Expert diagnosis mechanism method and system for digital dry-type transformer

Abstract

The invention belongs to the technical field of transformer diagnosis, and discloses an expert diagnosis mechanism method for a digital dry-type transformer, which can continuously monitor the running state of the transformer for 24 hours, and can reserve data to facilitate tracing, analysis and summarization after problems. And the overall and local operation states of the transformer are checked in combination with a 3D visualization effect. The danger caused by manual operation and maintenance is eliminated; and the manual operation and maintenance cost is reduced. And hardware loss and economic loss caused by abnormal conditions are reduced. Digitization and informatization transformation are achieved, and intelligent operation and maintenance are completed. A large model, machine learning and other novel science and technology are introduced, and a traditional mode is enabled.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer diagnosis, and in particular relates to an expert diagnosis mechanism method and system for digital dry-type transformers. Background Art

[0002] Transformers operate in a complex, high-voltage environment. Traditional manual inspections cannot monitor their operating status at all times, and operation and maintenance inspections carry safety risks. Digital transformers can monitor the operating status of transformers 24 hours a day and detect potential risks in a timely manner.

[0003] In digital transformers, manual operation and maintenance requires consideration of many factors, including noise, vibration, temperature, and other factors. It is difficult for ordinary operation and maintenance personnel to control whether their data is in a healthy state for the transformer. This patent can integrate these many factors and unify them to display the operating status of the transformer in a simple and clear way with a health score and estimated lifespan, reducing labor costs and operation and maintenance difficulties.

[0004] In today's world of ever-changing technology, the capabilities of large models need to be integrated into the power industry. This patent, through the empowerment of large models, can provide expert opinions for transformer diagnosis and guide operation and maintenance personnel to complete operation and maintenance work.

[0005] In the display of ordinary systems, there are multiple media display methods such as graphics, tables, curves, etc. This patent uses a 3D model to build a virtual model of the transformer, which can more intuitively view the overall and local states of the transformer.

[0006] Through the above analysis, the problems and defects of the existing technology are as follows:

[0007] (1) Traditional operation and maintenance methods consume a lot of manpower costs, and inspections inside transformers often involve dangerous operations such as meter reading, which cannot guarantee the safety of personnel.

[0008] (2) Traditional operation and maintenance methods cannot record historical status and data, cannot automatically detect abnormal situations, and can only complete operation and maintenance work through inefficient manual inspections.

[0009] (3) The traditional method of judging the operating status of the transformer is based on the experience of the operation and maintenance personnel, and the judgment level is also different, depending on the operation and maintenance experience of the operation and maintenance personnel on the transformer.

[0010] (4) Most traditional transformer status identification methods are based on rule verification and logical judgment. They do not use the massive historical data of the transformer's own operation to complete machine learning, and the identification method is relatively simple. Summary of the Invention

[0011] In view of the problems existing in the prior art, the present invention provides an expert diagnosis mechanism method for digital dry-type transformers.

[0012] The present invention is implemented as follows: an expert diagnosis mechanism method for digital dry-type transformers, comprising:

[0013] Step 1: Digital transformation of traditional transformers;

[0014] 1) Sensor types for digital dry-type transformers: This invention stipulates that the types of sensor equipment installed on digital transformers include vibration, partial discharge, arc, temperature, and noise sensors, and all sensors must support the Modbus communication protocol to ensure data collection and transmission with the system;

[0015] 2) Data collection and protocol conversion are implemented, and the transformer's sensor data is uploaded to the diagnostic platform. Through built-in algorithms and decision libraries, the sensor data is modeled and analyzed. By combining the generated model file with real-time data, the transformer's status can be diagnosed. A 3D model is built to display the sensor's real-time data, completing the transformer's digital transformation.

[0016] Step 2: Implement the diagnostic system using IoT technology;

[0017] Step 3, simulation model construction;

[0018] Step 4: Build the data model;

[0019] Step 5: Online diagnosis and lifespan estimation;

[0020] Step 6, expert opinion based on DeepSeek.

[0021] Furthermore, the diagnosis system is realized by using Internet of Things technology:

[0022] The design of the IoT system is divided into three layers: the perception layer, the edge terminal layer, and the platform layer;

[0023] Perception layer: Deploy sensors installed on transformers, including vibration, noise, partial discharge, arc, and temperature sensors; provide standard industrial communication protocols such as Modbus and RS-485;

[0024] Edge terminal layer: Sensors with local computing capabilities can collect data at high speeds and in real time. After preliminary cleaning and preprocessing of the data locally, the data is assembled into messages that conform to the unified protocol defined by the platform and uploaded to the platform.

[0025] Platform layer: provides rich application functions, including automatic meter reading, periodic meter reading, health diagnosis, life assessment, and data curves; helps operation and maintenance personnel to carry out inspections, data recording, and data analysis.

[0026] Furthermore, the simulation model is constructed:

[0027] a) Decompose the transformer model into core components: the iron core, windings, and insulation layer, and model the core components. Use metal texture mapping for the iron core to reflect the material properties of the silicon steel sheet. Simulate the gloss of the copper conductor for the windings and superimpose the current density texture.

[0028] b) Add dynamic change rules to static models through real-time parameters. For example, when the temperature rises, the corresponding area will heat up and change color. The vertex shader can be used to realize magnetic flux line animation; the electrical parameters and 3D materials can be linked in real time.

[0029] Furthermore, the data model is constructed

[0030] a) Implement feature engineering calculations for various sensor data, perform FFT spectrum analysis for vibration sensors, perform phase distribution pattern recognition for partial discharge sensors, implement pulse duration statistics for arc light, perform voiceprint feature extraction for noise, and calculate slope change for temperature; perform normalization processing (0-1 standardization) on all data;

[0031] b) After normalization, load the decision library configured by the system, obtain the weight configuration from the decision library, and multiply the normalized data by the weight parameter to make the random forest more inclined to split the node, thereby affecting the output of the health score;

[0032] c) The transformer status is classified into four categories (good, better, fair, and poor), and the random forest algorithm is selected from the classification algorithm. The historical operating data of the sensor is obtained. To set the final status for each data packet, the present invention proposes that DeepSeek intervene in the scoring, and DeepSeek assigns a score between 0 and 100 to each data packet and labels it with a score. The scored data is shuffled in order, 70% of the data is selected for model training, and 30% of the data is used for test evaluation (if DeepSeek does not have or has weak scoring capabilities, the llama factory can be used in combination with a customized diagnostic expert knowledge base to train the DeepSeek-xxB model to achieve scoring capabilities. This technical implementation is not described in this patent).

[0033] d) Model evaluation: Use mean square error and root mean square error to evaluate the model. The smaller the two indicators, the better the model training. Save the trained model with an overall accuracy higher than 85% and replace the model file in the project to complete the construction of the data model.

[0034] Furthermore, the online diagnosis and lifespan prediction:

[0035] a) The system sets a scheduled schedule to perform health diagnosis and life assessment on the transformer every five minutes. By acquiring real-time data from various sensors, the model calculates the final output health index;

[0036] b) After calculating the health score, calculate the remaining life of the equipment using the health score; use Weibull distribution fitting to estimate the life. Define the shape parameter beta and scale parameter eta used by Weibull. Set beta to 1.5 and the scale parameter based on the service life of the transformer. Its estimated life is X, and the calculation formula is as follows:

[0037] X=Wblinv(P,A,B)

[0038] P: represents the failure probability, which is the ratio of the current health score to the total health score;

[0039] A: Weibull scale parameter;

[0040] B: Weibull shape parameter.

[0041] Furthermore, the expert opinions based on DeepSeek:

[0042] DeepSeek is used to implement guidance after diagnosis; the guidance depends on the prompt words.

[0043] Another object of the present invention is to provide an expert diagnosis mechanism system for digital dry-type transformers, comprising:

[0044] Transformation module, used for digital transformation of traditional transformers;

[0045] IoT module, used to implement diagnostic systems using IoT technology;

[0046] Simulation module, used for building simulation models;

[0047] Data model building module, used for data model building;

[0048] Life prediction module, used for online diagnosis and life prediction;

[0049] Expert module for expert opinions based on DeepSeek.

[0050] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the expert diagnosis mechanism method for digital dry-type transformers.

[0051] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the expert diagnosis mechanism method for digital dry-type transformers.

[0052] Another object of the present invention is to provide an information data processing terminal, which is used to implement the expert diagnosis mechanism system for digital dry-type transformers.

[0053] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0054] This invention achieves 24-hour, uninterrupted monitoring of the operating status of dry-type transformers by building a continuous monitoring system based on the Internet of Things and machine learning, and structures the storage of all monitoring data. This data retention mechanism not only provides complete and accurate basic information for future fault tracing and statistical analysis, but also enables timely identification of deviations in operating trends by comparing real-time data with historical data, providing early warning of potential anomalies, significantly improving equipment reliability and operational safety.

[0055] This invention uses a 3D visualization engine to digitally reconstruct the entire transformer and key components (such as windings, insulation, and heat dissipation channels). Users can intuitively view the distribution and changes of various monitoring indicators in three-dimensional space. The real-time display of temperature, humidity, and partial discharge point distribution through a visualization panel not only improves the accuracy of fault location but also provides a more user-friendly interface for operation and maintenance personnel, making operation and maintenance decisions more intuitive and efficient.

[0056] This invention deeply integrates intelligent monitoring with automated operations and maintenance, replacing traditional manual inspections. This eliminates the dangerous work scenarios associated with on-site maintenance, such as overhead work and live-line maintenance, significantly reducing labor costs and safety risks. Furthermore, when the system detects an abnormality, it automatically triggers an alarm and an emergency work order, promptly notifying operations and maintenance personnel and assisting them in developing a rapid response plan, minimizing hardware and financial losses caused by equipment anomalies.

[0057] Based on digitalization and informatization, this invention introduces a large model (DeepSeek) and deep learning algorithms to establish a hybrid inference framework for dry-type transformer health diagnosis and lifespan assessment, combining expert knowledge with data-driven reasoning. The system uses DeepSeek to automatically score massive amounts of historical operation and maintenance data and uses high-quality samples as training sets for machine learning models, significantly reducing manual annotation costs. Furthermore, through continuous online learning and model iteration, diagnostic accuracy is continuously optimized, overcoming the technical bias of existing technologies that rely on manual judgment and experience, demonstrating significant innovation and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of an expert diagnosis mechanism method for digital dry-type transformers provided by an embodiment of the present invention.

[0059] Figure 2 This is a system structure diagram of an expert diagnosis mechanism for digital dry-type transformers provided by an embodiment of the present invention.

[0060] Figure 3 This is a structural diagram of a digital transformer provided by an embodiment of the present invention.

[0061] Figure 4 This is a diagram of an Internet of Things system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] like Figure 1 As shown, an embodiment of the present invention provides an expert diagnosis mechanism method for a digital dry-type transformer, comprising the following steps:

[0064] S101, digital transformation of traditional transformers;

[0065] 1) Sensor types for digital dry-type transformers: This invention stipulates that the types of sensor equipment installed on digital transformers include vibration, partial discharge, arc, temperature, and noise sensors, and all sensors must support the Modbus communication protocol to ensure data collection and transmission with the system;

[0066] 2) Data collection and protocol conversion are implemented, and the transformer's sensor data is uploaded to the diagnostic platform. Through built-in algorithms and decision libraries, the sensor data is modeled and analyzed. By combining the generated model file with real-time data, the transformer's status can be diagnosed. A 3D model is built to display the sensor's real-time data, completing the transformer's digital transformation.

[0067] S102, using IoT technology to implement a diagnostic system;

[0068] S103, simulation model construction;

[0069] S104, data model construction;

[0070] S105, online diagnosis and lifespan prediction;

[0071] S106, based on DeepSeek's expert opinion.

[0072] By deeply sensing multiple physical quantities such as partial discharge, vibration, temperature rise and acoustic noise in traditional dry-type transformers under operating conditions, and using the Modbus bus to achieve unified data access, the limitations of previous isolated signal acquisition have been completely broken, the real-time capture capability of key fault signs has been enhanced, and the diagnostic platform can obtain multi-dimensional holographic representation at the raw data level, fundamentally improving the technical bottlenecks of monitoring blind spots and delayed diagnosis.

[0073] The edge node completes parallel preprocessing of multi-sensor data under the high-bandwidth Modbus / RS-485 network, uses efficient time-domain feature filtering and wavelet denoising algorithms to eliminate noise interference on the hardware side, and packages the cleaned original waveform and normalized parameters into standardized messages and uploads them to the cloud, ensuring that subsequent modeling and simulation will not deviate due to data quality issues.

[0074] The 3D simulation module has a built-in fine-grained transformer structure model, applies metal texture mapping to the core silicon steel sheets, and uses a vertex shader to render magnetic flux line animation. At the same time, it maps local surface colors according to the real-time temperature slope, realizing the visualization of electric-thermal-magnetic multi-field coupling, effectively making up for the shortcomings of traditional 2D curve analysis in spatial distribution, and providing an intuitive basis for fault location.

[0075] Based on the spectrum and phase decoupling strategy of feature engineering, the vibration signal is shifted into the frequency domain through FFT to observe the harmonic components. The partial discharge signal relies on phase distribution pattern recognition to locate the discharge source. Arc light and noise statistics are used to extract pulse width distribution and soundprint features respectively. All indicators are normalized to zero and then multiplied with the preset decision library weights to construct a data evaluation model deeply coupled with the random forest classifier.

[0076] The online diagnostic engine automatically triggers the health index calculation every five minutes, and fits the real-time output status score with the historical health curve using the Weibull inverse function, dynamically calculating the remaining lifespan. This provides closed-loop support for operation and maintenance scheduling from second-level monitoring to year-level lifespan prediction, greatly improving the scientific nature of risk prevention and maintenance plans.

[0077] At the expert opinion generation level, the DeepSeek deep learning framework is connected to a dedicated knowledge graph. Through a prompt-word driven reasoning mechanism, it automatically combines and outputs diagnostic results with maintenance, inspection, and efficiency improvement suggestions. It can not only propose bracket reinforcement plans for local fatigue areas, but also perform systematic operation and maintenance optimization for long-term operating characteristics, achieving a leap from single fault diagnosis to a closed-loop operation and maintenance for the entire life cycle.

[0078] The embodiment of the present invention provides a diagnostic system using Internet of Things technology:

[0079] The design of the IoT system is divided into three layers: the perception layer, the edge terminal layer, and the platform layer;

[0080] Perception layer: Deploy sensors installed on transformers, including vibration, noise, partial discharge, arc, and temperature sensors; provide standard industrial communication protocols such as Modbus and RS-485;

[0081] Edge terminal layer: Sensors with local computing capabilities can collect data at high speeds and in real time. After preliminary cleaning and preprocessing of the data locally, the data is assembled into messages that conform to the unified protocol defined by the platform and uploaded to the platform.

[0082] Platform layer: provides rich application functions, including automatic meter reading, periodic meter reading, health diagnosis, life assessment, and data curves; helps operation and maintenance personnel to carry out inspections, data recording, and data analysis.

[0083] The simulation model provided by the embodiment of the present invention is constructed as follows:

[0084] a) Decompose the transformer model into core components: the iron core, windings, and insulation layer, and model the core components. Use metal texture mapping for the iron core to reflect the material properties of the silicon steel sheet. Simulate the gloss of the copper conductor for the windings and superimpose the current density texture.

[0085] b) Add dynamic change rules to static models through real-time parameters. For example, when the temperature rises, the corresponding area will heat up and change color. The vertex shader can be used to realize magnetic flux line animation; the electrical parameters and 3D materials can be linked in real time.

[0086] The data model provided by the embodiment of the present invention is constructed

[0087] a) Implement feature engineering calculations for various sensor data, perform FFT spectrum analysis for vibration sensors, perform phase distribution pattern recognition for partial discharge sensors, implement pulse duration statistics for arc light, perform voiceprint feature extraction for noise, and calculate slope change for temperature; perform normalization processing (0-1 standardization) on all data;

[0088] b) After normalization, load the decision library configured by the system, obtain the weight configuration from the decision library, and multiply the normalized data by the weight parameter to make the random forest more inclined to split the node, thereby affecting the output of the health score;

[0089] c) The transformer status is classified into four categories (good, better, fair, and poor), and the random forest algorithm is selected from the classification algorithm. The historical operating data of the sensor is obtained. To set the final status for each data packet, the present invention proposes that DeepSeek intervene in the scoring, and DeepSeek assigns a score between 0 and 100 to each data packet and labels it with a score. The scored data is shuffled in order, 70% of the data is selected for model training, and 30% of the data is used for test evaluation (if DeepSeek does not have or has weak scoring capabilities, the llama factory can be used in combination with a customized diagnostic expert knowledge base to train the DeepSeek-xxB model to achieve scoring capabilities. This technical implementation is not described in this patent).

[0090] d) Model evaluation: Use mean square error and root mean square error to evaluate the model. The smaller the two indicators, the better the model training. Save the trained model with an overall accuracy higher than 85% and replace the model file in the project to complete the construction of the data model.

[0091] Online diagnosis and lifespan prediction provided by embodiments of the present invention:

[0092] a) The system sets a scheduled schedule to perform health diagnosis and life assessment on the transformer every five minutes. By acquiring real-time data from various sensors, the model calculates the final output health index;

[0093] b) After calculating the health score, calculate the remaining life of the equipment using the health score; use Weibull distribution fitting to estimate the life. Define the shape parameter beta and scale parameter eta used by Weibull. Set beta to 1.5 and the scale parameter based on the service life of the transformer. Its estimated life is X, and the calculation formula is as follows:

[0094] X=Wblinv(P,A,B)

[0095] P: represents the failure probability, which is the ratio of the current health score to the total health score;

[0096] A: Weibull scale parameter;

[0097] B: Weibull shape parameter.

[0098] The DeepSeek-based expert opinions provided by the embodiments of the present invention:

[0099] DeepSeek is used to implement guidance opinions after diagnosis; the guidance opinions depend on the prompt words.

[0100] like Figure 2As shown, an embodiment of the present invention provides an expert diagnosis mechanism system for digital dry-type transformers, including:

[0101] Transformation module, used for digital transformation of traditional transformers;

[0102] IoT module, used to implement diagnostic systems using IoT technology;

[0103] Simulation module, used for building simulation models;

[0104] Data model building module, used for data model building;

[0105] Life prediction module, used for online diagnosis and life prediction;

[0106] Expert module for expert opinions based on DeepSeek.

[0107] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the expert diagnosis mechanism method for digital dry-type transformers.

[0108] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the expert diagnosis mechanism method for digital dry-type transformers.

[0109] Another object of the present invention is to provide an information data processing terminal, which is used to implement the expert diagnosis mechanism system for digital dry-type transformers.

[0110] The present invention is specifically implemented:

[0111] Digital dry-type transformer: A transformer that does not use insulating oil and is cooled by natural cooling or air cooling. By adding a variety of specific sensors, traditional transformers are transformed into digital transformers.

[0112] Random Forest Model: A decision tree-based ensemble learning algorithm that improves the accuracy and robustness of the model by building multiple trees and integrating their prediction results.

[0113] Decision library: stores the rules or decision tables required for data processing, and influences the final score determination through decision weights.

[0114] Prompt: Specific experts analyze prompt words, and the large model gives expert diagnostic opinions.

[0115] Figure 3This is the structure diagram of the digital transformer

[0116] Adding multiple sensors to traditional dry-type transformers, such as temperature, partial discharge, arc, vibration, and noise sensors, can clearly represent the transformer's surface characteristics. By sending data from these sensors to the diagnostic platform, online diagnosis of the digital transformer can be achieved through the platform's health diagnosis and life assessment functions.

[0117] Digital transformation of traditional transformers

[0118] Sensor types for digital dry-type transformers: This invention stipulates that the types of sensor equipment installed on digital transformers include vibration, partial discharge, arc, temperature, noise and other sensors, and all sensors need to support communication protocols such as Modbus to ensure data collection and transmission with the system.

[0119] Data collection and protocol conversion are implemented, and transformer sensor data is uploaded to the diagnostic platform. Using built-in algorithms and a decision library, sensor data is modeled and analyzed. By combining the generated model file with real-time data, the transformer's condition can be diagnosed. A 3D model is constructed to display real-time sensor data, completing the digital transformation of the transformer.

[0120] Implementing diagnostic systems using IoT technology

[0121] like Figure 4 The design of the Internet of Things system is divided into three layers: perception layer, edge terminal layer, and platform layer.

[0122] Perception layer: Deploy sensors installed on transformers, such as vibration, noise, partial discharge, arc, and temperature sensors. Provide standard industrial communication protocols such as Modbus and RS-485.

[0123] Edge terminal layer: Sensors with local computing capabilities can collect data at high speeds and in real time. After preliminary cleaning and preprocessing of the data locally, they are assembled into messages with a unified standard protocol defined by the platform and uploaded to the platform.

[0124] Platform layer: Provides a variety of application functions, such as automatic meter reading, periodic meter reading, health diagnosis, life assessment, data curves, etc. Helps operation and maintenance personnel to implement inspections, data recording, data analysis, etc.

[0125] Simulation model construction

[0126] The transformer model was decomposed into core components, including the core, windings, and insulation layer, and these core components were modeled. A metal texture map was used for the core to reflect the material properties of the silicon steel sheet; the windings were simulated using a copper conductor gloss map and a current density texture was superimposed.

[0127] Add dynamic change rules to static models through real-time parameters. For example, when the temperature rises, the corresponding area will heat up and change color. Use vertex shaders to realize magnetic flux line animation, etc. Realize real-time linkage between electrical parameters and 3D materials.

[0128] Data model construction

[0129] We perform feature engineering calculations on various sensor data, FFT spectrum analysis on vibration sensors, phase distribution pattern recognition on partial discharge sensors, pulse duration statistics on arc light, voiceprint feature extraction on noise, and slope change calculation on temperature. All data is normalized (0-1 standardization).

[0130] After normalization, the decision library of the system configuration is loaded, and the weight configuration is obtained from the decision library. By multiplying the normalized data by the weight parameter, the random forest is more inclined to split the nodes, thereby affecting the output of the health score.

[0131] The status of the transformer is divided into four types (good, better, average, and poor), so the random forest algorithm is selected in the classification algorithm. The historical operating data of the sensor is obtained. In order to set the final status for each packet of data, the present invention proposes that DeepSeek intervene in the scoring, and DeepSeek gives a score between 0-100 for each packet of data and adds a scoring label. The scored data is shuffled in order, 70% of the data is selected for model training, and 30% of the data is used for test evaluation (if DeepSeek does not have or has weak scoring capabilities, you can use llama factory combined with a custom diagnostic expert knowledge base to train the DeepSeek-xxB model to achieve scoring capabilities. This technical implementation is not described in this patent).

[0132] Model evaluation uses mean squared error and root mean squared error to evaluate the model. The smaller the two indicators, the better the model training. Save the trained model with an overall accuracy above 85% and replace the model file in the project to complete the data model construction.

[0133] Online diagnosis and lifespan prediction

[0134] The system is set up with a scheduled schedule to perform health diagnosis and life assessment on the transformer every five minutes. By acquiring real-time data from various sensors, the model calculates the final output health index. The health score rules are as follows:

[0135] Health Score Health status 0≤S<30 Poor 30≤S<60 generally 60≤S<80 better 80≤S≤100 good

[0136] After calculating the health score, the remaining life of the equipment is calculated using the health score. The life is estimated using a Weibull distribution fit. The shape parameter beta and scale parameter eta used by the Weibull distribution are defined. Beta is set to 1.5, and the scale parameter is set based on the age of the transformer. The estimated life is X, and the calculation formula is as follows:

[0137] X=Wblinv(P,A,B)

[0138] P: represents the failure probability, which is the ratio of the current health score to the total health score.

[0139] A: Weibull's scale parameter.

[0140] B: Weibull shape parameter.

[0141] Expert opinions based on DeepSeek

[0142] Health diagnosis is presented in the form of scores, which lacks guidance for operation and maintenance personnel. The present invention uses DeepSeek to implement guidance after diagnosis. Guidance depends on prompt words, which are defined as follows in the present invention:

[0143] ``````

[0144] #Character Setting

[0145] As a senior expert in the power industry, you have focused on dry-type transformer diagnosis for 20 years. You are familiar with standards such as IEEE C57.96 and IEC60076. You are good at combining multi-sensor data for cross-analysis, locating potential faults and proposing operational solutions.

[0146] #Input command

[0147] Please process the sensor data (temperature, vibration, noise, partial discharge, arc flash and comprehensive score) provided by the user according to the following structured process:

[0148] Step 1: Verify data integrity and mark abnormal parameters that exceed thresholds (reference standards: temperature > 130°C, partial discharge > 20pC, vibration velocity > 4.5mm / s);

[0149] Step 2: Analyze the correlation of multi-sensor data (e.g., whether high temperature is accompanied by an increase in partial discharge, and whether the vibration and noise spectra match);

[0150] Step 3: Sort by fault priority (arc > partial discharge > temperature > vibration > noise) to determine whether core risks such as insulation degradation and mechanical looseness are involved;

[0151] Step 4: Based on the comprehensive score (<60 points requires urgent intervention), output 3-5 professional opinions.

[0152] Output requirements:

[0153] Each comment must include the problem location, root cause, and recommended solutions (e.g., "replace the winding insulation layer," "tighten the bolts to 25 N·m").

[0154] References to industry standard clauses (example: “According to IEC 60599, DGA oil chromatography-assisted diagnosis is recommended”);

[0155] The language should be concise, the technical details should be clear, and the total word count should be ≤ 500 words.

[0156] ``````

[0157] Examples of DeepSeek expert opinions are as follows:

[0158] ``````

[0159] 1. Expert opinions on dry-type transformer diagnosis

[0160] Abnormally high temperature (138°C)

[0161] Cause location: Thermal aging of H-class insulation material (temperature rise rate >6°C / day for 5 consecutive days, exceeding the daily average temperature rise limit allowed by IEEE C57.96).

[0162] Disposal suggestions:

[0163] Immediately reduce the load to 85% of the rated capacity and use an infrared thermal imager to review the hotspot distribution within 72 hours (according to IEEE C57.96 Section 8.2);

[0164] If hot spots persist, it is recommended to replace the winding insulation within 1 month.

[0165] Abnormal partial discharge (28pC)

[0166] Correlation analysis: PD pulses clustered on the high-voltage side of phase B and were accompanied by 1kHz high-frequency vibration (amplitude 0.15mm), which is consistent with the characteristics of epoxy resin cracking (refer to the IEC 60599 partial discharge pattern library).

[0167] Disposal suggestions:

[0168] Use ultrasonic locator (frequency band 40-80kHz) to accurately calibrate the defect location;

[0169] Replace damaged coils within 3 months and perform a power frequency withstand voltage test (IEC 60270 requires ≥35kV / 1min).

[0170] Vibration exceeds the standard (5.2mm / s)

[0171] Root cause diagnosis: Loose clamp bolts (measured torque 18 N·m, lower than the 25 N·m specified in IEC 60076-11), causing core resonance.

[0172] Disposal suggestions:

[0173] Re-tighten the bolts to 25 N·m and apply thread locking agent (such as Loctite 243);

[0174] Monitor the vibration spectrum weekly to ensure that the amplitude in the frequency band below 2kHz is ≤3.8mm / s.

[0175] Comprehensive score 58 points

[0176] Risk rating: Critical failure of insulation system (estimated remaining life 8 months, based on temperature-partial discharge accelerated aging model).

[0177] Long-term strategy:

[0178] Start the procurement process of winding spare parts (delivery cycle needs to be less than 6 months);

[0179] Deploy an online partial discharge monitoring device (sampling rate ≥ 10MHz) to track the PD pulse phase distribution in real time (PRPD mode).

[0180] 2. Execution Priority and Standard Reference

[0181] Emergency operation: Immediately perform load shedding (complete today) and tighten bolts (within 24 hours);

[0182] Reference standards: IEEE C57.96 (temperature rise), IEC 60270 (partial discharge), IEC 60076-11 (mechanical strength);

[0183] Output verification: Total word count is 487 words, which meets technical specifications.

[0184] The large model outputs opinions and displays them on the system interface, providing guidance. This enables every operator to follow the guidance and avoids differences in operation and maintenance capabilities caused by different operation and maintenance levels.

[0185] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0186] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. An expert diagnosis mechanism method for digital dry-type transformers, characterized in that: The method comprises the following steps: Step 1: Digitally transform traditional dry-type transformers; Step 2: The IoT diagnostic system is implemented based on IoT technologies at the perception layer, edge terminal layer, and platform layer. Step 3: Simulation model construction: 3D simulation modeling of the core components of the dry-type transformer is carried out and real-time parameter linkage is achieved; Step 4: Data model construction performs feature engineering calculations and normalization on various sensor data, and implements state assessment by loading decision library weights and training random forest models; Step 5: The online diagnosis and life prediction system performs real-time health diagnosis according to the predetermined scheduling frequency and calculates the remaining life based on the inverse Weibull distribution function; Step 6: Expert opinions generate targeted diagnostic guidance based on DeepSeek.

2. The method according to claim 1, characterized in that Step one of the digital transformation involves installing vibration sensors, phase-distributed partial discharge sensors, arc sensors, temperature sensors, and noise sensors on dry-type transformers. These sensors all support the Modbus communication protocol to ensure reliable data collection and transmission.

3. The method according to claim 1, characterized in that Step 2: The implementation of the IoT diagnostic system includes deploying multiple types of sensors at the perception layer, performing preliminary data cleaning and standardized packaging at the edge terminal layer, and providing application functions such as automatic meter reading, regular meter reading, health diagnosis, and life assessment at the platform layer.

4. The method according to claim 1, characterized in that Step three of the simulation model building involves splitting the dry-type transformer into core, winding, and insulation layer components and establishing three-dimensional models for each. The core uses metal texture mapping to simulate material properties, and the winding uses copper material gloss texture. Dynamic rules are implemented through vertex shaders to achieve temperature color change and magnetic flux line animation.

5. The method according to claim 1, characterized in that: Step 4: Data model building includes fast Fourier transform spectrum analysis of vibration data, phase pattern recognition of partial discharge data, pulse duration statistics of arc data, voiceprint feature extraction of noise data, slope change calculation of temperature data, and then zero-to-one normalization of all data, and obtaining weights from the decision library and multiplying them with the normalized data for training the random forest model; the method also includes DeepSeek giving a score to each data as a label, selecting 70% training set and 30% test set and evaluating the model performance with mean square error and root mean square error.

6. The method according to claim 1, characterized in that Step five, online diagnosis and life estimation, involves the system obtaining real-time sensor data every five minutes to calculate the health score, and calling the inverse Weibull distribution function to determine the remaining life based on the ratio of the current health score to the total health score.

7. The method according to claim 1, characterized in that: Step 6: Expert opinion generation. The DeepSeek interactive module based on prompt words outputs targeted maintenance and repair guidance according to the diagnosis results.

8. An expert diagnosis mechanism system for digital dry-type transformers, characterized in that: The system includes a digital transformation module, an Internet of Things module, a simulation module, a data model module, an online diagnosis module and an expert opinion module. The digital transformation module is responsible for installing various types of sensors, the Internet of Things module is responsible for data collection and transmission, the simulation module is responsible for three-dimensional model construction and real-time linkage, the data model module is responsible for feature engineering and diagnostic model training, the online diagnosis module is responsible for health assessment and life expectancy estimation, and the expert opinion module is responsible for generating DeepSeek diagnostic guidance.

9. The system according to claim 8, characterized in that The digital transformation module is used to add vibration sensors, phase-distributed partial discharge sensors, arc sensors, temperature sensors and noise sensors to the dry-type transformer, and enable the sensors to upload collected data to the diagnostic platform through the Modbus protocol.

10. A non-volatile storage medium storing thereon an instruction program executable by a processor to implement the method of claim 1.

Images