environment is comprehensively evaluated by analyzing the circulating current and the loss between the transformers, and the intelligent transformer is maintained according to the health state.

Description

Technical Field

[0001] The present invention relates to the technical field of transformers, and more particularly to an intelligent transformer for a power system. Background Art

[0002] With the rapid development of power systems, transformers, as key energy conversion equipment in power systems, have a significant impact on the stability of the entire power grid due to their safe and reliable operation. Traditional transformers primarily perform voltage conversion and lack real-time monitoring, diagnosis, and early warning capabilities.

[0003] The existing Chinese patent with publication number CN116937815A discloses an intelligent maintenance system for power transformers, including: a control terminal, which is the main control end of the system and is used to issue execution commands; a receiving module, which is used to receive real-time operating data of the transformer; a loading module, which is used to load operating status safety parameters of the transformer service equipment; and an analysis module, which is used to receive the real-time operating data of the transformer received in the receiving module and the operating status safety parameters of the transformer service equipment loaded in the loading module. The invention can combine the received real-time operating data of the transformer with the loaded operating status safety parameters of the transformer service equipment to analyze and determine whether the operating status of the power transformer is safe, thereby achieving the purpose of maintaining the operating safety of the power transformer.

[0004] However, large data centers need to adopt an N+X redundant transformer architecture, which will result in multiple transformers being connected in parallel. The failure of a single transformer may lead to the risk of cascading downtime. Existing technologies only evaluate single machines and do not consider the impact of electrical coupling between multiple transformers. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent transformer for a power system in order to solve the above problems.

[0006] The present invention provides an intelligent transformer for a power system, comprising: A data receiving module, used for collecting transformer data of each transformer; The parameter analysis module calculates the load factor of each transformer based on the transformer data, identifies the current operating condition based on the load factor and transformer data, selects the corresponding safety threshold based on the operating condition type, and calculates the comprehensive health index of each transformer; Collaborative analysis module analyzes the coupling effect between parallel transformers and corrects the comprehensive health index of each transformer to obtain the health index; The abnormality location module analyzes the health indicators of multiple transformers and compares them with safety thresholds to identify abnormal transformers and the degree of abnormality. The early warning maintenance module generates early warning information and maintenance strategies based on abnormal transformers and their abnormality levels.

[0007] Furthermore, the specific steps of selecting the corresponding safety threshold include: Step 201, obtaining a load factor by calculating a ratio of actual apparent power to a rated capacity of a transformer in transformer data, where the actual apparent power is the square root of the sum of the squares of the active power in the transformer data and the reactive power in the transformer data; Step 202: Identify the current transformer operating condition based on the load factor and the harmonic content in the transformer data; Step 203: Select a corresponding safety threshold according to the identified working condition type.

[0008] Furthermore, the calculation method of the comprehensive health index of the transformer includes: Calculate the hotspot temperature rise coefficient, which is the ratio of the winding temperature in the transformer data to the rated winding temperature; Use the Arrhenius equation to calculate the life consumption rate. Set the default values ​​of the activation energy-related parameters and reference temperature in the Arrhenius equation according to the transformer's implementation standards. Convert the winding temperature from Celsius to Kelvin, input it into the Arrhenius equation, and output the life consumption rate. The activation energy-related parameter is the ratio of the activation energy to the Boltzmann constant. The weighted sum of the hot spot temperature rise coefficient, life consumption rate, and the ratio of the current in the transformer data to the rated current of each transformer is calculated. The weighted sum is expressed as the comprehensive health index of each transformer.

[0009] Furthermore, the specific steps of the collaborative analysis module include: Step 301, calculating the absolute value of the current difference between two adjacent transformers to obtain the circulating current value; Step 302 , evaluating the circulating current loss by heat loss generated by the circulating current value on the equivalent resistance of the transformer, wherein the circulating current loss is the square of the circulating current value times the number of phases multiplied by the equivalent resistance in the transformer data, and the number of phases multiplied is related to the number of phases of the transformer; Step 303: The health index is equal to the comprehensive health index plus a weighted value of the ratio of circulating current loss to rated power.

[0010] Furthermore, the steps of the anomaly location module include: Step 401, calculating the mean and standard deviation of all transformer health indicators to obtain the group mean and group standard deviation; Step 402: Determine whether each transformer is abnormal. If any condition is met, it is determined to be an abnormal transformer. The conditions include: Calculate the degree of deviation of the transformer's health index from the group mean. When the degree of deviation exceeds the preset multiple of the group standard deviation; When the health indicator of the transformer exceeds the safety threshold under the current working conditions; Step 403: For the abnormal transformer identified, the degree of abnormality is quantified using the Mahalanobis distance method.

[0011] Furthermore, methods for quantifying the degree of abnormality using the Mahalanobis distance method include: Construct a parameter vector for each transformer. The parameter vector includes the following: health indicators, oil temperature, winding temperature, load factor, and circulating current percentage from transformer data. The circulating current percentage is the ratio of the circulating current value to the rated current. Standardize all parameters of the parameter vector to obtain a standardized parameter vector. For the normal transformer group, the mean vector and covariance matrix of the normal transformer group are calculated based on the standardized parameter vector, and the Mahalanobis distance of each abnormal transformer is calculated based on the mean vector and covariance matrix. That is, the difference between the standardized parameter vector of the abnormal transformer and the mean vector of the normal transformer group is weighted by the inverse matrix of the covariance matrix, and the square root is taken to obtain the final Mahalanobis distance value. The Mahalanobis distance value represents the degree of abnormality, where the normal transformer group is the set of transformers that have not been judged as abnormal.

[0012] Furthermore, the elements in the covariance matrix include the covariances between the parameters, and the covariances between the parameters include: health index and oil temperature, health index and winding temperature, health index and load rate, health index and circulation percentage, oil temperature and winding temperature, oil temperature and load rate, oil temperature and circulation percentage, winding temperature and load rate, winding temperature and circulation percentage, and load rate and circulation percentage.

[0013] Furthermore, the specific steps of the early warning maintenance module include: Based on the degree of abnormality and the number of abnormal transformers, combined with the number of redundant transformers, the warning level is divided into four levels: normal, low-level warning, medium-level warning, and high-level warning; For different warning levels, maintenance strategies are generated by combining the Mahalanobis distance value and the warning level. The maintenance strategies include: When the warning level is normal, no action is taken; When the warning level is low, reduce the load of the abnormal transformer until the warning level is reduced to normal; When the warning level is medium, load transfer is implemented to distribute the load of the abnormal transformer to the normal transformer until the warning level is reduced to low. When the warning level is high, load switching is performed to transfer the load from the abnormal transformer to the normal transformer or redundant transformer; and the abnormal transformer is isolated.

[0014] Furthermore, when all transformers are normal or the Mahalanobis distance values ​​of all abnormal transformers are less than the minimum value for the low-level warning, and the transformer group maintains full redundancy, the warning level is determined to be normal; full redundancy is the N+X structure where X remains unchanged, where N represents the minimum number of transformers required for normal operation and X represents the number of redundant transformers; When the Mahalanobis distance value of any abnormal transformer is between the minimum value and the maximum value of the low-level warning, the warning level is low-level warning; When the Mahalanobis distance value of any abnormal transformer is between the minimum value of the intermediate warning and the maximum value of the intermediate warning, or the Mahalanobis distance values ​​of two or more abnormal transformers exceed the minimum value of the low-level warning at the same time, or X=1, the warning level is intermediate warning; When the Mahalanobis distance value of any abnormal transformer exceeds the maximum value of the intermediate warning, or the Mahalanobis distance values ​​of two or more abnormal transformers simultaneously exceed the minimum value of the intermediate warning, or X=0, the warning level is high warning; The minimum value of the low-level warning is less than the maximum value of the low-level warning, the maximum value of the low-level warning is equal to the minimum value of the intermediate warning, and the minimum value of the intermediate warning is less than the maximum value of the intermediate warning; If multiple warning levels are met at the same time, the highest warning level will be output.

[0015] Furthermore, the transformer data includes real-time operating parameters, rated parameters and safety thresholds, and the safety thresholds include light-load operating condition thresholds, full-load operating condition thresholds, harmonic environment thresholds and normal operating condition thresholds.

[0016] The beneficial effects of the present invention are: by comprehensively analyzing multiple operating parameters, a health indicator evaluation system is established to comprehensively evaluate the operating status of the transformer and improve monitoring accuracy and reliability; by analyzing the circulating current and its loss between transformers, the health status of the equipment in the parallel operation environment of multiple transformers is comprehensively evaluated.

[0017] It can intelligently identify transformer operating conditions based on load rate and harmonic content, select corresponding safety thresholds, and improve monitoring accuracy. It uses statistical analysis methods such as Mahalanobis distance to identify abnormal transformers based on group data analysis, thereby improving the accuracy of fault location. It provides multi-level early warning and differentiated maintenance strategies based on the degree of abnormality and redundant architecture characteristics to optimize maintenance resource allocation. It provides load optimization suggestions for N+X redundant transformer architectures, ensuring reliability while improving energy efficiency. It automatically extracts complex features through deep learning methods to achieve high-precision fault diagnosis and predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is an example of a module of an intelligent transformer for a power system according to the present invention. Figure 1 ; Figure 2 This is an example diagram of a parameter analysis module for an intelligent transformer used in a power system according to the present invention; Figure 3 This is an example diagram of an abnormality location module for an intelligent transformer in a power system according to the present invention; Figure 4 This is an example of a module of an intelligent transformer for a power system according to the present invention. Figure 2 . DETAILED DESCRIPTION

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0020] Example 1: Smart transformers for power systems, such as Figure 1 Shown, including: The data receiving module 101 is used to collect transformer data of each transformer, where the transformer data includes real-time operating parameters, rated parameters and safety thresholds.

[0021] The operating parameters specifically include: Voltage, which is the voltage value of the primary and secondary sides of the transformer; Current, which is the current value of the primary and secondary sides of the transformer; Active power; Reactive power; Winding temperature, which is the actual temperature of the transformer winding; Oil temperature, which is the actual temperature of the transformer oil; Harmonic content, which is the harmonic distortion rate of the power grid; The rated parameters of the transformer include: rated capacity, rated voltage, rated current, rated frequency, rated winding temperature, and rated power equivalent resistance; And the safety thresholds under each working condition: light load condition threshold, full load condition threshold, harmonic environment threshold and normal working condition threshold.

[0022] After preliminary verification, the collected data is transmitted to the parameter analysis module 102 for further processing.

[0023] The parameter analysis module 102 calculates the load rate of each transformer based on the collected transformer data, identifies the current working condition based on the load rate and transformer data, selects the corresponding safety threshold based on the working condition, and finally calculates the comprehensive health index of each transformer.

[0024] Specific as Figure 2 Shown, including: Step 201: Load factor is a key indicator for measuring transformer load levels. The load factor is calculated by calculating the ratio of actual apparent power to the transformer's rated capacity. Specifically, the process involves first obtaining the transformer's active power and reactive power from the data receiving module 101, then calculating the actual apparent power (the square root of the sum of the squares of the active power and the reactive power), and finally dividing the actual apparent power by the transformer's rated capacity to obtain the load factor.

[0025] The load factor is a dimensionless ratio that reflects the current load level of the transformer. A higher load factor indicates that the transformer load is closer to its rated capacity; a lower load factor indicates that the transformer load level is lighter.

[0026] Step 202: Based on the calculated load factor and the harmonic content in the transformer data, identify the current transformer operating condition type: When the load rate is less than the light load threshold, it is determined to be a light load condition; When the load rate is greater than the full load threshold, it is determined to be a full load condition; When the total harmonic distortion rate is greater than the harmonic threshold, it is determined to be a harmonic environment condition; When the load rate is greater than the light load threshold and less than the full load threshold, and the total harmonic distortion rate is less than the harmonic threshold, it is determined to be a normal operating condition.

[0027] These thresholds are initially set based on power industry standards and transformer characteristics. Specifically, the default value for the light-load threshold is 0.3, based on engineering practices and traditional standards. The default value for the full-load threshold is 0.7, reflecting the upper limit of the transformer's long-term safe operation load. The default value for the harmonic threshold is 5%, based on industry standards for power system harmonic limits. In practice, these thresholds are fine-tuned based on specific transformer models and application scenarios.

[0028] The operating condition identification results will be passed to subsequent modules for dynamic safety threshold selection and health indicator calculation, ensuring accurate assessment and early warning based on the actual operating status of the transformer.

[0029] Step 203, select the corresponding safety threshold according to the identified working condition: under light load conditions, the safety threshold is the light load condition threshold; under full load conditions, the safety threshold is the full load condition threshold; under harmonic environment conditions, the safety threshold is the harmonic environment threshold; under normal conditions, the safety threshold is the normal condition threshold; for complex working conditions, such as when both full load and harmonic environment conditions are met, the most stringent safety threshold is adopted, and the safety threshold is the minimum value of each threshold.

[0030] Dynamically selecting the safety threshold can adapt to the operating characteristics of the transformer under different working conditions and improve the accuracy of monitoring.

[0031] Step 204: The transformer comprehensive health index is a comprehensive indicator used to assess the operating status of each transformer. It involves a weighted calculation of multiple parameters. First, the hotspot temperature rise coefficient (HTRC) is calculated: the ratio of the winding temperature to the rated winding temperature. The HTRC is a dimensionless value. When it approaches or exceeds 1, it indicates that the transformer temperature is approaching or exceeding the design limit, potentially posing an overheating risk.

[0032] Next, the life consumption rate is calculated to assess the aging rate of the transformer insulation material at the current temperature. This calculation is based on the Arrhenius equation, which describes the effect of temperature on chemical reaction rates. During the calculation, default values ​​for the activation energy parameters and reference temperature in the Arrhenius equation are set according to the transformer's implementation standard. The activation energy parameter, representing the ratio of the activation energy to the Boltzmann constant, is a key parameter in the Arrhenius equation. The winding temperature is also converted from degrees Celsius to degrees Kelvin. The life consumption rate is a dimensionless value. When it is 1, the insulation material ages at the current winding temperature at the same rate as at the reference temperature. A value greater than 1 indicates accelerated aging, and a value less than 1 indicates reduced aging. The activation energy parameters and reference temperature are set based on the recommended values ​​for the relationship between transformer hotspot temperature and life in the IEEE C57.91-2011 standard. The default values ​​for the activation energy parameters are 15000K, and the default value for the reference temperature is 383K (110°C).

[0033] A comprehensive health index is calculated for each transformer. This index takes into account the weighted sum of three factors: the hotspot temperature rise coefficient, the life consumption rate, and the ratio of current to rated current. Each of these three factors is assigned a different weighting coefficient to reflect its impact on the transformer's health. The weighting coefficients are initially set based on historical operating data analysis. Typically, the hotspot temperature rise coefficient is weighted at 0.4, the life consumption rate at 0.3, and the current ratio at 0.3. These weights can be dynamically adjusted based on actual operating data.

[0034] The comprehensive health index is a dimensionless value, typically between 0 and 10, with larger values ​​indicating that the transformer condition deviates further from the normal range. The calculated comprehensive health index of each transformer is passed to the collaborative analysis module 103 for further correction.

[0035] The collaborative analysis module 103 mainly analyzes the coupling effect between parallel transformers and modifies the comprehensive health index of each transformer to obtain the health index. In the transformer redundancy architecture, the mutual influence between transformers is a factor that cannot be ignored.

[0036] In step 301, when transformers are operated in parallel, circulating currents are generated between them due to slight differences in parameters such as impedance and voltage. The circulating current value is calculated by calculating the absolute value of the current difference between two adjacent transformers. The presence of circulating currents can cause additional losses and temperature rise in the transformers, making it an important parameter for evaluating the operating status of parallel transformers.

[0037] Step 302 evaluates the circulating current loss by calculating the heat loss caused by the circulating current on the transformer's equivalent resistance. The circulating current loss is calculated as the square of the circulating current value multiplied by the equivalent resistance in the transformer data. The phase multiple is determined based on the number of phases in the transformer. For example, for a three-phase transformer, the phase multiple is 3. The equivalent resistance value typically ranges from 0.01Ω to 0.1Ω, depending on the transformer's capacity and design characteristics. The calculated circulating current loss is used to assess the impact of the circulating current on the transformer's operating efficiency and lifespan.

[0038] Step 303: After accounting for circulating current, the comprehensive health index of each transformer is corrected. The resulting health index is equal to the comprehensive health index plus the weighted value of the ratio of circulating current loss to rated power. The weighting factor for the ratio of circulating current loss to rated power is initially set to 0.2, determined based on analysis of historical operating data, and reflects the impact of circulating current loss on the transformer's health. The corrected comprehensive health index is used as the health index.

[0039] The revised comprehensive health index more comprehensively reflects the actual status of the transformer in the parallel operation environment and is passed to the anomaly detection and fault location module 104 for analysis. In this way, transformers that may be at potential risk due to circulating current influence can be identified, even if their individual parameters are within normal range. Failure to identify the circulating current influence may result in the following consequences: The additional losses caused by circulating current will accelerate the aging of transformer insulation and be ignored, resulting in a significant shortening of transformer life; The heat generated by the circulating current may cause local hot spots, which may lead to insulation breakdown failure if they exist for a long time. When the load changes dynamically, the circulating current may suddenly increase, causing the protection device to malfunction or refuse to operate; Circulation loss will reduce overall energy efficiency and increase operating costs; Failure to detect circulating current problems in a timely manner may result in missing the optimal maintenance opportunity, ultimately leading to catastrophic failure of the transformer and affecting the reliability and safety of the entire power system.

[0040] The abnormality location module 104 performs group data analysis on the health indicators of multiple transformers to determine abnormal transformers and the degree of abnormality. In a multi-transformer system, abnormal equipment can be identified more accurately through comparative analysis.

[0041] Specific as Figure 3 Shown, including: Step 401 first calculates the average health index of all transformers to obtain the group mean. This is done by adding the health indexes of all transformers and dividing it by the total number of transformers. This average represents the overall health of the transformer group under the current operating conditions and serves as an important reference for determining whether a single transformer is abnormal.

[0042] Next, the standard deviation of the health indicators is calculated to obtain the group standard deviation, which is the square root of the average of the sum of the squares of the differences between each transformer's health indicator and the group mean. The standard deviation reflects the degree of dispersion in the health status of the transformer group. A smaller standard deviation indicates closer health status among the transformers, while a larger standard deviation indicates greater differences in health status between transformers.

[0043] These statistics are calculated based on data from all online transformers and are typically updated every 5-15 minutes to ensure they reflect the dynamic changes in the transformer population. The results are stored in the system database and used both for current anomaly determination and as historical data for future analysis.

[0044] Step 402: Determine whether each transformer is abnormal based on two conditions: The first condition is based on statistical principles. The deviation of the transformer health indicator from the population mean is calculated. When the deviation exceeds a preset multiple of the population standard deviation (the default multiple is 2), the threshold is considered statistically significant. This determination is based on normal distribution theory. In a normal distribution, the probability of data falling outside the range of ±2 standard deviations of the mean is approximately 5%, so these data points are considered statistically significant outliers. A threshold of 2 standard deviations is used as the initial threshold, which can be adjusted through system settings and is typically between 1.5 and 3, depending on the user's desired sensitivity to outliers.

[0045] The second condition is based on engineering practice principles. This occurs when the transformer's health indicators exceed the safety threshold under the current operating conditions. Safety thresholds are pre-set based on the transformer model, operating environment, and industry standards, and are dynamically adjusted by the threshold adjustment module. Initial safety thresholds are typically set based on the transformer manufacturer's technical specifications and power industry standards. The light-load threshold is approximately 7.5, the full-load threshold is approximately 8.5, the harmonic threshold is approximately 8.0, and the normal threshold is approximately 8.0.

[0046] Using "OR" logic, any condition that meets it is considered an anomaly. This design enhances the system's sensitivity and enables it to capture different types of anomalies. The judgment result is stored as a Boolean value and passed to the anomaly source location step. The Boolean value can be considered normal or abnormal.

[0047] Step 403: For the transformer determined to be abnormal, the degree of abnormality is further quantified using the Mahalanobis distance method.

[0048] The Mahalanobis distance is a multidimensional distance calculation method that accounts for intervariate correlations. It treats the multiple parameters of each transformer as a point in multidimensional space and calculates the distance from that point to the center of the normal transformer population. Unlike the simple Euclidean distance, the Mahalanobis distance considers the intervariate correlations between parameters through the covariance matrix, enabling more accurate identification of outliers in multidimensional space.

[0049] In actual calculations, we first construct a parameter vector for each transformer. The parameter vector includes health indicators, oil temperature, winding temperature, load factor, and circulating current percentage. Specifically, it includes the following elements for multiple transformers: Health index, reflecting the comprehensive health status of the transformer; load rate, the ratio of actual load to rated capacity; circulating current percentage, the ratio of circulating current value to rated current; For example, the parameter vector of a transformer is [7.2, 65°C, 85°C, 0.85, 8%], which means that its health index is 7.2, the oil temperature is 65°C, the winding temperature is 85°C, the load factor is 85%, and the circulating current is 8% of the rated current.

[0050] Since these parameters have different dimensions and numerical ranges, direct calculation will cause the parameters with larger values ​​(such as temperature) to dominate the covariance matrix, while the influence of parameters with smaller values ​​(such as load rate) will be weakened. To solve this problem, all parameters are standardized before calculating the Mahalanobis distance to obtain a standardized parameter vector: Z-score normalization is applied to each parameter, subtracting the original parameter value from the mean of the parameter in the normal transformer population and dividing it by its standard deviation.

[0051] After normalization, all parameters are converted to dimensionless values ​​with similar numerical ranges, ensuring that each parameter has the same weight in subsequent calculations.

[0052] The normal transformer group is a set of transformers that are not judged to be abnormal.

[0053] For a normal transformer population, calculate its mean vector and covariance matrix: The mean vector is the arithmetic mean of the normalized parameter vectors of all normal transformers. For example, if there are 10 normal transformers, the normalized parameters of these 10 transformers are averaged to obtain the mean vector.

[0054] The covariance matrix is ​​obtained by calculating the covariance between each standardized parameter. The diagonal elements represent the variance of each parameter, and the off-diagonal elements represent the covariance between parameters, reflecting the correlation between the parameters.

[0055] In this embodiment, the covariances between the parameters included in the covariance matrix include: health index and oil temperature, health index and winding temperature, health index and load rate, health index and circulation percentage, oil temperature and winding temperature, oil temperature and load rate, oil temperature and circulation percentage, winding temperature and load rate, winding temperature and circulation percentage, and load rate and circulation percentage.

[0056] If there are 10 normal transformers, each with five parameters (health index, oil temperature, winding temperature, load factor, and circulating current percentage), the covariance matrix is ​​a 5×5 matrix. The (i, j)th element of this matrix represents the covariance between the i-th and j-th standardized parameters, calculated based on the data from these 10 transformers. The covariance matrix is ​​updated every 15 minutes to adapt to dynamic changes in the transformer's operating status.

[0057] After obtaining the mean vector and covariance matrix, the Mahalanobis distance of each abnormal transformer is calculated based on the mean vector and covariance matrix. That is, the difference between the standardized parameter vector of the abnormal transformer and the mean vector of the normal transformer group is weighted by the inverse matrix of the covariance matrix, and finally the square root is taken to obtain the final Mahalanobis distance value. The weighted calculation by the inverse matrix of the covariance matrix is ​​to use the inverse matrix of the covariance matrix as a weight to adjust the contribution of different parameters and suppress the interference of parameters with high correlation or small variance.

[0058] The result of the Mahalanobis distance calculation is a dimensionless value, usually in the range of 0-10, but theoretically there is no upper limit. The larger the value, the higher the degree of abnormality.

[0059] The anomaly detection results are passed to the threshold adjustment module and the early warning maintenance module 105 for subsequent parameter optimization and early warning generation. The anomaly detection results are also recorded in the historical database for long-term trend analysis and model optimization.

[0060] The early warning maintenance module 105 generates differentiated early warning information and maintenance strategies based on abnormal transformers and their abnormality levels, targeting the characteristics of redundant transformer architecture, to ensure system reliability while optimizing maintenance resource allocation.

[0061] Specifically include: Step 501: Based on the Mahalanobis distance value (abnormality level) and the number of abnormal transformers, combined with the characteristics of the redundant transformer architecture, the warning level is divided into four levels: normal, low warning, medium warning, and high warning: When all transformers are normal or the Mahalanobis distance values ​​of all abnormal transformers are less than the minimum value for the low-level warning, and the transformer group maintains full redundancy (X in the N+X architecture remains unchanged), the warning level is determined to be normal; When the Mahalanobis distance value of any abnormal transformer is between the low-level warning minimum value and the low-level warning maximum value, the default values ​​of the low-level warning minimum value and the low-level warning maximum value are 3.0 and 5.0 respectively, or when the redundancy is slightly reduced, which means that X in the N+X architecture is reduced by 1, but X ≥ 2, the warning level is low warning; When the Mahalanobis distance value of any abnormal transformer is between the minimum value of the intermediate warning and the maximum value of the intermediate warning (the default values ​​of the minimum and maximum values ​​of the intermediate warning are 5.0 and 7.0 respectively), or when the Mahalanobis distance values ​​of two or more abnormal transformers exceed the minimum value of the low-level warning at the same time, or when the redundancy is significantly reduced, the redundancy is significantly reduced to X=1 in the N+X architecture, and the warning level is medium warning; When the Mahalanobis distance value of any abnormal transformer exceeds the maximum value of the intermediate warning, or the Mahalanobis distance values ​​of two or more abnormal transformers simultaneously exceed the minimum value of the intermediate warning, or when the redundancy completely disappears, the redundancy completely disappears, which means X=0 in the N+X architecture, and the warning level is advanced warning.

[0062] If multiple warning levels are met at the same time, the highest warning level will be output; the determination of the warning level comprehensively considers the abnormality degree of a single transformer, the number of abnormal transformers and the redundancy status of the entire transformer group.

[0063] In the N+X architecture, N represents the minimum number of transformers required for normal operation, and X represents the number of redundant transformers. For example, an N+2 architecture means that the system requires N transformers for normal operation, and two redundant transformers are provided, for a total of N+2 transformers. This architecture ensures that the system can maintain normal operation even if some transformers fail.

[0064] Step 502: For different warning levels, differentiated maintenance strategies are automatically generated based on the Mahalanobis distance value and the redundant architecture characteristics. The maintenance strategies include: When the warning level is normal, perform routine monitoring and scheduled maintenance as planned, and no special treatment is required.

[0065] When the warning level is low, optimize load distribution and reduce the load on the abnormal transformer. The load reduction ratio is proportional to the Mahalanobis distance value, which is three times the Mahalanobis distance value divided by 100, usually 10%-20%. Arrange non-emergency maintenance, which can be performed during the planned maintenance window. It is recommended to complete the maintenance within 30 days without affecting the normal operation of the system. Prepare spare parts and update the maintenance plan. When the warning level is medium, load transfer will be implemented to distribute the load of the abnormal transformer to the normal transformer or redundant transformer; priority maintenance will be arranged, and it is recommended to be completed within 7 days; the standby transformer will be started to restore system redundancy.

[0066] When the warning level is high, load switching should be performed immediately to transfer critical loads from the abnormal transformer; emergency maintenance should be arranged, and it is recommended to be handled within 24 hours; the abnormal transformer should be isolated to prevent cascading failures; the emergency plan should be activated and temporary equipment should be deployed to ensure power supply reliability.

[0067] Step 503: Provide load optimization suggestions based on the Mahalanobis distance value for the redundant transformer architecture to improve energy efficiency while ensuring reliability: When the Mahalanobis distance values ​​of all abnormal transformers are lower than 2.0, it is recommended to implement a load balancing strategy to make the load rates of each transformer close and extend the overall service life.

[0068] When an abnormal transformer's Mahalanobis distance exceeds 3.0, a progressive load reduction strategy is implemented based on the extent to which the Mahalanobis distance exceeds the threshold, ensuring the abnormal transformer operates within a safe range. Specifically, a 10% load reduction is recommended for every 1.0 point above the threshold.

[0069] Evaluate the current redundant capacity in real time and use the Mahalanobis distance value to predict the possible risk of insufficient capacity.

[0070] During low-load periods, it is recommended to appropriately concentrate the load so that some transformers operate at full load, and only select transformers with Mahalanobis distance values ​​lower than 1.5. The remaining transformers should reduce their load or be in standby mode to improve overall energy efficiency.

[0071] For example, in an N+2 architecture, if a transformer is detected with a Mahalanobis distance value of 4.5, it is recommended to reduce the load of the transformer by 15% and distribute the load to the healthy transformer with the lowest Mahalanobis distance value. It is also recommended to schedule maintenance during the next maintenance window.

[0072] This redundant architecture performance evaluation based on Mahalanobis distance values ​​and historical data can continuously optimize early warning and maintenance strategies, and improve the reliability and economy of the entire transformer group.

[0073] Example 2: Based on Example 1, this example introduces a deep learning method to improve the accuracy and prediction ability of transformer health status assessment, such as Figure 4 As shown, including: Data receiving module 601 is similar to that in Example 1 and is used to collect transformer data of each transformer. In addition to the parameters mentioned in Example 1, this embodiment also adds the following real-time operating parameters: Partial discharge data, an indicator of insulation degradation; Transformer vibration data, an indicator of the transformer's mechanical condition; Gas analysis data refers to the dissolved gas content in oil, including H2, CH4, C2H2, etc.

[0074] Data sources include transformer-built-in sensors, online monitoring equipment, and periodic manual sampling and analysis. Data is collected in real time via industrial Ethernet, MODBUS, or wireless sensor networks. Sampling frequency is set based on parameter importance. Key parameters like temperature may be sampled every five minutes, while gas analysis data may be updated daily.

[0075] The data processing module 602 processes the collected transformer data to obtain time window data; the time window data improves the data quality and provides high-quality input for the fault prediction model.

[0076] Mainly include: Outlier detection and processing are performed on the raw transformer data, removing or correcting anomalous data points to generate corrected data. Outlier detection uses a method based on mean and standard deviation. Data points that deviate from the mean by more than three standard deviations are marked as outliers. These outliers are corrected using methods such as linear interpolation or neighbor replacement.

[0077] Initial threshold settings are based on statistical analysis of historical data. For example, the threshold for determining abnormality for temperature parameters is initially set at a value exceeding three standard deviations from the mean. For parameters with larger fluctuations, such as load current, the threshold may be set at 3.5 standard deviations. These thresholds are automatically adjusted quarterly based on actual operating conditions.

[0078] Normalizing data from different dimensions to the same scale facilitates model training. Normalization uses a min-max normalization method to map each parameter data to a range of 0 to 1. For example, temperature data may range from -20°C to 150°C, which is then mapped to a range of 0 to 1 through normalization.

[0079] Normalization parameters are determined based on the device's technical specifications. For example, the minimum and maximum temperatures are based on the device's minimum and maximum operating temperatures, respectively. These normalization parameters are updated regularly to adapt to changes in data distribution.

[0080] The continuous time series data in the normalized data is divided into time windows of fixed length to obtain time window data, which serves as the input for the fault prediction model. Data from the most recent period (e.g., 24 hours) is organized into time windows. The window length is set based on the changing characteristics of different parameters. Fast-changing parameters such as temperature may use a 4-hour window, while slowly changing parameters such as gas content in oil may use a 7-day window.

[0081] The sliding step size of the time window is typically set to 25% of the window length to ensure sufficient overlap between adjacent windows to capture continuously changing trends. The window size and sliding step size can be dynamically adjusted through the parameter optimization module to achieve optimal performance.

[0082] The fault prediction module 603 is the core of this embodiment; a fault prediction model is constructed, the fault prediction model inputs time window data, and outputs health indicator prediction values ​​and fault type probability distribution.

[0083] This example uses a hybrid model that combines a long short-term memory network (LSTM) with a convolutional neural network (CNN). It includes the following layers: Input layer, receiving time window data; Convolutional layer, using one-dimensional convolution to extract local features; LSTM layer, which captures temporal features and has 128 hidden units; Fully connected layer, mapping the output of the LSTM layer to the final prediction result; The output layer outputs the health indicator prediction value and fault type probability distribution. The health indicator prediction value is a value between 0 and 1, indicating the overall health status of the transformer. The closer the value is to 1, the better the health status. The fault type probability distribution is the probability value of four main fault types, including insulation aging, overheating fault, mechanical fault and circulating current loss. The probability value of each fault type is between 0 and 1, and the sum of the four probabilities is 1.

[0084] The model is implemented using the TensorFlow framework and deployed on edge computing devices to ensure real-time responsiveness. The convolutional layer uses the ReLU activation function to extract local features, the LSTM layer captures long-term dependencies, and the fully connected layer maps features to prediction results.

[0085] Model training uses a composite loss function that considers both health indicator prediction error and fault type classification error. The health indicator prediction loss uses mean squared error (MSE), which is the average of the squares of the difference between the predicted and true values. The fault type classification loss uses cross-entropy loss, which is the cross entropy between the true label and the predicted probability distribution.

[0086] The total loss function is a weighted combination of the health indicator prediction loss and the fault type classification loss. The initial weight coefficient for the health indicator prediction loss is set to 0.7, and the initial weight coefficient for the fault type classification loss is set to 0.3. The weight ratio of the two losses is initially set to 7:3, and this ratio will be dynamically adjusted based on the actual prediction results.

[0087] The weights of the loss function are initially set based on expert experience and optimized using a grid search method on the validation set. These weights are re-evaluated quarterly using the latest data to ensure optimal model performance.

[0088] The model is trained using mini-batch stochastic gradient descent with a batch size of 64, an initial learning rate of 0.001, and 100 training rounds. To prevent overfitting, L2 regularization and early stopping are used.

[0089] Training data sources include historical operating data, simulation data, and publicly available industry datasets. The initial model is pre-trained on a comprehensive dataset encompassing various fault scenarios and then fine-tuned using historical data from specific transformers. The model is updated monthly with newly collected data to ensure it adapts to equipment aging and environmental changes.

[0090] The fault diagnosis module 604 performs a comprehensive diagnosis of the health status of the transformer based on the output of the fault prediction model and combines domain knowledge to obtain the severity of the fault.

[0091] The fault diagnosis module 604 can identify the following typical fault types: Insulation aging, mainly manifested by increased partial discharge and abnormal dissolved gas in the oil; Overheating failure, mainly manifested as abnormal hot spot temperature and increased oil temperature; Mechanical failure, mainly manifested as abnormal vibration and increased noise; Circulating current loss is mainly manifested as current imbalance and abnormal temperature rise between transformers.

[0092] The probability distribution of fault types is calculated using the Softmax function in the model's output layer, with the fault type with the highest probability being selected as the diagnosis result. The initial fault identification threshold is set at 0.6, meaning that when the probability of a particular fault type exceeds 60%, it is considered a fault of that type. This threshold can be adjusted based on actual operational experience and is typically between 0.5 and 0.7.

[0093] The fault signature database is based on power industry standards and historical fault cases, and contains typical characteristic patterns for various fault types. A fuzzy matching algorithm is used to compare the current status with the fault signature database to improve diagnostic accuracy.

[0094] Fault severity assessment is based on health indicators and fault probability, and the fault severity index is calculated. The assessment process comprehensively considers health indicators and the maximum fault type probability.

[0095] Fault severity is divided into four levels: minor fault, moderate fault, severe fault, and critical fault, among which: The fault severity index interval of a minor fault is [0, 0.3); the fault severity index interval of a moderate fault is [0.3, 0.6); the fault severity index interval of a severe fault is [0.6, 0.8); and the fault severity index interval of a critical fault is [0.8, 1.0].

[0096] Severity thresholds are based on power industry standards and equipment manufacturer recommendations, and are adjusted based on the criticality of the equipment. The thresholds for critical transformers may be more stringent, for example, with the lower limit of the "serious fault" level adjusted to 0.55.

[0097] The early warning control module 605 generates early warning information and control suggestions of different levels based on the severity of the fault.

[0098] Based on the severity of the fault, the warning levels are divided into low-level warning, medium-level warning, high-level warning, and emergency warning; among them, low-level warning corresponds to minor faults, prompting the operation and maintenance personnel to pay attention; medium-level warning corresponds to medium faults, and it is recommended to arrange maintenance; high-level warning corresponds to serious faults, and it is recommended to repair immediately; emergency warning corresponds to critical faults, and it is recommended to shut down immediately for processing.

[0099] Warning thresholds are initially set based on industry standards and adjusted based on the criticality of the equipment and the operating environment. For example, for transformers supporting critical operations, the trigger threshold for a medium-level warning might be lowered from 0.3 to 0.25, enabling earlier warnings. These thresholds are regularly evaluated and adjusted based on historical warning accuracy.

[0100] Specific control recommendations are generated for different fault types and severities. For example, for insulation aging, it is recommended to check the insulation condition and replace the insulation material if necessary; for overheating, it is recommended to reduce the load and increase cooling; for mechanical failures, it is recommended to check fasteners and identify vibration sources; for circulating current loss, it is recommended to check and adjust the transformer impedance matching.

[0101] Control recommendations are generated based on an expert knowledge base containing troubleshooting methods and best practices for various fault types. Using case-based reasoning, we extract solutions applicable to current situations from historical success stories. The expert knowledge base is updated quarterly to incorporate new experience and technological advancements.

[0102] This embodiment significantly improves the accuracy of transformer health status assessment and fault prediction capabilities by introducing deep learning methods. Compared with Example 1, this embodiment has the following advantages: Powerful feature extraction capabilities enable the fault prediction model to automatically extract complex timing features and correlations between parameters; high-precision fault diagnosis can identify multiple fault types and assess fault severity; and predictive maintenance can predict potential faults in advance, providing support for preventive maintenance.

[0103] This embodiment is particularly suitable for critical places with extremely high requirements on the operating status of transformers, such as large data centers, financial institutions, medical facilities, etc.

[0104] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. An intelligent transformer for a power system, characterized in that: include: A data receiving module, used for collecting transformer data of each transformer; The parameter analysis module calculates the load factor of each transformer based on the transformer data, identifies the current operating condition based on the load factor and transformer data, selects the corresponding safety threshold based on the operating condition type, and calculates the comprehensive health index of each transformer; Collaborative analysis module analyzes the coupling effect between parallel transformers and corrects the comprehensive health index of each transformer to obtain the health index; The abnormality location module analyzes the health indicators of multiple transformers and compares them with safety thresholds to identify abnormal transformers and the degree of abnormality. The early warning maintenance module generates early warning information and maintenance strategies based on abnormal transformers and their abnormality levels.

2. The intelligent transformer for a power system according to claim 1, characterized in that: The specific steps for selecting the appropriate safety threshold include: Step 201, obtaining a load factor by calculating a ratio of actual apparent power to a rated capacity of a transformer in transformer data, where the actual apparent power is the square root of the sum of the squares of the active power in the transformer data and the reactive power in the transformer data; Step 202: Identify the current transformer operating condition based on the load factor and the harmonic content in the transformer data; Step 203: Select a corresponding safety threshold according to the identified working condition type.

3. The intelligent transformer for a power system according to claim 1, characterized in that: The calculation method of the comprehensive health index of the transformer includes: Calculate the hotspot temperature rise coefficient, which is the ratio of the winding temperature in the transformer data to the rated winding temperature; Use the Arrhenius equation to calculate the life consumption rate. Set the default values ​​of the activation energy-related parameters and reference temperature in the Arrhenius equation according to the transformer's implementation standards. Convert the winding temperature from Celsius to Kelvin, input it into the Arrhenius equation, and output the life consumption rate. The activation energy-related parameter is the ratio of the activation energy to the Boltzmann constant. The weighted sum of the hot spot temperature rise coefficient, life consumption rate, and the ratio of the current in the transformer data to the rated current of each transformer is calculated. The weighted sum is expressed as the comprehensive health index of each transformer.

4. The intelligent transformer for a power system according to claim 1, characterized in that: The specific steps of the collaborative analysis module include: Step 301, calculating the absolute value of the current difference between two adjacent transformers to obtain the circulating current value; Step 302 , evaluating the circulating current loss by heat loss generated by the circulating current value on the equivalent resistance of the transformer, wherein the circulating current loss is the square of the circulating current value times the number of phases multiplied by the equivalent resistance in the transformer data, and the number of phases multiplied is related to the number of phases of the transformer; Step 303: The health index is equal to the comprehensive health index plus a weighted value of the ratio of circulating current loss to rated power.

5. The intelligent transformer for a power system according to claim 1, characterized in that: The steps of the anomaly location module include: Step 401, calculating the mean and standard deviation of all transformer health indicators to obtain the group mean and group standard deviation; Step 402: Determine whether each transformer is abnormal. If any condition is met, it is determined to be an abnormal transformer. The conditions include: Calculate the degree of deviation of the transformer's health index from the group mean. When the degree of deviation exceeds the preset multiple of the group standard deviation; When the health indicator of the transformer exceeds the safety threshold under the current working conditions; Step 403: For the abnormal transformer identified, the degree of abnormality is quantified using the Mahalanobis distance method.

6. The intelligent transformer for a power system according to claim 5, characterized in that: Methods for quantifying the degree of abnormality using the Mahalanobis distance method include: Construct a parameter vector for each transformer. The parameter vector includes the following: health indicators, oil temperature, winding temperature, load factor, and circulating current percentage from transformer data. The circulating current percentage is the ratio of the circulating current value to the rated current. Standardize all parameters of the parameter vector to obtain a standardized parameter vector. For the normal transformer group, the mean vector and covariance matrix of the normal transformer group are calculated based on the standardized parameter vector, and the Mahalanobis distance of each abnormal transformer is calculated based on the mean vector and covariance matrix. That is, the difference between the standardized parameter vector of the abnormal transformer and the mean vector of the normal transformer group is weighted by the inverse matrix of the covariance matrix, and the square root is taken to obtain the final Mahalanobis distance value. The Mahalanobis distance value represents the degree of abnormality, where the normal transformer group is the set of transformers that have not been judged as abnormal.

7. The intelligent transformer for a power system according to claim 6, characterized in that: The elements in the covariance matrix contain the covariances between the parameters, and the covariances between the parameters include: health index and oil temperature, health index and winding temperature, health index and load rate, health index and circulation percentage, oil temperature and winding temperature, oil temperature and load rate, oil temperature and circulation percentage, winding temperature and load rate, winding temperature and circulation percentage, and load rate and circulation percentage.

8. The intelligent transformer for a power system according to claim 1, characterized in that: The specific steps of the early warning maintenance module include: Based on the degree of abnormality and the number of abnormal transformers, combined with the number of redundant transformers, the warning level is divided into four levels: normal, low-level warning, medium-level warning, and high-level warning; For different warning levels, maintenance strategies are generated by combining the Mahalanobis distance value and the warning level. The maintenance strategies include: When the warning level is normal, no action is taken; When the warning level is low, reduce the load of the abnormal transformer until the warning level is reduced to normal; When the warning level is medium, load transfer is implemented to distribute the load of the abnormal transformer to the normal transformer until the warning level is reduced to low. When the warning level is high, load switching is performed to transfer the load from the abnormal transformer to the normal transformer or redundant transformer; and the abnormal transformer is isolated.

9. The intelligent transformer for a power system according to claim 8, characterized in that: The warning level is considered normal when all transformers are normal or the Mahalanobis distance values ​​of all abnormal transformers are less than the minimum value for the low-level warning, and the transformer group maintains full redundancy. Full redundancy is achieved by keeping X constant in the N+X structure, where N represents the minimum number of transformers required for normal operation and X represents the number of redundant transformers. When the Mahalanobis distance value of any abnormal transformer is between the minimum value and the maximum value of the low-level warning, the warning level is low-level warning; When the Mahalanobis distance value of any abnormal transformer is between the minimum value of the intermediate warning and the maximum value of the intermediate warning, or the Mahalanobis distance values ​​of two or more abnormal transformers exceed the minimum value of the low-level warning at the same time, or X=1, the warning level is intermediate warning; When the Mahalanobis distance value of any abnormal transformer exceeds the maximum value of the intermediate warning, or the Mahalanobis distance values ​​of two or more abnormal transformers simultaneously exceed the minimum value of the intermediate warning, or X=0, the warning level is high warning; The minimum value of the low-level warning is less than the maximum value of the low-level warning, the maximum value of the low-level warning is equal to the minimum value of the intermediate warning, and the minimum value of the intermediate warning is less than the maximum value of the intermediate warning; If multiple warning levels are met at the same time, the highest warning level will be output.

10. The intelligent transformer for a power system according to claim 1, characterized in that: Transformer data includes real-time operating parameters, rated parameters and safety thresholds. The safety thresholds include light-load operating thresholds, full-load operating thresholds, harmonic environment thresholds and normal operating thresholds.

Images