A method and system for intelligent identification of cooling defects in main transformers

Abstract

This invention discloses an intelligent identification method and system for cooling defects in main transformers. The method includes: real-time acquisition of temperature monitoring data, cooling device monitoring data, and operating condition correction data to establish a dual-modal dataset; extraction of temperature state features, cooling state features, and temperature-cooling response consistency features to obtain a dual-modal inspection state feature vector, and calculation of a comprehensive score for temperature anomalies and a comprehensive score for cooling state anomalies; calculation of robust identification parameters based on historical normal samples using a priori key cooling defect pattern library; collaborative identification by calling multiple identification agents to obtain cooling defect type, identification confidence level, anomaly evaluation value, and identification evidence chain; and further calculation of risk value and handling intensity to determine the risk level and match handling rules to obtain a handling strategy. This invention improves the identification priority, robustness, and targeted handling of key cooling defects while keeping the false alarm rate under control.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to an intelligent identification method and system for cooling defects in main transformers. Background Technology

[0002] The main transformer is a core piece of equipment in power systems and substations, and its safe and stable operation is directly related to the reliability of power supply. During the operation of the main transformer, the upper oil temperature and winding temperature are important parameters reflecting its thermal state. Coolers, fans, radiators, and backup cooling devices are crucial equipment for maintaining the safe temperature rise of the main transformer. When the cooling system malfunctions, the main transformer may experience a continuous rise in oil temperature and winding temperature, shorten insulation life, and in severe cases, may lead to equipment failure.

[0003] Existing methods for identifying cooling defects in main transformers have the following shortcomings: First, traditional inspections rely mainly on manual checks of temperature gauges, cooler operation status, fan operation status, and radiator contamination, making the identification results highly dependent on the experience of maintenance personnel. Second, existing temperature alarms often use fixed thresholds, typically only judging whether oil or winding temperatures exceed limits, lacking an assessment of the correlation between temperature changes and cooling device response, easily misjudging temperature rises caused by normal load increases as cooling faults. Third, main transformer temperature is significantly affected by load levels, ambient temperature, cooler switching status, operating mode, and measurement noise; actual operating data distribution may deviate from historical normal sample distribution, leading to false alarms or missed alarms by ordinary data-driven identification models under complex operating conditions. Fourth, cooling system defects have significant risk differences; critical cooling defects such as fan stoppage, cooler failure to engage, standby cooler unresponsiveness, severe radiator contamination, and persistent insufficient cooling effect, if not identified in a timely manner, may lead to serious consequences. Fifth, critical cooling defect samples are usually few, making it difficult for ordinary machine learning models to fully learn the characteristics of such defects, easily resulting in missed detections in the early stages. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method and system for intelligent identification of cooling defects in main transformers, which addresses the above-mentioned problems in the prior art. The method and system use strongly correlated temperature modes and cooling state modes as the core data foundation to construct temperature-cooling response consistency features and integrate prior key cooling defect information and a multi-bar identification mechanism to improve the identification priority and robustness of key cooling defects under the premise of controlled false alarm rate.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for intelligent identification of cooling defects in main transformers includes the following steps: Real-time acquisition of temperature monitoring data, cooling device monitoring data, and operating condition correction data of the main transformer; establishment of temperature mode dataset and cooling state mode dataset. Temperature state features, cooling state features, and temperature-cooling response consistency features are extracted from the temperature modal dataset and cooling state modal dataset to obtain a dual-modal inspection state feature vector, and a comprehensive score for temperature anomaly and a comprehensive score for cooling state anomaly are calculated. Using historical normal samples of the dual-modal inspection state feature vector, the current dual-modal inspection state feature vector, the comprehensive score of temperature anomaly and the comprehensive score of cooling state anomaly, and combined with the prior critical cooling defect pattern library, the sub-Bruk bar identification parameters are calculated. The sub-Bruk bar identification parameters include the critical defect priority identification score and the anomaly evaluation function. The prior critical cooling defect pattern library contains multiple types of critical cooling defect patterns. Each type of critical cooling defect pattern includes a defect name, defect feature contour, defect feature weight, evidence chain requirements and handling rules. The defect feature contour is composed of parameters in the dual-modal inspection state feature vector. Using different identification agents, collaborative identification is performed based on the current dual-modal inspection state feature vector, the prior key cooling defect pattern library, the key defect priority identification score, and the anomaly evaluation function to obtain the cooling defect type, identification confidence, anomaly evaluation value, and identification evidence chain. Based on the cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, calculate the risk value and handling intensity. Determine the defect risk level based on the risk value, and match the handling intensity with the handling rules in the prior critical cooling defect pattern library to obtain the handling strategy.

[0006] Furthermore, when extracting temperature state features, cooling state features, and temperature-cooling response consistency features from the temperature modal dataset and cooling state modal dataset, the specific steps include: calculating the oil temperature anomaly degree, winding temperature anomaly degree, and deviation between the field temperature and the background temperature from the temperature modal dataset, and using these, along with the ambient temperature and main transformer load level from the temperature modal dataset, as well as the oil temperature change rate, winding temperature change rate, and temperature anomaly duration from the temperature monitoring data, as temperature state features; and calculating the cooler operation rate, fan effective operating rate, and fan power consumption from the cooling state modal dataset. Flow deviation and radiator contamination index, together with radiator defect signal, standby radiator activation status and standby radiator response time in the cooling state modal dataset, are used as cooling state features. The expected cooling response is calculated by weighting radiator activation rate, fan effective operating rate, radiator contamination index and fan current deviation. Based on the expected cooling response, the oil temperature-cooling response inconsistency and winding temperature-cooling response inconsistency are calculated. The oil temperature response before and after cooling activation, the winding temperature response before and after cooling activation and the temperature-cooling state dual-modal comprehensive inconsistency index are also calculated and used as temperature-cooling response consistency features.

[0007] Furthermore, the oil temperature anomaly is calculated by subtracting the reference oil temperature under the same load and ambient temperature from the upper layer oil temperature in the temperature modal dataset, and then dividing by the difference between the upper layer oil temperature allowable limit and the reference oil temperature. The winding temperature anomaly is calculated by subtracting the reference winding temperature from the winding temperature in the temperature mode dataset, and then dividing by the difference between the allowable limit of the winding temperature and the reference winding temperature. The deviation between the field temperature and the background temperature is the absolute value of the difference between the field thermometer reading and the background monitored temperature in the temperature modal dataset; The cooler utilization rate is the ratio of the number of cooler groups already in operation to the total number of cooler groups in the cooling state modal dataset. The effective fan operating rate is the ratio of the actual number of operating fans in the cooling state modal dataset to the number of fans commanded to be activated; The fan current deviation is the absolute value of the difference between the fan motor current and the fan rated operating current in the cooling state modal data set, divided by the fan rated operating current. The radiator fouling index is the ratio of the fouled area of ​​the radiator to the effective heat dissipation area of ​​the radiator in the cooling state modal data. The oil temperature-cooling response inconsistency is the result of subtracting the expected cooling response from the normalized oil temperature change rate. The normalized oil temperature change rate is the result of subtracting the oil temperature change rate benchmark value under normal operating conditions from the oil temperature change rate in the temperature modal dataset, and then dividing by the difference between the oil temperature change rate warning limit and the benchmark value. The winding temperature-cooling response inconsistency is calculated by subtracting the expected cooling response from the normalized winding temperature change rate. The normalized winding temperature change rate is calculated by subtracting the winding temperature change rate reference value under normal operating conditions from the winding temperature change rate in the temperature modal dataset, and then dividing by the difference between the winding temperature change rate warning limit and the reference value. The oil temperature response before and after cooling is calculated by subtracting the upper oil temperature before cooling from the upper oil temperature after cooling is applied. The winding temperature response before and after cooling is calculated by subtracting the winding temperature before cooling from the winding temperature after cooling is applied. The temperature-cooling state dual-modal comprehensive inconsistency index is the result of a weighted sum of the oil temperature-cooling response inconsistency, the winding temperature-cooling response inconsistency, the oil temperature response before and after cooling input, the winding temperature response before and after cooling input, the radiator contamination index, and the degree of ineffective fan operation. The sum of the degree of ineffective fan operation and the effective fan operation rate is 1.

[0008] Furthermore, the mathematical expressions for calculating the comprehensive score of temperature anomaly and the comprehensive score of cooling condition anomaly are as follows:

[0009]

[0010] in, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. Indicates the degree of oil temperature abnormality. Indicates the degree of winding temperature anomaly. This represents the normalized rate of change of oil temperature. This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly. Indicates the cooler utilization rate. Indicates the effective operating rate of the fan. Indicates fan current deviation. This indicates the radiator's dirtiness index. This indicates a cooler defect signal. This indicates the normalized standby cooler response time. to This indicates the weighting of temperature anomaly assessment for different parameters. to This represents the weighting of the cooling status evaluation for different parameters.

[0011] Furthermore, when calculating the sub-Bruker identification parameters using historical normal samples of the dual-modal inspection state feature vector, the current dual-modal inspection state feature vector, the comprehensive score of temperature anomaly, and the comprehensive score of cooling state anomaly, combined with the prior key cooling defect pattern library, the following steps are included: An empirical distribution is constructed based on historical normal samples, expressed as follows:

[0012] in This represents the historical normal sample empirical distribution. Indicates concentration on the sample Point mass distribution at [location] This represents the i-th historical normal sample. This indicates the consistency characteristics of temperature-cooling response under normal conditions; Construct a set with an uncertain distribution, as shown in the following expression:

[0013] in, Represents a set with an uncertain distribution. This represents the actual operational disturbance distribution. This represents the distance between the actual operational disturbance distribution and the historical normal sample empirical distribution. Indicates the allowable distribution offset radius; Using the defect feature contours in the prior critical cooling defect pattern library as prior defect contours, the matching degree between the current state and the i-th type of critical defect is calculated, as shown in the following expression:

[0014]

[0015] in, This represents the degree of matching between the current state and the i-th type of critical defect pattern. Represents the feature matching function. Represents the feature profile of the i-th type of defect. This represents the feature vector of the dual-modal inspection state; The robustness index for critical defect detection is calculated using the following expression:

[0016] in, This represents the robustness index for detecting the i-th type of critical defect; The critical defect priority detection score is calculated using the following expression:

[0017]

[0018] in, This indicates the priority detection score for the i-th type of critical defect. This represents the risk weight of the i-th type of critical defect. This represents the robustness weight of the i-th type of critical defect. This represents the distribution offset risk correction term. Indicates the risk weight of distribution offset; The anomaly evaluation function and its corresponding constraints are constructed as follows:

[0019]

[0020]

[0021] in, This indicates the anomaly evaluation value of the current state. This represents the maximum score among all critical defect priority detection scores. This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. , , , These represent the weight coefficients of the corresponding evaluation items. Indicates the combined inconsistency index of temperature-cooling state dual modes; Indicates the threshold for abnormal alarms. Indicates the maximum permissible false alarm rate. Indicates normal operating status; This represents the prior defect profile of the i-th type of critical cooling defect in the prior critical cooling defect pattern library. This represents the feature vector of the current dual-modal inspection state. This represents the actual operational disturbance distribution. This represents the anomaly evaluation function. This represents the expected minimum detection rate for the i-th type of critical cooling defect. This indicates the loss in detection performance caused by the offset of the operational disturbance distribution.

[0022] Furthermore, the different identification agents include a primary screening agent, a temperature expert identification agent, a cooling expert identification agent, a priori critical defect matching agent, and a case verification agent. When using these different identification agents for collaborative identification based on the current bimodal inspection state feature vector, a priori critical cooling defect pattern library, critical defect priority identification scoring, and anomaly evaluation function, the specific methods include: The primary screening agent performs initial screening of anomaly types based on the oil temperature anomaly degree, winding temperature anomaly degree, temperature-cooling state dual-modal comprehensive inconsistency index, fan ineffective operation degree, and anomaly evaluation function of the dual-modal inspection state feature vector, and outputs the confidence score of the primary screening. When the confidence score of the primary screening is greater than the preset threshold, the temperature expert identification agent, the cooling expert identification agent, the prior key defect matching agent, and the case verification agent are triggered to perform parallel collaborative identification. The temperature expert identification agent identifies temperature defects and outputs the temperature expert identification confidence level based on the oil temperature anomaly degree, winding temperature anomaly degree, normalized oil temperature change rate, normalized winding temperature change rate, temperature anomaly duration, and field-background temperature deviation of the dual-modal inspection state feature vector. The cooling expert identification agent identifies cooling defects based on the cooler availability rate, fan effective operating rate, fan current deviation, radiator dirt index, cooler defect signal, standby cooler availability status, and normalized standby cooler response time of the dual-modal inspection state feature vector, and outputs the cooling expert identification confidence score. The prior critical defect matching agent outputs the most likely critical defect type and the prior critical defect matching confidence level based on the critical defect priority identification score, and takes the most likely critical defect type as the cooling defect type. The case verification agent outputs the case verification confidence score based on the similarity between the current bimodal inspection state feature vector and the historical case features; The outputs of each agent are weighted and fused to obtain the identification confidence level.

[0023] Furthermore, the mathematical expression for the identification confidence level is as follows:

[0024] in, This indicates the final identification confidence level. Indicates the confidence level of the initial screening. Indicates the reliability of temperature expert identification; This indicates the confidence level of the cooling expert's identification; This indicates the prior critical defect matching confidence level; This indicates the confidence level of the case validation. to These represent the fusion weights of the identification results of each agent; The mathematical expression for the confidence level of the primary screening is as follows:

[0025] In the formula, Indicates the confidence level of the initial screening; This represents the normalization function, used to map the scores to the [0,1] interval; Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; M represents the temperature-cooling state dual-mode comprehensive inconsistency index; Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; This represents the anomaly evaluation function; to This represents the weight coefficient of each evaluation item in the primary screening agent; The mathematical expression for the reliability of temperature expert identification is as follows:

[0026] in, This indicates the confidence level of the temperature expert's identification. Represents the normalization function; Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; This represents the normalized rate of change of oil temperature; This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly; This indicates the deviation between the ambient temperature and the background temperature; to This represents the weighting coefficients of each evaluation item in the temperature expert's identification agent; The mathematical expression for the confidence level identified by cooling experts is as follows:

[0027] in, This indicates the confidence level of the cooling expert's identification. Represents the normalization function; Indicates the cooler utilization rate; This indicates that the cooler is not fully engaged. Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; Indicates fan current deviation; Indicates the radiator's dirtiness index; This indicates a cooler defect signal; This indicates that the standby cooler is in operation; This indicates that the standby cooler is not in operation; This indicates the normalized standby cooler response time; to This represents the weighting coefficients of each evaluation item in the intelligent agent identified by the cooling expert; The mathematical expression for the prior critical defect matching confidence is as follows:

[0028] in, represents the priority detection score for the i-th type of critical cooling defect, and K represents the total number of critical cooling failure modes; The mathematical expression for the confidence level in case validation is as follows:

[0029]

[0030] in, Indicates the confidence level of the case validation; This represents the case similarity threshold. This indicates the similarity between the current case and historical cases. This represents the feature vector of the historical case state.

[0031] Furthermore, when calculating the risk value and handling intensity based on the aforementioned cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, the calculation of the risk value includes the following steps: Determine the corresponding chain of evidence requirements in the prior critical cooling defect pattern library based on the type of cooling defect; The value of evidence chain integrity is obtained by dividing the number of valid evidence in the identified evidence chain by the number of evidence required for the corresponding evidence chain in the prior critical cooling defect pattern library. The risk value is calculated by weighting the evidence chain integrity value, temperature anomaly comprehensive score, cooling state anomaly comprehensive score, anomaly evaluation value, current dual-modal inspection state feature vector, and prior critical cooling defect pattern library. The mathematical expression is as follows:

[0032] Where R represents the cooling defect risk value, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. This represents the risk weight of the i-th type of critical defect in the prior critical cooling defect pattern library. This represents the combined inconsistency index of the temperature-cooling state dual-modal characteristics in the current dual-modal inspection state feature vector. This represents the load correction factor. This represents the ambient temperature correction factor. This indicates the anomaly evaluation value of the current state. A value representing the completeness of the chain of evidence.

[0033] Furthermore, when calculating the risk value and handling intensity based on the cooling defect type, abnormal evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, the step of calculating the handling intensity is included. Specifically, the weighted sum of the risk value, critical defect priority identification score, and evidence chain integrity value is calculated to obtain the handling intensity value.

[0034] The present invention also proposes an intelligent identification system for cooling defects in a main transformer, comprising a processor and a computer-readable storage medium interconnected with each other, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the intelligent identification method for cooling defects in the main transformer.

[0035] Compared with the prior art, the advantages of the present invention are as follows: (1) In view of the technical problems that traditional inspections rely on human experience and the identification results are greatly affected by subjectivity, the present invention adopts the method of acquiring temperature monitoring data, cooling device monitoring data and operating condition correction data in real time to establish a dual-modal dataset, and calling multiple identification agents to perform collaborative identification, thereby transforming manual inspections into automated identification based on dual-modal physical quantity monitoring data, reducing the dependence of identification results on the experience of operation and maintenance personnel, and improving the objectivity and consistency of identification.

[0036] (2) In view of the technical problem that the fixed temperature threshold lacks the correlation judgment between temperature and cooling device response, and the temperature rise caused by normal load increase is easily misjudged as cooling failure, the present invention adopts the means of extracting temperature state characteristics, cooling state characteristics and temperature-cooling response consistency characteristics, transforming the simple judgment of whether the temperature exceeds the limit into the correlation identification method of judging whether the temperature change and the cooling device response match. The temperature-cooling response consistency characteristics reflect the matching relationship between physical thermal response and cooling device action, reducing false alarms caused by normal load increase and ambient temperature change, and improving the ability to identify early cooling defects.

[0037] (3) In response to the technical problems of data distribution deviation leading to false alarms or missed alarms in complex working conditions and insufficient priority of key cooling defect identification, this invention adopts the method of combining a prior key cooling defect pattern library and calculating the sub-Bru bar identification parameters based on the feature vectors of historical normal inspection samples and current dual-modal inspection states. Within the uncertain distribution set, the false alarm rate of normal state is constrained not to exceed the preset threshold, so that the model can maintain the stability of key cooling defect detection results under the conditions of load change, ambient temperature change, cooler switching change and measurement noise change, and achieve the effect of controlled false alarm rate and priority identification of key defects.

[0038] (4) In response to the technical problem of lack of closed-loop handling of identification results, the present invention adopts a means to calculate risk value and handling intensity based on cooling defect type, abnormal evaluation value, identification evidence chain, key defect priority identification score, temperature abnormality comprehensive score, cooling state abnormality comprehensive score, and evidence chain requirements and handling rules in the prior key cooling defect pattern library, and then determine the risk level and match the handling rules to obtain the handling strategy, thereby realizing closed-loop management from dual-modal data acquisition, intelligent identification to physical operation and maintenance handling. Attached Figure Description

[0039] Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0041] To address the issues of false alarms and missed alarms due to fixed temperature thresholds, the disconnect between temperature data and cooling status data, insufficient priority for identifying critical cooling defects, and inadequate robustness under complex operating conditions in existing technologies, this embodiment proposes an intelligent identification method for cooling defects in main transformers. Based on strongly correlated temperature and cooling status modes as core data, a temperature-cooling response consistency feature is constructed. Furthermore, prior information on critical cooling defects and a robust identification mechanism are integrated to improve the identification priority and robustness of critical cooling defects while keeping the false alarm rate under control.

[0042] like Figure 1 As shown, the method includes the following steps: S1. Construct a main transformer inspection object model. Based on the main transformer's structural composition, temperature rise mechanism, and cooling system failure mechanism, the main transformer is divided into temperature inspection objects, cooling device inspection objects, and operating condition correction objects, resulting in a temperature-cooling state dual-mode inspection object model.

[0043] S2, Collect dual-mode inspection data of the main transformer. Based on the temperature-cooling state dual-mode inspection object model, temperature mode data and cooling state mode data are collected to obtain the dual-mode raw inspection dataset. Specifically, in this embodiment, the dual-mode raw inspection dataset includes a temperature mode dataset and a cooling state mode dataset. According to the temperature-cooling state dual-mode inspection object model, the temperature monitoring quantity, cooling device monitoring quantity, and operating condition correction quantity of the main transformer are acquired in real time to establish the temperature mode dataset and the cooling state mode dataset.

[0044] S3. Extract the main transformer inspection status features. Based on the dual-modal original inspection dataset, calculate the temperature status features, cooling status features, and temperature-cooling response consistency features to obtain the main transformer dual-modal inspection status feature vector. Specifically, in this embodiment, extract the temperature status features, cooling status features, and temperature-cooling response consistency features from the temperature mode dataset and the cooling status mode dataset to obtain the dual-modal inspection status feature vector, and calculate the comprehensive score for temperature anomalies and the comprehensive score for cooling status anomalies.

[0045] S4. Construct a priori critical cooling defect pattern library. Based on the aforementioned state feature vector, operation and maintenance procedures, historical defect records, expert experience, and accident cases, a priori critical cooling defect pattern library is constructed. In this embodiment, the priori critical cooling defect pattern library contains multiple types of critical cooling defect patterns. Each type of critical cooling defect pattern includes a defect name, defect feature profile, defect feature weight, evidence chain requirements, and handling rules. The defect feature profile is composed of parameters from the dual-modal inspection state feature vector to achieve a structured expression of high-risk defects in the main transformer cooling system.

[0046] S5. Construct a sub-Bruker identification model. Based on historical normal inspection samples, the current dual-modal inspection state feature vector, and a priori critical cooling defect pattern library, construct a distribution uncertainty set, a critical defect detection robustness index, a critical defect priority identification score, and an anomaly evaluation function. Specifically, in this embodiment, using historical normal samples of the dual-modal inspection state feature vector, the current dual-modal inspection state feature vector, the comprehensive score of temperature anomaly, and the comprehensive score of cooling state anomaly, combined with the priori critical cooling defect pattern library, calculate the sub-Bruker identification parameters.

[0047] S6. Perform multi-agent collaborative identification. Based on the bimodal inspection state feature vector, the prior critical cooling defect pattern library, the critical defect priority identification score, and the anomaly evaluation function, a primary screening agent, a temperature expert identification agent, a cooling expert identification agent, a prior critical defect matching agent, and a case verification agent are used for collaborative identification to obtain the cooling defect type, identification confidence level, anomaly evaluation value, and identification evidence chain. Specifically, this embodiment uses different identification agents to perform collaborative identification based on the current bimodal inspection state feature vector, the prior critical cooling defect pattern library, the critical defect priority identification score, and the anomaly evaluation function to obtain the cooling defect type, identification confidence level, anomaly evaluation value, and identification evidence chain.

[0048] S7, Generate Risk Level and Handling Strategy. Based on the cooling defect type, identification confidence level, anomaly evaluation value, critical defect priority identification score, and identification evidence chain, calculate the risk value and handling intensity, and output the risk level and handling strategy. Specifically, in this embodiment, the risk value and handling intensity are calculated based on the cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score. The defect risk level is determined based on the risk value, and the handling intensity is matched with the handling rules in the prior critical cooling defect pattern library to obtain the handling strategy.

[0049] The above steps form a continuous data transmission link: the inspection object model output by S1 is used to limit the data acquisition range of S2; the bimodal raw inspection data output by S2 is used for feature extraction in S3; the state feature vector output by S3 is used for defect pattern construction in S4, robust identification modeling in S5, and collaborative identification in S6; the prior critical cooling defect pattern library output by S4 is used for critical defect matching in S5, defect type identification in S6, and handling strategy generation in S7; the critical defect priority identification score and anomaly evaluation function output by S5 are used for identification fusion in S6 and risk calculation in S7; the identification results output by S6 are used to generate risk levels and handling strategies in S7.

[0050] This embodiment also proposes an intelligent identification system for cooling defects in main transformers, including a processor and a computer-readable storage medium interconnected with each other. The computer-readable storage medium stores a computer program, which is executed by the processor to implement the steps of the intelligent identification method for cooling defects in main transformers described in this embodiment. The system includes: The inspection object modeling module is used to construct a dual-modal inspection object model of the main transformer, which is in both temperature and cooling states. A dual-mode data acquisition module is used to acquire temperature mode data and cooling status mode data of the main transformer; The state feature extraction module is used to extract temperature state features, cooling state features, and temperature-cooling response consistency features. The Prior Critical Defect Pattern Library module is used to build and store critical cooling defect patterns of the main transformer. The Bruker identification module is used to construct distribution uncertainty sets, robustness indicators for key defect detection, priority identification scoring for key defects, and anomaly evaluation functions; The multi-agent collaborative identification module is used to obtain identification results through primary screening, temperature expert identification, cooling expert identification, prior critical defect matching, and case verification.

[0051] The risk assessment and response decision-making module is used to calculate the risk value, risk level, response intensity, and response strategy based on the identification results.

[0052] The modules form a continuous data flow in the following order: "Inspection Object Modeling—Dual-Modal Data Acquisition—State Feature Extraction—Priority Defect Pattern Construction—Sub-Bruker Bar Identification—Multi-Agent Collaborative Identification—Risk Assessment and Disposal Decision-Making." The output of the Inspection Object Modeling module serves as the input to the Dual-Modal Data Acquisition module and the State Feature Extraction module; the output of the Dual-Modal Data Acquisition module serves as the input to the State Feature Extraction module; the output of the State Feature Extraction module serves as the input to the Priority Key Defect Pattern Library module, the Sub-Bruker Bar Identification module, the Multi-Agent Collaborative Identification module, and the Risk Assessment and Disposal Decision-Making module; the output of the Priority Key Defect Pattern Library module serves as the input to the Sub-Bruker Bar Identification module, the Multi-Agent Collaborative Identification module, and the Risk Assessment and Disposal Decision-Making module; the output of the Sub-Bruker Bar Identification module serves as the input to the Multi-Agent Collaborative Identification module and the Risk Assessment and Disposal Decision-Making module; and the output of the Multi-Agent Collaborative Identification module serves as the input to the Risk Assessment and Disposal Decision-Making module. The output of the preceding module serves as the input to the following module, ensuring that the diagnostic process has clear data transfer relationships, evidence traceability, and closed-loop disposal capabilities.

[0053] The following provides a detailed explanation of each step.

[0054] In step S1 of this embodiment, based on the main transformer's structural composition, temperature rise mechanism, and cooling system failure mechanism, the inspection object modeling module is called to divide the main transformer into temperature inspection objects, cooling device inspection objects, and operating condition correction objects, thereby obtaining the dual-modal inspection object model required for subsequent data acquisition and feature extraction.

[0055] In this step, the following inspection object model is constructed:

[0056] in, This represents the dual-mode inspection object model of the main transformer. This represents the set of temperature monitoring objects. This represents the set of objects to be inspected for the cooling system. This represents the set of objects for operational condition correction.

[0057] The set of temperature monitoring objects is as follows:

[0058] in, Indicates the temperature of the upper oil layer. Indicates the winding temperature. This indicates the reading of the on-site thermometer. This indicates the temperature being monitored in the background. Indicates the rate of change of oil temperature. This represents the rate of change of winding temperature. Indicates the duration of the temperature anomaly.

[0059] The set of objects to be inspected for cooling devices is as follows:

[0060] in, This indicates the number of cooler units that are already in operation. Indicates the total number of cooler groups. Indicates that the fan is in command status. Indicates the actual operating status of the fan. Indicates the fan motor current. Indicates the degree of dirtiness of the radiator. This indicates a cooler defect signal. This indicates that the standby cooler is in operation. This indicates the response time of the standby cooler.

[0061] The set of objects for operational condition correction is as follows:

[0062] in, Indicates the main transformer load level. Indicates ambient temperature. This indicates the cooling control method or operating mode.

[0063] The temperature-cooling state dual-modal inspection object model output in this step It is used in step S2 to determine the data acquisition range of temperature mode and cooling state mode, in step S3 to determine the state feature extraction object, and in step S4 to determine the feature composition of the prior key cooling defect mode.

[0064] In this embodiment, step S2 is based on the temperature-cooling state dual-modal inspection object model obtained in step S1. It calls the dual-modal data acquisition module to collect temperature modal data and cooling state modal data to obtain the dual-modal raw inspection dataset for subsequent state feature extraction.

[0065] The temperature modal dataset is represented as follows:

[0066] in, Represents a temperature modal dataset. This indicates the temperature of the upper oil layer at time t. This represents the winding temperature at time t. This represents the on-site thermometer reading at time t. This represents the temperature monitored in the background at time t. This represents the ambient temperature at time t. This indicates the main transformer load level at time t.

[0067] Based on continuous sampling data, the rate of change of oil temperature can be calculated:

[0068] in, This represents the rate of change of oil temperature at time t. This indicates the upper oil temperature at the previous sampling time. This indicates the time interval between two consecutive samples.

[0069] Similarly, based on continuously sampled data, the rate of change of winding temperature can be calculated:

[0070] in, This represents the rate of change of winding temperature at time t. This indicates the winding temperature at the previous sampling time. This indicates the time interval between two consecutive samples.

[0071] The cooling state modal dataset is represented as follows:

[0072] in, This represents a dataset of cooling state modalities. This indicates the fan's rated operating current. Indicates the area of ​​dirt on the radiator. This indicates the effective heat dissipation area of ​​the radiator; the meanings of the other parameters are the same as in step S1.

[0073] In this embodiment, step S3 is based on the dual-modal raw inspection data obtained in step S2. The state feature extraction module is called to normalize, differentiate, calculate the deviation, and model the response consistency of the temperature data and cooling state data, so as to obtain temperature state features, cooling state features, and temperature-cooling response consistency features.

[0074] Specifically, when extracting temperature state features, cooling state features, and temperature-cooling response consistency features from the temperature modal dataset and cooling state modal dataset, the process includes: 1. Calculate the oil temperature anomaly degree based on the temperature modal dataset. Winding temperature anomaly Deviation between ambient temperature and background temperature The ambient temperature in the temperature modality dataset and main transformer load level And the rate of change of oil temperature in the temperature monitoring data. Winding temperature change rate Duration of temperature anomaly Together, they serve as characteristics of the temperature state, resulting in the following temperature state feature vector:

[0075] in: The oil temperature anomaly is calculated by subtracting the reference oil temperature under the same load and ambient temperature from the upper layer oil temperature in the temperature modal dataset, and then dividing by the difference between the allowable limit of the upper layer oil temperature and the reference oil temperature. The mathematical expression is as follows:

[0076] in, Indicates the degree of oil temperature abnormality. This indicates the reference oil temperature under similar loads and ambient temperatures. Indicates the allowable limit for the upper oil temperature; The winding temperature anomaly is calculated by subtracting the reference winding temperature from the winding temperature in the temperature mode dataset, and then dividing by the difference between the allowable winding temperature limit and the reference winding temperature. The mathematical expression is as follows:

[0077] in, Indicates the degree of winding temperature anomaly. This indicates the reference winding temperature under similar loads and ambient temperatures. Indicates the allowable temperature limit of the winding; The deviation between the on-site temperature and the background temperature is the absolute value of the difference between the on-site thermometer reading and the background monitored temperature in the temperature modal dataset, and its mathematical expression is: , This indicates the reading of the on-site thermometer. This indicates the temperature being monitored in the background.

[0078] 2. The cooler utilization rate is calculated based on the cooling state modal dataset. Fan effective operating rate Fan current deviation Radiator dirt index The cooler defect signal in the cooling state modal dataset The standby cooler is in operation. and backup cooler response time Together, these are used as features of the cooling state, resulting in the following cooling state feature vector:

[0079] in: The cooler utilization rate is the number of cooler groups already in operation in the cooling state modal dataset. Total number of cooler groups The ratio, mathematically expressed as follows: ; The effective fan operating rate is the actual number of operating fans in the cooling state modality dataset. Number of fans to be deployed according to instructions The ratio, mathematically expressed as follows: ; The fan current deviation is the fan motor current in the cooling state mode data. With the fan's rated operating current The absolute value of the difference divided by the fan's rated operating current The mathematical expression is as follows: ; The radiator fouling index is the amount of fouled area of ​​the radiator in the cooling state modal data. With the effective heat dissipation area of ​​the radiator The ratio, mathematically expressed as follows: ; 3. The expected cooling response is calculated by weighting the cooler activation rate, fan effective operating rate, radiator contamination index, and fan current deviation. Based on the expected cooling response, the oil temperature-cooling response inconsistency and winding temperature-cooling response inconsistency are calculated. The oil temperature response before and after cooling activation, the winding temperature response before and after cooling activation, and the temperature-cooling state dual-modal comprehensive inconsistency index are also calculated and used as the temperature-cooling response consistency characteristic. The specific process is as follows: To avoid dimensional differences between the temperature change rate and the cooling status index, the oil temperature change rate and the winding temperature change rate are normalized before constructing the temperature-cooling response consistency characteristics.

[0080] The normalized oil temperature change rate is calculated by subtracting the oil temperature change rate benchmark value under normal operating conditions from the oil temperature change rate in the temperature modal dataset, and then dividing by the difference between the oil temperature change rate warning limit and the benchmark value.

[0081] The normalized winding temperature change rate is calculated by subtracting the baseline value of the winding temperature change rate under normal operating conditions from the winding temperature change rate in the temperature modal dataset, and then dividing by the difference between the warning limit and the baseline value of the winding temperature change rate.

[0082] in, This represents the normalized rate of change of oil temperature. This represents the normalized rate of change of winding temperature. and These represent the reference values ​​for the rate of change of oil temperature and the rate of change of winding temperature under normal operating conditions, respectively. and These represent the warning limits for the rate of change of oil temperature and the rate of change of winding temperature, respectively.

[0083] The expected cooling response is further constructed as follows:

[0084] in, Indicates the expected cooling response amount. Indicates the weight of the cooler's availability rate. Indicates the weight of the effective operating rate of the fan. This indicates the weight of the radiator dirt index. This indicates the weight of the fan current deviation.

[0085] The oil temperature-cooling response inconsistency is calculated by subtracting the expected cooling response from the normalized oil temperature change rate. ; The winding temperature-cooling response inconsistency is calculated by subtracting the expected cooling response from the normalized winding temperature change rate. ; The oil temperature response before and after cooling is calculated by subtracting the upper oil temperature before cooling from the upper oil temperature after cooling is applied. ; The winding temperature response before and after cooling is calculated by subtracting the winding temperature before cooling from the winding temperature after cooling is applied. ; The temperature-cooling state dual-modal comprehensive inconsistency index is the result of a weighted sum of the inconsistency between oil temperature and cooling response, the inconsistency between winding temperature and cooling response, the oil temperature response before and after cooling activation, the winding temperature response before and after cooling activation, the radiator contamination index, and the degree of ineffective fan operation.

[0086] in, This indicates the degree to which the fan is not operating effectively, and the difference between the degree of fan ineffective operation and the effective fan operation rate. The sum of is 1.

[0087] Therefore, the temperature-cooling response consistency feature vector is obtained as follows:

[0088] The final output is the dual-modal inspection state feature vector:

[0089] In step S3, the mathematical expressions for calculating the comprehensive score of temperature anomaly and the comprehensive score of cooling state anomaly are as follows:

[0090]

[0091] in, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. Indicates the degree of oil temperature abnormality. Indicates the degree of winding temperature anomaly. This represents the normalized rate of change of oil temperature. This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly. Indicates the cooler utilization rate. Indicates the effective operating rate of the fan. Indicates fan current deviation. This indicates the radiator's dirtiness index. This indicates a cooler defect signal. This indicates the normalized standby cooler response time. to This indicates the weighting of temperature anomaly assessment for different parameters. to This represents the weighting of the cooling status evaluation for different parameters.

[0092] Step S3 Output , , In step S4, it is used to construct the defect feature profile; in step S5, it is used to calculate the critical defect matching degree and the anomaly evaluation function; in step S6, it is used for multi-agent identification; and in step S7, it is used for risk value calculation.

[0093] Step S4 of this embodiment, based on the state feature vector obtained in step S3, and combined with operation and maintenance procedures, historical defect records, expert experience, and accident cases, calls the prior critical defect pattern library module to perform a structured expression of high-risk defects in the main transformer cooling system, thus obtaining the prior critical defect pattern library used for subsequent robust identification and critical defect matching:

[0094] in, This represents a priori critical defect pattern library. Let K represent the i-th type of critical cooling defect mode, and K represent the number of critical defect types.

[0095] The patterns for each type of critical defect are as follows:

[0096] in, Indicate the fault name, Represents the feature outline of the defect. Represents the defect feature weights. Indicates the chain of evidence requirement, This indicates the rules for handling the situation.

[0097] The defect feature profile is composed of parameters in the dual-modal inspection state feature vector. Taking the fan stop-transition cooling anomaly as an example, its defect feature profile is as follows:

[0098] Its feature weights are:

[0099] in, Indicates the weight of oil temperature anomaly. Indicates the weight of winding temperature anomaly. Indicates the weight of the effective operating rate of the fan. Indicates the weighting of fan current deviation. This indicates the weight of the overall inconsistency index.

[0100] Chain of evidence requirements These include situations where the cooler is in operation, the fan is not running effectively, the fan current is abnormal, the oil temperature or winding temperature continues to rise, and the temperature-cooling response inconsistency index increases.

[0101] Disposal rules This includes putting the backup cooler into operation, checking the fan motor, checking the fan control circuit, reducing the load if necessary, or requesting a power outage for maintenance.

[0102] Table 1 shows some of the contents of the prior critical defect pattern library.

[0103] Table 1

[0104] Step S4 output Among them, the defect feature contour In step S5, the prior defect profile and defect feature weights are used. In step S5, the priority detection score for critical defects is calculated, and the chain of evidence is required. Steps S6 and S7 are used to calculate the completeness of the chain of evidence and the disposal rules. Step S7 is used to generate a disposal strategy.

[0105] In this embodiment, step S5, based on the current state feature vector obtained in step S3, the prior critical defect pattern library obtained in step S4, and historical normal inspection samples, calls the sub-Bruker identification module to construct a distribution uncertainty set, a critical defect detection robustness index, a critical defect priority identification score, and an anomaly evaluation function, thereby obtaining the identification quantities required for subsequent multi-agent identification and risk assessment. When calculating the sub-Bruker identification parameters using historical normal samples of the dual-modal inspection state feature vector, the current dual-modal inspection state feature vector, the comprehensive score for temperature anomalies, and the comprehensive score for cooling state anomalies, combined with the prior critical cooling defect pattern library, the following steps are included: 1. Construct an empirical distribution based on historical normal samples.

[0106] Let the historical normal sample be:

[0107] Among them, among them, Let N represent the i-th historical normal sample, and N represent the number of historical normal samples. This indicates the temperature-cooling response disturbance characteristics under normal conditions.

[0108] The empirical distribution expression is constructed as follows:

[0109] in This represents the historical normal sample empirical distribution. Indicates concentration on the sample The point mass distribution at that location.

[0110] 2. Construct a set with an uncertain distribution, expressed as follows:

[0111] in, Represents a set with an uncertain distribution. This represents the actual operational disturbance distribution. This represents the distance between the actual operational disturbance distribution and the historical normal sample empirical distribution. Indicates the allowed distribution offset radius.

[0112] 3. Calculate the matching degree. Using the defect feature contours in the prior critical cooling defect pattern library from step S4 as prior defect contours, calculate the matching degree between the current state and the i-th type of critical defect, as shown in the following expression:

[0113]

[0114] in, This represents the degree of matching between the current state and the i-th type of critical defect pattern. Represents the feature matching function. Represents the feature profile of the i-th type of defect. This represents the feature vector of the dual-modal inspection state; As a preferred approach, the matching function employs weighted cosine similarity:

[0115] In the formula, The current state represents the matching degree between the current state and the i-th type of critical cooling defect mode; X represents the feature vector of the current dual-modal inspection state; Represents the prior fault profile of the i-th type of critical cooling defect; The feature weight vector representing the i-th type of critical cooling defect; This indicates element-wise multiplication; Represents the vector norm.

[0116] 4. Calculate the robustness index for key defect detection, as shown in the following expression:

[0117] in, This indicates the robustness index for detecting the i-th type of critical defect. The larger the value, the smaller the deviation of the current working condition from the historical normal samples, and the more stable the detection of the i-th type of critical defect.

[0118] 5. Calculate the priority detection score for critical defects, using the following expression:

[0119]

[0120] in, This indicates the priority detection score for the i-th type of critical defect. This represents the risk weight of the i-th type of critical defect. This represents the robustness weight of the i-th type of critical defect. This represents the distribution offset risk correction term. This represents the distribution offset risk weight.

[0121] 6. Construct the anomaly evaluation function and the corresponding constraints.

[0122] The mathematical expression for the anomaly evaluation function is as follows:

[0123] in, This indicates the anomaly evaluation value of the current state. This represents the maximum score among all critical defect priority detection scores. This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. , , , These represent the weight coefficients of the corresponding evaluation items. It represents the combined inconsistency index of the temperature-cooling state dual-mode.

[0124] Set false alarm rate constraints:

[0125] in, Indicates the threshold for abnormal alarms. Indicates the maximum permissible false alarm rate. This indicates that the system is in normal operating condition.

[0126] While ensuring the false alarm rate is controlled under normal conditions, further constraints are set for critical defect detection. For the i-th type of critical cooling defect in the prior critical cooling defect pattern library, its prior defect profile is as follows: The current dual-modal inspection state feature vector is X. After superimposing the current state with the key defect contour, the anomaly evaluation function should satisfy:

[0127] in, This represents the prior defect profile of the i-th type of critical cooling defect in the prior critical cooling defect pattern library. This represents the feature vector of the current dual-modal inspection state. This represents the actual operational disturbance distribution. This represents the anomaly evaluation function. This represents the expected minimum detection rate for the i-th type of critical cooling defect. This represents the loss in detection performance caused by the offset of operational disturbance distribution. Through this constraint, the model can still prioritize the detection capability of key cooling defects such as fan stoppage, cooler failure to engage, standby cooler non-response, severe radiator contamination, and persistent insufficient cooling effect under conditions of load variation, ambient temperature variation, cooler switching variation, and measurement noise variation.

[0128] in, The settings can be configured based on the risk level of the critical cooling defect. For high-risk defects such as fan failure, cooler activation failure, standby cooler unresponsiveness, severely contaminated radiators, and persistently insufficient cooling, higher settings should be applied. To increase its detection priority; for relatively low-risk abnormal modes such as false alarms caused by temperature sensor malfunctions, a relatively low priority can be set. .

[0129] The loss of detection performance It can be determined based on the distribution offset distance, for example:

[0130] In the formula, This represents the sensitivity coefficient of the i-th type of critical cooling defect to the distribution shift; Indicates the actual operational disturbance distribution Compared with historical normal sample empirical distribution The distance between them. Therefore, when operating conditions, ambient temperature, load level, or cooler switching status change, the model can dynamically adjust the key defect detection requirements according to the degree of distribution offset, thereby improving the robust identification capability of key cooling defects while keeping the false alarm rate under control.

[0131] Step S5 output:

[0132] in, and In step S6, it is used for multi-agent collaborative identification, and in step S7, it is used for risk level and treatment intensity calculation.

[0133] In this embodiment, step S6, based on the state features obtained in step S3, the prior critical defect pattern library obtained in step S4, and the critical defect matching degree, robustness index, priority detection score, and anomaly evaluation function obtained in step S5, calls the multi-agent collaborative identification module to complete the initial screening, temperature identification, cooling identification, critical defect matching, and case verification, and obtains the final cooling defect identification result.

[0134] In step S6, the different identification agents include a primary screening agent, a temperature expert identification agent, a cooling expert identification agent, a priori critical defect matching agent, and a case verification agent. When using different identification agents for collaborative identification based on the current bimodal inspection state feature vector, the priori critical cooling defect pattern library, the critical defect priority identification score, and the anomaly evaluation function, the specific steps include: 1. The primary screening agent performs initial screening of anomaly types based on the oil temperature anomaly degree, winding temperature anomaly degree, temperature-cooling state dual-modal comprehensive inconsistency index, fan ineffective operation degree, and anomaly evaluation function of the dual-modal inspection state feature vector, and outputs the confidence score of the primary screening. When the confidence score of the primary screening is greater than the preset threshold, the temperature expert identification agent, the cooling expert identification agent, the prior key defect matching agent, and the case verification agent are triggered to perform parallel collaborative identification.

[0135] Specifically, the primary screening agent is based on , M and Perform rapid screening and output the initial abnormality type, initial screening confidence level, and initial screening evidence. The initial screening confidence level is expressed as:

[0136] In the formula, Indicates the confidence level of the initial screening; This represents the normalization function, used to map the scores to the [0,1] interval; Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; M represents the temperature-cooling state dual-mode comprehensive inconsistency index; Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; This represents the anomaly evaluation function; to This represents the weight coefficient of each evaluation item in the primary screening agent.

[0137] when Greater than the preset initial screening threshold At that time, the primary screening agent outputs a suspected cooling defect and transmits the current state characteristics to the temperature expert identification agent and the cooling expert identification agent for further analysis.

[0138] 2. The temperature expert identification agent identifies temperature defects based on the oil temperature anomaly degree, winding temperature anomaly degree, normalized oil temperature change rate, normalized winding temperature change rate, temperature anomaly duration, and field-background temperature deviation of the dual-modal inspection state feature vector, and outputs the temperature expert identification confidence score.

[0139] Specifically, temperature experts identify the temperature state characteristics of intelligent agents. To determine whether a temperature defect truly exists, whether it is continuously developing, and whether it may be caused by an anomaly at the measuring point. Calculate the confidence level for temperature expert identification:

[0140] in, This indicates the confidence level of the temperature expert's identification. This represents the normalization function. Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; This represents the normalized rate of change of oil temperature; This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly; This indicates the deviation between the ambient temperature and the background temperature; to This represents the weighting coefficients of each evaluation item in the temperature expert's identification agent.

[0141] in, The negative sign is used because a large deviation between the on-site temperature and the backend temperature indicates that the temperature anomaly may originate from a problem with the temperature measurement point or the backend data acquisition channel, rather than necessarily a genuine cooling defect. Therefore, The larger the value, the lower the confidence level of the true temperature anomaly can be, while triggering the judgment of abnormal false alarms of the temperature sensor.

[0142] 3. The cooling expert identification agent identifies cooling defects based on the cooler availability rate, fan effective operating rate, fan current deviation, radiator dirt index, cooler defect signal, standby cooler availability status, and normalized standby cooler response time from the dual-modal inspection state feature vector, and outputs the cooling expert identification confidence score.

[0143] Specifically, cooling experts identify the characteristics of the cooling state used by intelligent agents. This determines whether the cooling device is in operation as required, whether it is functioning effectively, and whether it has sufficient cooling capacity. Its identification confidence level is expressed as:

[0144] in, This indicates the confidence level of the cooling expert's identification. This represents the normalization function. Indicates the cooler utilization rate; This indicates that the cooler is not fully engaged. Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; Indicates fan current deviation; Indicates the radiator's dirtiness index; This indicates a cooler defect signal; This indicates that the standby cooler is in operation; This indicates that the standby cooler is not in operation; This indicates the normalized standby cooler response time; to This represents the weighting coefficients of each evaluation item in the intelligent agent identified by the cooling expert.

[0145] in, It can be represented as:

[0146] In the formula, This indicates the actual response time of the standby cooler. This indicates the maximum allowable response time for the backup cooler. When A higher level indicates an abnormally significant cooling condition; when and At the same time, a higher value indicates a strong consistency between temperature anomalies and cooling condition anomalies, increasing the credibility of cooling defects.

[0147] 4. The prior critical defect matching agent outputs the most likely critical defect type and the prior critical defect matching confidence level based on the critical defect priority identification score, and takes the most likely critical defect type as the cooling defect type.

[0148] Specifically, the prior critical defect matching agent uses the output of step S5. This determines which prior critical cooling defect mode the current state is closer to. This is then combined with the critical defect priority detection score obtained in step S5. Identify the most likely critical defect type:

[0149] in, This indicates the type of critical defect obtained from the final match. This indicates the priority detection score for the i-th type of critical cooling defect.

[0150] The prior critical fault matching confidence level The calculation method is as follows:

[0151] in, This indicates the priority detection score for the i-th type of critical cooling defect. K represents the total number of critical cooling failure modes.

[0152] 5. The case verification agent outputs the case verification confidence score based on the similarity between the current bimodal inspection state feature vector and the historical case features; The outputs of each agent are weighted and fused to obtain the identification confidence level.

[0153] Specifically, the case verification agent uses the current state feature X and historical case features. Calculate case similarity:

[0154] in, This indicates the similarity between the current case and historical cases. This represents the state feature vector of historical cases. The historical case database consists of historical cooling defect records, maintenance reports, inspection records, and handling results. Each historical case includes at least the defect type, state feature vector, identification evidence chain, handling measures, and handling results. Represents the feature vector of the current state; The confidence level of case validation can be expressed as:

[0155] In the formula, Indicates the confidence level of the case validation; This represents the case similarity threshold. When... When, it indicates that the current identification result is supported by historical cases; when Furthermore, when the risk level is high, the system will trigger a manual review prompt.

[0156] In step S6, the outputs of each agent are weighted and fused to obtain the final identification confidence score. The mathematical expression for the identification confidence score is as follows:

[0157] in, This indicates the final identification confidence level. Indicates the confidence level of the initial screening. Indicates the reliability of temperature expert identification; This indicates the confidence level of the cooling expert's identification; This indicates the prior critical defect matching confidence level; This indicates the confidence level of the case validation. to These represent the fusion weights of the identification results of each agent; where .

[0158] Step S6 output:

[0159] in, This indicates the final identification result. Indicates the type of cooling defect. This indicates the final identification confidence level. Indicates an abnormal evaluation value. This indicates the identification of the evidence chain. This output is used in step S7 for risk value calculation, risk level determination, and response strategy generation.

[0160] It should be noted that the primary screening agent, temperature expert identification agent, cooling expert identification agent, prior key defect matching agent, and case verification agent in step S6 can all be obtained by using conventional agent construction methods in the field or by training based on conventional machine learning models. Their specific model structures, training dataset construction methods, and training algorithms are common knowledge in the field and are not points of improvement. Therefore, the detailed training process and model structure of the identification network inside each agent will not be elaborated.

[0161] In this embodiment, step S7, based on the final cooling defect type, identification confidence level, anomaly evaluation value, and identification evidence chain output in step S6, and combined with the key defect priority detection score output in step S5 and the evidence chain requirements and handling rules in step S4, calls the risk assessment and handling decision module to calculate the risk value and handling intensity, and finally outputs the risk level and handling strategy. The specific implementation process is as follows: First, based on the aforementioned cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, the risk value and handling intensity are calculated, specifically including: Determine the corresponding chain of evidence requirements in the prior critical cooling defect pattern library based on the type of cooling defect; The completeness of the evidence chain is obtained by dividing the number of valid pieces of evidence in the identified evidence chain by the number of pieces of evidence required for the corresponding evidence chain in the prior critical cooling defect pattern library. The mathematical expression is as follows:

[0162] in, Indicating the completeness of the chain of evidence, This indicates the number of valid pieces of evidence that satisfy the current diagnosis. This indicates the amount of evidence required for the corresponding prior critical defect pattern. The risk value is calculated by weighting the evidence chain integrity value, temperature anomaly comprehensive score, cooling state anomaly comprehensive score, anomaly evaluation value, current dual-modal inspection state feature vector, and prior critical cooling defect pattern library. The mathematical expression is as follows:

[0163] Where R represents the cooling defect risk value, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. This represents the risk weight of the i-th type of critical defect in the prior critical cooling defect pattern library. This represents the temperature-cooling state z-modal comprehensive inconsistency index in the current dual-modal inspection state feature vector. This represents the load correction factor. This represents the ambient temperature correction factor. This indicates the anomaly evaluation value of the current state. A value representing the completeness of the chain of evidence.

[0164] The intensity of action is calculated by weighting the risk value, the critical defect priority identification score, and the completeness of the chain of evidence. The mathematical expression for the intensity of action is as follows:

[0165] in, Indicates the intensity of the treatment. This indicates a priority score for detecting critical defects.

[0166] Then, based on the risk value R, the defect risk level is determined, as shown in Table 2.

[0167] Table 2

[0168] At the same time, according to the intensity of treatment Handling rules in prior critical defect patterns The resulting disposal strategy is shown in Table 3.

[0169] Table 3

[0170] Final output:

[0171] Where Z represents the final diagnosis and treatment output, Indicates the risk level. This indicates the handling strategy.

[0172] In summary, this invention discloses an intelligent identification method and system for cooling defects in main transformers. The core identification inputs are the temperature mode and the cooling state mode of the main transformer. The temperature mode characterizes the thermal response of the cooling defects, while the cooling state mode characterizes the operational response of the cooling system. These two modes constitute the dual-mode combination with the strongest cause-effect correlation in cooling defect identification. The temperature mode includes data such as upper oil temperature, winding temperature, and temperature change rate. The cooling state mode includes data such as the number of cooler units in operation, fan activation commands, actual fan operating status, and standby cooler response status. First, a dual-modal inspection model of the main transformer's temperature and cooling status is constructed. Then, dual-modal inspection data is collected, and temperature status features, cooling status features, and temperature-cooling response consistency features are extracted. Next, a priori critical cooling defect pattern library is constructed, and a robust identification model is established based on historical normal inspection samples, current status features, and the priori critical cooling defect pattern library to obtain critical defect detection robustness indicators, critical defect priority identification scores, and anomaly evaluation functions. Subsequently, collaborative identification is performed through a primary screening agent, a temperature expert identification agent, a cooling expert identification agent, a priori critical defect matching agent, and a case verification agent. Finally, cooling defect types, identification evidence chains, risk levels, and handling strategies are generated. This invention can improve the identification priority, robustness, and targeted handling of critical cooling defects such as fan stoppage, cooler activation failure, standby cooler non-response, severe radiator contamination, and persistent insufficient cooling effect, while keeping the false alarm rate under control.

[0173] This invention employs two strongly correlated modes—temperature mode and cooling state mode—as core identification inputs, reducing system complexity and engineering deployment difficulties caused by excessive fusion of multi-source data. By constructing temperature-cooling response consistency features, it transforms the traditional method of simply judging "whether the temperature exceeds the limit" into an association identification method that judges "whether the temperature change matches the cooling device response," thereby reducing false alarms caused by factors such as normal load increases and ambient temperature changes, and improving the ability to identify early cooling defects. By constructing a priori key cooling defect pattern library, it includes fan stoppage, cooler activation failure, and standby cooler non-response. High-risk defects such as severely dirty radiators and persistently insufficient cooling are prioritized for identification. Combined with the sub-Bruker identification model, the stability of key cooling defect detection results is maintained under conditions of load variation, ambient temperature variation, radiator switching variation, and measurement noise variation. Through a multi-agent collaborative identification mechanism, the results of primary screening, temperature expert identification, cooling expert identification, key defect matching, and case verification are integrated to improve the credibility and interpretability of the identification results. The identification results are then transformed into risk level, handling intensity, and handling strategy, realizing closed-loop management from dual-modal data collection and intelligent identification to operation and maintenance handling.

[0174] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0175] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of cooling defects in a main transformer, characterized in that, Includes the following steps: Real-time acquisition of temperature monitoring data, cooling device monitoring data, and operating condition correction data of the main transformer; establishment of temperature mode dataset and cooling state mode dataset. Temperature state features, cooling state features, and temperature-cooling response consistency features are extracted from the temperature modal dataset and cooling state modal dataset to obtain a dual-modal inspection state feature vector, and a comprehensive score for temperature anomaly and a comprehensive score for cooling state anomaly are calculated. Using historical normal samples of the dual-modal inspection state feature vector, the current dual-modal inspection state feature vector, the comprehensive score of temperature anomaly and the comprehensive score of cooling state anomaly, and combined with the prior critical cooling defect pattern library, the sub-Bruk bar identification parameters are calculated. The sub-Bruk bar identification parameters include the critical defect priority identification score and the anomaly evaluation function. The prior critical cooling defect pattern library contains multiple types of critical cooling defect patterns. Each type of critical cooling defect pattern includes a defect name, defect feature contour, defect feature weight, evidence chain requirements and handling rules. The defect feature contour is composed of parameters in the dual-modal inspection state feature vector. Using different identification agents, collaborative identification is performed based on the current dual-modal inspection state feature vector, the prior key cooling defect pattern library, the key defect priority identification score, and the anomaly evaluation function to obtain the cooling defect type, identification confidence, anomaly evaluation value, and identification evidence chain. Based on the cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, calculate the risk value and handling intensity. Determine the defect risk level based on the risk value, and match the handling intensity with the handling rules in the prior critical cooling defect pattern library to obtain the handling strategy.

2. The intelligent identification method for cooling defects in main transformers according to claim 1, characterized in that, When extracting temperature state features, cooling state features, and temperature-cooling response consistency features from the temperature modal dataset and cooling state modal dataset, the specific steps include: calculating oil temperature anomaly, winding temperature anomaly, and the deviation between the field temperature and the background temperature based on the temperature modal dataset, and using these together with the ambient temperature and main transformer load level in the temperature modal dataset, as well as the oil temperature change rate, winding temperature change rate, and temperature anomaly duration in the temperature monitoring data, as temperature state features; calculating cooler activation rate, fan effective operating rate, fan current deviation, and radiator contamination index based on the cooling state modal dataset, and using these together with the cooler defect signal, standby cooler activation status, and standby cooler response time in the cooling state modal dataset as cooling state features; weighting the cooler activation rate, fan effective operating rate, radiator contamination index, and fan current deviation to calculate the expected cooling response quantity, calculating the oil temperature-cooling response inconsistency and winding temperature-cooling response inconsistency based on the expected cooling response quantity, and calculating the oil temperature response before and after cooling activation, the winding temperature response before and after cooling activation, and the temperature-cooling state dual-modal comprehensive inconsistency index, all of which are used as temperature-cooling response consistency features.

3. The intelligent identification method for cooling defects in main transformers according to claim 2, characterized in that: The oil temperature anomaly is calculated by subtracting the reference oil temperature under the same load and ambient temperature from the upper layer oil temperature in the temperature modal dataset, and then dividing by the difference between the allowable limit of the upper layer oil temperature and the reference oil temperature. The winding temperature anomaly is calculated by subtracting the reference winding temperature from the winding temperature in the temperature mode dataset, and then dividing by the difference between the allowable limit of the winding temperature and the reference winding temperature. The deviation between the field temperature and the background temperature is the absolute value of the difference between the field thermometer reading and the background monitored temperature in the temperature modal dataset; The cooler utilization rate is the ratio of the number of cooler groups already in operation to the total number of cooler groups in the cooling state modal dataset. The effective fan operating rate is the ratio of the actual number of operating fans in the cooling state modal dataset to the number of fans commanded to be activated; The fan current deviation is the absolute value of the difference between the fan motor current and the fan rated operating current in the cooling state modal data set, divided by the fan rated operating current. The radiator fouling index is the ratio of the fouled area of ​​the radiator to the effective heat dissipation area of ​​the radiator in the cooling state modal data. The oil temperature-cooling response inconsistency is the result of subtracting the expected cooling response from the normalized oil temperature change rate. The normalized oil temperature change rate is the result of subtracting the oil temperature change rate benchmark value under normal operating conditions from the oil temperature change rate in the temperature modal dataset, and then dividing by the difference between the oil temperature change rate warning limit and the benchmark value. The winding temperature-cooling response inconsistency is calculated by subtracting the expected cooling response from the normalized winding temperature change rate. The normalized winding temperature change rate is calculated by subtracting the winding temperature change rate reference value under normal operating conditions from the winding temperature change rate in the temperature modal dataset, and then dividing by the difference between the winding temperature change rate warning limit and the reference value. The oil temperature response before and after cooling is calculated by subtracting the upper oil temperature before cooling from the upper oil temperature after cooling is applied. The winding temperature response before and after cooling is calculated by subtracting the winding temperature before cooling from the winding temperature after cooling is applied. The temperature-cooling state dual-modal comprehensive inconsistency index is the result of a weighted sum of the oil temperature-cooling response inconsistency, the winding temperature-cooling response inconsistency, the oil temperature response before and after cooling input, the winding temperature response before and after cooling input, the radiator contamination index, and the degree of ineffective fan operation. The sum of the degree of ineffective fan operation and the effective fan operation rate is 1.

4. The intelligent identification method for cooling defects in main transformers according to claim 3, characterized in that, The mathematical expressions for calculating the comprehensive score of temperature anomaly and the comprehensive score of cooling condition anomaly are as follows: in, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. Indicates the degree of oil temperature abnormality. Indicates the degree of winding temperature anomaly. This represents the normalized rate of change of oil temperature. This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly. Indicates the cooler utilization rate. Indicates the effective operating rate of the fan. Indicates fan current deviation. Indicates the radiator's dirtiness index. This indicates a cooler defect signal. This indicates the normalized standby cooler response time. to This indicates the weighting of temperature anomaly assessment for different parameters. to This represents the weighting of the cooling status evaluation for different parameters.

5. The intelligent identification method for cooling defects in main transformers according to claim 4, characterized in that, When calculating the sub-Bruker identification parameters using historical normal samples of dual-modal inspection state feature vectors, current dual-modal inspection state feature vectors, comprehensive scores of temperature anomalies and cooling state anomalies, combined with a priori critical cooling defect pattern library, the following steps are included: An empirical distribution is constructed based on historical normal samples, expressed as follows: in This represents the historical normal sample empirical distribution. Indicates concentration on the sample Point mass distribution at [location] This represents the i-th historical normal sample. This indicates the consistency characteristics of temperature-cooling response under normal conditions; Construct a set with an uncertain distribution, as shown in the following expression: in, Represents a set with an uncertain distribution. This represents the actual operational disturbance distribution. This represents the distance between the actual operational disturbance distribution and the historical normal sample empirical distribution. Indicates the allowable distribution offset radius; Using the defect feature contours in the prior critical cooling defect pattern library as prior defect contours, the matching degree between the current state and the i-th type of critical defect is calculated, as shown in the following expression: in, This represents the degree of matching between the current state and the i-th type of critical defect pattern. Represents the feature matching function. Represents the feature profile of the i-th type of defect. This represents the feature vector of the dual-modal inspection state; The robustness index for critical defect detection is calculated using the following expression: in, This represents the robustness index for detecting the i-th type of critical defect; The critical defect priority detection score is calculated using the following expression: in, This indicates the priority detection score for the i-th type of critical defect. This represents the risk weight of the i-th type of critical defect. This represents the robustness weight of the i-th type of critical defect. This represents the distribution offset risk correction term. Indicates the risk weight of distribution offset; The anomaly evaluation function and its corresponding constraints are constructed as follows: in, This indicates the anomaly evaluation value of the current state. This represents the maximum score among all critical defect priority detection scores. This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. , , , These represent the weight coefficients of the corresponding evaluation items. Indicates the combined inconsistency index of temperature-cooling state dual modes; Indicates the threshold for abnormal alarms. Indicates the maximum permissible false alarm rate. Indicates normal operating status; This represents the prior defect profile of the i-th type of critical cooling defect in the prior critical cooling defect pattern library. This represents the feature vector of the current dual-modal inspection state. This represents the actual operational disturbance distribution. This represents the anomaly evaluation function. This represents the expected minimum detection rate for the i-th type of critical cooling defect. This indicates the loss in detection performance caused by the offset of the operational disturbance distribution.

6. The intelligent identification method for cooling defects in main transformers according to claim 5, characterized in that, Different identification agents include a primary screening agent, a temperature expert identification agent, a cooling expert identification agent, a priori critical defect matching agent, and a case verification agent. When using these different identification agents for collaborative identification based on the current bimodal inspection state feature vector, a priori critical cooling defect pattern library, critical defect priority identification scoring, and anomaly evaluation function, the specific methods include: The primary screening agent performs initial screening of anomaly types based on the oil temperature anomaly degree, winding temperature anomaly degree, temperature-cooling state dual-modal comprehensive inconsistency index, fan ineffective operation degree, and anomaly evaluation function of the dual-modal inspection state feature vector, and outputs the confidence score of the primary screening. When the confidence score of the primary screening is greater than the preset threshold, the temperature expert identification agent, the cooling expert identification agent, the prior key defect matching agent, and the case verification agent are triggered to perform parallel collaborative identification. The temperature expert identification agent identifies temperature defects and outputs the temperature expert identification confidence level based on the oil temperature anomaly degree, winding temperature anomaly degree, normalized oil temperature change rate, normalized winding temperature change rate, temperature anomaly duration, and field-background temperature deviation of the dual-modal inspection state feature vector. The cooling expert identification agent identifies cooling defects based on the cooler availability rate, fan effective operating rate, fan current deviation, radiator dirt index, cooler defect signal, standby cooler availability status, and normalized standby cooler response time of the dual-modal inspection state feature vector, and outputs the cooling expert identification confidence score. The prior critical defect matching agent outputs the most likely critical defect type and the prior critical defect matching confidence level based on the critical defect priority identification score, and takes the most likely critical defect type as the cooling defect type. The case verification agent outputs the case verification confidence score based on the similarity between the current bimodal inspection state feature vector and the historical case features; The outputs of each agent are weighted and fused to obtain the identification confidence level.

7. The intelligent identification method for cooling defects in main transformers according to claim 6, characterized in that, The mathematical expression for identifying confidence level is as follows: in, This indicates the final identification confidence level. Indicates the confidence level of the initial screening. Indicates the reliability of temperature expert identification; Indicates the confidence level of the cooling expert's identification; This indicates the prior critical defect matching confidence level; This indicates the confidence level of the case validation. to These represent the fusion weights of the identification results of each agent; The mathematical expression for the confidence level of the primary screening is as follows: In the formula, Indicates the confidence level of the initial screening; This represents the normalization function, used to map the scores to the [0,1] interval; Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; M represents the temperature-cooling state dual-modal comprehensive inconsistency index; Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; This represents the anomaly evaluation function; to This represents the weight coefficient of each evaluation item in the primary screening agent; The mathematical expression for the reliability of temperature expert identification is as follows: in, This indicates the confidence level of the temperature expert's identification. Represents the normalization function; Indicates the degree of oil temperature abnormality; Indicates the degree of winding temperature anomaly; This represents the normalized rate of change of oil temperature; This represents the normalized rate of change of winding temperature. Indicates the duration of the temperature anomaly; This indicates the deviation between the ambient temperature and the background temperature. to This represents the weighting coefficients of each evaluation item in the temperature expert's identification agent; The mathematical expression for the confidence level identified by cooling experts is as follows: in, This indicates the confidence level of the cooling expert's identification. Represents the normalization function; Indicates the cooler utilization rate; This indicates that the cooler is not fully engaged. Indicates the effective operating rate of the fan; This indicates the degree to which the fan is not operating effectively; Indicates fan current deviation; Indicates the radiator's dirtiness index; This indicates a cooler defect signal; This indicates that the standby cooler is in operation; This indicates that the standby cooler is not in operation; This indicates the normalized standby cooler response time; to This represents the weighting coefficients of each evaluation item in the intelligent agent identified by the cooling expert; The mathematical expression for the prior critical defect matching confidence is as follows: in, represents the priority detection score for the i-th type of critical cooling defect, and K represents the total number of critical cooling failure modes; The mathematical expression for the confidence level in case validation is as follows: in, Indicates the confidence level of the case validation; This represents the case similarity threshold. This indicates the similarity between the current case and historical cases. This represents the feature vector of the historical case state.

8. The intelligent identification method for cooling defects in main transformers according to claim 7, characterized in that, When calculating the risk value and handling intensity based on the aforementioned cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, the step of calculating the risk value is included, specifically: Determine the corresponding chain of evidence requirements in the prior critical cooling defect pattern library based on the type of cooling defect; The value of evidence chain integrity is obtained by dividing the number of valid evidence in the identified evidence chain by the number of evidence required for the corresponding evidence chain in the prior critical cooling defect pattern library. The risk value is calculated by weighting the evidence chain integrity value, temperature anomaly comprehensive score, cooling state anomaly comprehensive score, anomaly evaluation value, current dual-modal inspection state feature vector, and prior critical cooling defect pattern library. The mathematical expression is as follows: Where R represents the cooling defect risk value, This indicates a comprehensive score for temperature anomalies. The overall score indicates an abnormal cooling condition. This represents the risk weight of the i-th type of critical defect in the prior critical cooling defect pattern library. This represents the combined inconsistency index of the temperature-cooling state dual-modal characteristics in the current dual-modal inspection state feature vector. This represents the load correction factor. This represents the ambient temperature correction factor. This indicates the anomaly evaluation value of the current state. A value representing the completeness of the chain of evidence.

9. The intelligent identification method for cooling defects in main transformers according to claim 8, characterized in that, When calculating the risk value and handling intensity based on the cooling defect type, anomaly evaluation value, identification evidence chain, critical defect priority identification score, temperature anomaly comprehensive score, and cooling state anomaly comprehensive score, the step of calculating the handling intensity includes calculating the weighted sum of the risk value, critical defect priority identification score, and evidence chain integrity value to obtain the handling intensity value.

10. A smart identification system for cooling defects in a main transformer, characterized in that, The device includes an interconnected processor and a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and the computer program is executed by the processor to implement the steps of the intelligent identification method for cooling defects of the main transformer according to any one of claims 1 to 9.

Images