An online monitoring device for transformer oil particles based on spectral and conductivity detection

Abstract

This invention provides an online monitoring device for particulate matter in transformer oil based on spectral and conductivity detection, enabling efficient, accurate, and continuous monitoring of particulate impurities in transformer oil. It includes a sampling module, a sensor module, an analysis and processing unit, and a communication module. The sampling module includes a micro-pump for continuous oil sample collection and an adjustable flow valve to ensure sample stability. The sensor module includes a spectral sensor for detecting optical properties in the oil sample and a conductivity sensor for monitoring changes in conductivity and detecting the electrical properties of different impurities in the oil. The microprocessor processes the sensor data and determines the impurity type and concentration. The communication module uses wireless transmission technology to send the impurity type and concentration obtained from the analysis and processing unit to the monitoring center in real time. This device is easy to operate, provides accurate and reliable detection results, and is suitable for online monitoring and maintenance of transformer oil.

Description

Technical Field

[0001] This invention relates to the field of power transformer testing technology, specifically to an online monitoring device for transformer oil particles based on spectral and conductivity detection. Background Technology

[0002] During operation, the insulating oil of power transformers becomes contaminated with various impurities. These impurities include tiny particles, moisture, gases, and other contaminants, all of which negatively impact the performance of the insulating oil. Particulate matter, in particular, significantly affects the transformer's insulation performance, reducing its withstand voltage level and increasing the risk of partial discharge, thus threatening the safe and stable operation of the transformer. The importance of keeping the insulating oil clean has long been recognized, but existing testing methods have many limitations.

[0003] Traditional testing methods typically rely on periodic sampling and laboratory analysis. This approach is not only complex and inefficient, but also fails to reflect the real-time condition of transformer oil. In practical applications, the operating environment of transformers is complex and variable, and the impurity content in the oil may change significantly within a short period. Failure to detect and address these changes in a timely manner can lead to transformer failures or even power system safety accidents.

[0004] Existing methods for detecting impurities in transformer oil primarily rely on offline detection technology, typically requiring multiple steps such as sampling, transportation, and laboratory analysis—a cumbersome and time-consuming process. Because it's offline, it struggles to reflect changes in impurities in the transformer oil in a timely manner, failing to meet the requirements for real-time monitoring of transformer operating status. Furthermore, existing detection equipment is usually bulky, expensive, and has high maintenance costs, making it unsuitable for field applications. It also requires professional operation, increasing labor costs and operational risks. Therefore, developing a device capable of real-time online detection of impurities in transformer oil is crucial. Such a device should possess high sensitivity and accuracy, continuously monitoring the state of the insulating oil without shutting down the transformer, and promptly detecting and warning of potential faults. This would not only improve the operational reliability of the transformer but also extend its service life and reduce maintenance and repair costs.

[0005] By using real-time online monitoring devices, the safety and efficiency of transformer operation will be significantly improved. The application of this technology will have a profound impact on the stable operation of the power system and represents an important direction for the development of modern power equipment monitoring technology. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide an online particle monitoring device for transformer oil, which can realize real-time detection of different impurity particles in the oil and has the advantages of low maintenance cost and simple operation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: an online monitoring device for transformer oil particles based on spectral and conductivity detection, the device comprising: a sampling module, a sensor module, an analysis and processing unit, and a communication module.

[0008] The sampling module achieves continuous oil sample collection through a micro-pump and adjustable flow valve, providing a stable and continuous supply of oil samples. The sensor module includes a spectral sensor and a conductivity sensor, used to detect the optical and electrical properties of the oil sample. Through multi-angle and multi-index detection methods, the accuracy and comprehensiveness of the detection are improved. The spectral sensor detects the optical properties of the oil sample, while the conductivity sensor monitors changes in the oil sample's conductivity and detects the electrical properties of different impurities. The analysis and processing unit integrates a microprocessor, enabling rapid processing and analysis of sensor data to accurately determine the type and concentration of impurities in the oil sample. The communication module uses wireless transmission technology to send the processed data to the monitoring center in real time, ensuring the real-time performance and reliability of data transmission.

[0009] Furthermore, the integrated microprocessor includes an impurity type and concentration discrimination method, which includes a discrimination calculation model. The discrimination calculation model includes a classification module and a concentration fitting module. The classification module is used to determine the type of impurity in the oil sample, and the concentration fitting module is used to calculate the impurity concentration corresponding to the determined impurity type, thereby outputting the impurity type and concentration.

[0010] The discriminant calculation model includes an offline training phase and an online detection phase.

[0011] Offline training phase: Collect a number of offline data, including spectral data obtained by detecting different impurities in transformer oil through a spectral sensor and conductivity data obtained by detecting transformer oil under different impurity concentration conditions through a conductivity sensor. Label the spectral data and conductivity data of each impurity.

[0012] Multiple linear regression and principal component analysis algorithms were used to process the spectral data to calculate the concentration of each different impurity, denoted as impurity concentration one. Gradient descent was used to process the conductivity data to calculate the concentration of each different impurity, denoted as impurity concentration two. Then, impurity concentration one and impurity concentration two for each impurity were fused to obtain the final concentration for each impurity. Several final concentrations for each impurity were used as training set one, and the types of impurities in training set one were used as labels for the training samples of each type.

[0013] Use several conductivity data for each impurity as training set two, and use the actual concentration of each impurity in training set two as the label of the corresponding concentration training sample.

[0014] The classification module is trained using the first training set. When the loss no longer decreases, the trained classification module is obtained. The concentration fitting module is trained using the second training set. When the loss no longer decreases, the trained concentration fitting module is obtained. After the classification module and concentration fitting module are trained, the trained discriminant calculation model is obtained.

[0015] Online detection phase: Data collected by spectral and conductivity sensors is input into the trained discrimination calculation model, which outputs the type and concentration of different particles.

[0016] Furthermore, during the offline training phase, the classification module uses the Support Vector Machine (SVM) algorithm to construct a hyperplane to separate impurity data of different categories and minimize the hinge loss function. The concentration fitting module employs a regression analysis algorithm, optimizing model parameters through least squares or gradient descent to ensure accurate fitting of the relationship between conductivity data and impurity concentration.

[0017] The training samples are input into the classification module. The Support Vector Machine (SVM) algorithm separates the data of different categories by constructing a hyperplane, which is represented as:

[0018] w·x+b=0 (1)

[0019] Where w is the normal vector, b is the bias term, and x is the input data.

[0020] SVM uses the hinge loss function, defined as:

[0021]

[0022] in It is the training sample of the i-th class in training set 1. The predicted value, This represents the category label of the i-th category training sample.

[0023] The goal of training the classification module is to minimize the following loss function:

[0024]

[0025] Where C is the regularization parameter, used to control the complexity of the classification module and the degree of fit to the training set, and n represents the number of class training samples in training set one.

[0026] The concentration fitting module uses a regression analysis algorithm to calculate the concentration of each impurity:

[0027]

[0028] Where Y is the concentration of each impurity. Let βj be the j-th training sample with concentration in training set 2, where β0, β1, ..., βj. m ε is the regression coefficient, and ε is the error term.

[0029] The goal of training the concentration fitting module is to minimize the loss function J(θ):

[0030]

[0031] Where m is the total number of concentration training samples in training set two. Let j be the actual impurity concentration of the j-th concentration training sample. These are the predicted values ​​from the concentration fitting module.

[0032] During training, the gradient descent algorithm is used to optimize the parameters of the classification module and the concentration fitting module, updating the SVM weight vector and the coefficients of the regression analysis algorithm, as shown in the following formula:

[0033]

[0034] Where θ represents the model parameters, which are the weight vector in SVM and the regression coefficients in regression analysis algorithms; α is the learning rate. It is the gradient of the loss function with respect to the parameters.

[0035] Furthermore, during the offline training phase, the method for processing the spectral data using multiple linear regression and principal component analysis algorithms to calculate different impurity concentrations is as follows:

[0036] D concentration2 =X·β+ε1 (7)

[0037] Among them, D concentration1 Let ε be the impurity concentration, X be the spectral data obtained from the spectral sensor, β be the regression coefficient, and ε1 be the error term.

[0038] Furthermore, during the offline training phase, the gradient descent algorithm is used to process the conductivity data to calculate the concentrations of different impurities. The method is as follows:

[0039] The concentration fitting model uses a linear regression algorithm to fit the relationship between conductivity data and impurity concentration:

[0040] D concentration2 =β0 + β1·x1 + β2·x2 + ... + β n ·x n (8)

[0041] Among them, D concentration2 Let the impurity concentrations be two, x1, x2, ..., x n The conductivity data are obtained from the conductivity sensor, β0,β1,…,βn These are the regression coefficients, which are obtained during the offline training phase using the least squares method or gradient descent algorithm.

[0042] Furthermore, during the offline training phase, the concentration-D obtained from the spectral sensor data of different impurities is calculated. concentration1 Concentration 2D calculated from conductivity sensor data concentration2 The mixture is then fused to obtain the final concentration of each impurity. The fusion formula is as follows:

[0043] D combined =α·D concentration1 +(1-α)·D concentration2 (9)

[0044] Among them, D combined This represents the final concentration of each impurity, and α is the fusion weight.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] 1. This invention, through the combination of a sampling module, a spectral sensor, and a conductivity sensor, enables the detection of impurity characteristics in transformer oil from multiple angles and with multiple indicators. Specifically, the spectral sensor can detect the optical properties of the oil sample and identify different types of impurities; the conductivity sensor can detect the electrical properties of the oil sample and reflect the content of conductive impurities in the oil sample. This multi-dimensional detection method greatly improves the accuracy and comprehensiveness of the detection, providing more detailed oil sample analysis data.

[0047] 2. The communication module of this invention adopts advanced wireless transmission technology, supporting Wi-Fi and LoRa communication methods. Wireless transmission technology ensures that processed data is sent to the monitoring center in real time, enabling the monitoring center to receive and process oil sample test data in real time, ensuring the real-time performance and reliability of data transmission. This real-time data transmission capability allows the monitoring center to promptly grasp the transformer's operating status, effectively preventing and promptly handling potential faults.

[0048] 3. The analysis and processing unit of this invention integrates a high-performance microprocessor (model: MCU-500) and an impurity type and concentration discrimination method. The microprocessor possesses efficient data processing capabilities, enabling it to quickly process large amounts of sensor data. The impurity type and concentration discrimination method can accurately analyze sensor data to determine the type and concentration of impurities in the oil sample, thereby generating a reliable test report. This efficient data processing capability not only improves the detection speed but also ensures the accuracy and reliability of the test results.

[0049] 4. The sampling module of this invention employs a micro pump (MP-100) and an adjustable flow valve (FV-200), enabling stable and continuous oil sample collection. The micro pump has a flow rate range of 0.1-1 L / min, ensuring stable oil sample extraction under different operating conditions. The adjustable flow valve, through a manually adjustable knob, allows for precise flow rate control, ensuring the stability and continuity of the oil sample and providing high-quality oil samples for sensor detection.

[0050] 5. The various modules of this invention are compactly designed and easy to install at the sampling port of the transformer oil tank. Installation and commissioning are simple and convenient, ensuring a secure and leak-free connection. The operating status of each module is easy to check and maintain; users can recalibrate the sensors as needed to maintain detection accuracy. The operating status of the micro-pump and flow valve can be checked and maintained regularly, and replaced promptly to ensure the continuity and stability of oil sample collection.

[0051] 6. This invention can detect not only the optical properties of oil samples but also their electrical properties. Through comprehensive processing by the analysis unit, a detailed test report can be generated, providing comprehensive monitoring and analysis of the transformer's operating status. Unlike traditional single-detection methods, this invention provides a more comprehensive and multifunctional oil sample testing solution, improving the overall level of transformer oil testing.

[0052] 7. Existing detection methods suffer from drawbacks such as the inability to perform real-time online detection, complex and time-consuming detection procedures, high maintenance costs, and complex operation. These shortcomings make it difficult to meet the demand for efficient and accurate monitoring of impurity particles in transformer oil. The detection device of this invention can not only detect impurity particles in oil in real time, but also identify different types of particles. It is easy to operate and has low maintenance costs, making it a promising solution with broad application prospects and market value. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a block diagram of the online transformer oil particle monitoring device of the present invention.

[0055] Figure 2 This describes the principle of particle type and concentration detection.

[0056] Figure 3 This is a schematic diagram showing the installation location of the detection device.

[0057] Among them, 1-transformer oil tank; 2-transformer insulating oil; 3-impurity particles; 4-transformer coil; 5-sampling module. Detailed Implementation

[0058] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0059] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. "If" means one or more, unless otherwise explicitly specified.

[0060] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0061] An online monitoring device for transformer oil particles based on spectral and conductivity detection includes the following components:

[0062] Sampling module: includes a micro pump and an adjustable flow valve. The micro pump is used to achieve continuous oil sample collection, and the adjustable flow valve is used to ensure the stability of the oil sample.

[0063] Sensor module: includes a spectral sensor and a conductivity sensor. The spectral sensor is used to detect the optical properties of the oil sample, and the conductivity sensor is used to monitor the conductivity changes of the oil sample and detect the electrical properties of different impurities in the oil.

[0064] Analysis and processing unit: Integrated microprocessor method, wherein the microprocessor is used to process sensor data and determine the type and concentration of impurities.

[0065] Communication module: It uses wireless transmission technology to send the impurity types and concentrations obtained by the analysis and processing unit to the monitoring center in real time.

[0066] In practice, the sampling module includes a miniature pump of model MP-100 and an adjustable flow valve of model FV-200. The miniature pump has a flow rate range of 0.1-1L / min and can continuously and stably extract oil samples. The combined use of the miniature pump MP-100 and the adjustable flow valve FV-200 ensures the stability and continuous collection of oil samples, providing high-quality samples for subsequent sensor detection.

[0067] In specific implementation, the sensor module includes a SP-300 spectral sensor and an EC-100 conductivity sensor. The spectral sensor has a wavelength range of 200-800 nm. The SP-300 spectral sensor is used to detect the spectral characteristics of the oil sample. By analyzing the spectral response of the oil sample, the SP-300 spectral sensor can identify different types of impurities and provide detailed optical detection data. The EC-100 conductivity sensor is used to detect changes in the conductivity of the oil sample. By monitoring the conductivity of the oil sample, the EC-100 conductivity sensor can reflect the content of conductive impurities in the oil sample and provide supplementary electrical detection data. The conductivity sensor has a measurement range of 0-200 μS / cm and monitors changes in the conductivity of the oil sample.

[0068] In practical implementation, the analysis and processing unit integrates a microprocessor MCU-500 and an impurity type and concentration discrimination method to process sensor data and determine the type and concentration of impurities in the oil sample. The MCU-500's high-efficiency computing power and dedicated algorithms can accurately analyze sensor data, determine the type and concentration of impurities in the oil sample, and provide reliable detection results. The microprocessor has high-performance computing capabilities, enabling it to quickly process large amounts of data and accurately identify impurities in the oil through its built-in algorithms, generating a reliable detection report.

[0069] In specific implementation, the communication module is model WT-400, supporting Wi-Fi and LoRa communication methods to ensure real-time data transmission and reliable reception, enabling the monitoring center to promptly grasp the transformer's operating status. The WT-400 transmits the impurity types and concentrations obtained from the analysis and processing unit to the monitoring center in real time. The WT-400 wireless transmission module supports multiple communication methods, ensuring real-time data transmission and reliable reception, enabling the monitoring center to promptly grasp the transformer's operating status.

[0070] Specifically, the integrated microprocessor includes an impurity type and concentration discrimination method, which includes a discrimination calculation model. The discrimination calculation model includes a classification module and a concentration fitting module. The classification module is used to determine the type of impurity in the oil sample, and the concentration fitting module is used to calculate the impurity concentration corresponding to the determined impurity type, thereby outputting the impurity type and concentration.

[0071] The discrimination calculation model includes an offline training phase and an online detection phase;

[0072] Offline training phase: Collect a number of offline data, including spectral data obtained by detecting different impurities in transformer oil through a spectral sensor and conductivity data obtained by detecting transformer oil under different impurity concentration conditions through a conductivity sensor. Label the spectral data and conductivity data of each impurity.

[0073] Multiple linear regression and principal component analysis algorithms were used to process the spectral data to calculate the concentration of each different impurity, denoted as impurity concentration one. Gradient descent was used to process the conductivity data to calculate the concentration of each different impurity, denoted as impurity concentration two. Then, impurity concentration one and impurity concentration two for each impurity were fused to obtain the final concentration for each impurity. Several final concentrations for each impurity were used as training set one, and the types of impurities in training set one were used as labels for the training samples of each type.

[0074] Use several conductivity data for each impurity as training set two, and use the actual concentration of each impurity in training set two as the label of the corresponding concentration training sample.

[0075] The classification module is trained using the first training set. When the loss no longer decreases, the trained classification module is obtained. The concentration fitting module is trained using the second training set. When the loss no longer decreases, the trained concentration fitting module is obtained. After the classification module and concentration fitting module are trained, the trained discriminant calculation model is obtained.

[0076] Online detection phase: Data collected by spectral and conductivity sensors is input into the trained discrimination calculation model, which outputs the type and concentration of different particles.

[0077] Specifically, during the model training phase, a large amount of offline data needs to be collected first. This training data includes spectral and conductivity data detected by spectral and conductivity sensors, covering different concentrations of common impurities in transformer oil, and corresponding to known impurity types and their concentration labels.

[0078] During the offline training phase, the classification module uses the Support Vector Machine (SVM) algorithm to construct a hyperplane to separate impurity data of different categories and minimize the hinge loss function. The concentration fitting module employs a regression analysis algorithm, optimizing model parameters through least squares or gradient descent to ensure accurate fitting of the relationship between conductivity data and impurity concentration.

[0079] The training samples are input into the classification module. The Support Vector Machine (SVM) algorithm separates the data of different categories by constructing a hyperplane, which is represented as:

[0080] w·x+b=0 (1)

[0081] Where w is the normal vector, b is the bias term, and x is the input data.

[0082] SVM uses the hinge loss function, defined as:

[0083]

[0084] in It is the training sample of the i-th class in training set 1. The predicted value, This represents the category label of the i-th category training sample.

[0085] The goal of training the classification module is to minimize the following loss function:

[0086]

[0087] Here, C is the regularization parameter, used to control the complexity of the classification module and its fit to the training set, and n represents the number of training samples of each class in the training set. In this way, SVM can learn the optimal classification boundary for different types of impurities.

[0088] The concentration fitting module uses a regression analysis algorithm to calculate the concentration of each impurity:

[0089]

[0090] Where Y is the concentration of each impurity. Let βj be the j-th training sample with concentration in training set 2, where β0, β1, ..., βj. m These are the regression coefficients, and ε is the error term. To determine the optimal model parameters β0, β1, ..., β... m The model is trained using the least squares method or gradient descent method.

[0091] The goal of training the concentration fitting module is to minimize the loss function J(θ): J(θ) uses the mean squared error (MSE) as the loss function.

[0092]

[0093] Where m is the total number of concentration training samples in training set two. Let j be the actual impurity concentration of the j-th concentration training sample. These are the predicted values ​​from the concentration fitting module.

[0094] After multiple iterations, the model finally determined the optimal parameters β0,β1,…,β mAt this point, the model can accurately fit the relationship between conductivity and impurity concentration. The final regression analysis model can be applied to online monitoring devices, predicting the concentration of each impurity in transformer oil by inputting real-time conductivity data. This model accurately reflects the relationship between conductivity and impurity concentration, ensuring the accuracy and reliability of online detection of particulate matter in transformer oil.

[0095] During training, the gradient descent algorithm is used to optimize the parameters of the classification module and the concentration fitting module, updating the SVM weight vector and the coefficients of the regression analysis algorithm, as shown in the following formula:

[0096]

[0097] Where θ represents the model parameters, which are the weight vector in SVM and the regression coefficients in regression analysis algorithms; α is the learning rate. It is the gradient of the loss function with respect to the parameters.

[0098] In the structure of the discriminant calculation model, the classification model and the concentration fitting model work closely together. The classification model determines the type of impurities in the oil sample based on the input sensor data and outputs specific impurity category labels, which serve as the basis for subsequent concentration calculations. After receiving the impurity type label, the concentration fitting model uses spectral and conductivity data to accurately calculate the concentration of the corresponding impurity. The organic combination of these two models constitutes the core function of the discriminant calculation model, enabling it to comprehensively output the type and concentration of impurities.

[0099] After training, the entire discriminant calculation model is validated and optimized to ensure its accuracy and robustness, and finally, the model is written into the microprocessor. In practical applications, the microprocessor relies on the discriminant calculation model to achieve real-time monitoring and analysis of impurities in transformer oil, accurately determine the types of impurities in the oil, and precisely calculate their concentrations, thereby providing reliable data support for transformer condition monitoring and maintenance.

[0100] Specifically, during the offline training phase, the method for processing spectral data using multiple linear regression and principal component analysis algorithms to calculate different impurity concentrations is as follows:

[0101] D concentration2 =X·β+ε1 (7)

[0102] Among them, D concentration1 Let ε be the impurity concentration, X be the spectral data obtained from the spectral sensor, β be the regression coefficient, and ε1 be the error term.

[0103] Specifically, during the offline training phase, the gradient descent algorithm is used to process the conductivity data to calculate the concentrations of different impurities. The method is as follows:

[0104] The concentration fitting model uses a linear regression algorithm to fit the relationship between conductivity data and impurity concentration:

[0105] D concentration2 =β0 + β1·x1 + β2·x2 + ... + β n ·x n (8)

[0106] Among them, D concentration2 Let the impurity concentrations be two, x1, x2, ..., x n The conductivity data are obtained from the conductivity sensor, β0,β1,…,β n These are the regression coefficients, which are obtained during the offline training phase using the least squares method or gradient descent algorithm.

[0107] Specifically, during the offline training phase, the concentration-D obtained from the spectral sensor data of different impurities is calculated. concentration1 Concentration 2D calculated from conductivity sensor data concentration2 The mixture is then fused to obtain the final concentration of each impurity. The fusion formula is as follows:

[0108] D combined =α·D concentration1 +(1-α)·D concentration2 (9)

[0109] Among them, D combined This represents the final concentration of each impurity, and α is the fusion weight, the specific value of which can be adjusted according to the accuracy and characteristics of different sensors.

[0110] This invention also provides a method for using an online monitoring device for transformer oil particles based on spectral and conductivity detection, comprising:

[0111] Step 1: Install and connect components

[0112] First, the sampling module (a miniature pump of model MP-100 and an adjustable flow valve of model FV-200) was installed at the sampling port of the transformer oil tank, ensuring a secure and leak-free connection. Then, the sensor modules (a spectral sensor of model SP-300 and a conductivity sensor of model EC-100) were connected to the output of the sampling module, along with the analysis and processing unit (microprocessor MCU-500) and the communication module (model WT-400), ensuring a complete system connection. Preferably, all connecting parts are made of oil-resistant materials to prevent leakage and contamination.

[0113] Step 2: Start the equipment and collect oil samples

[0114] The micro pump (MP-100) is started, preferably with the flow rate set at 0.5 L / min. The flow rate is precisely controlled by an adjustable flow valve (FV-200) to ensure the stability and continuity of the oil sample. The oil sample is pumped from the sampling module to the sensor module. Preferably, a filter screen is used to filter the oil sample to prevent particles from clogging the pipeline.

[0115] Step 3: Sensor Module Testing

[0116] Before initiating detection, initialize the spectral sensor (model: SP-300) and conductivity sensor (model: EC-100) to ensure they are operating at their optimal condition. Preferably, calibrate the sensors using a known standard oil sample and check the accuracy of the sensor response.

[0117] After system startup, the sensor module continuously monitors the oil sample, and the analysis and processing unit (model: MCU-500) processes the data in real time to generate an accurate detection report. Preferably, in the offline stage, the spectral sensor performs optical detection on the oil sample within the wavelength range of 200-800nm, and uses multiple linear regression and principal component analysis algorithms to process the spectral data to calculate the concentration of each different impurity, denoted as impurity concentration one. The conductivity sensor measures the conductivity change of the oil sample, ranging from 0-200μS / cm, and uses a gradient descent algorithm to process the conductivity data to calculate the concentration of each different impurity, denoted as impurity concentration two. The impurity concentration one and impurity concentration two corresponding to each impurity are fused to obtain the final concentration of each impurity; several final concentrations of each impurity are used as training set one, and the types of impurities in training set one are used as labels for the training samples. The trained discrimination calculation model is written into the microprocessor MCU-500.

[0118] Furthermore, the spectral sensor (model: SP-300) performs optical detection on the oil sample within a wavelength range of 200-800 nm to identify spectral characteristics in the oil sample. Preferably, the spectral sensor is initially set up using a calibration standard. A conductivity sensor (model: EC-100) is used to monitor changes in the conductivity of the oil sample, with a measurement range of 0-200 μS / cm, reflecting the content of conductive impurities in the oil sample and providing supplementary electrical detection data. The conductivity sensor should be calibrated periodically to maintain accuracy.

[0119] Step 4: Data Processing and Analysis

[0120] In the online phase, the data collected by the spectral sensor (model: SP-300) and the conductivity sensor (model: EC-100) are transmitted in real time to the analysis and processing unit (model: MCU-500). The unit then uses the discrimination calculation model trained in the offline phase to process the data and output the determined impurity type and concentration in the oil sample.

[0121] Furthermore, the analysis and processing unit (model: MCU-500) integrates all analysis results, generates a test report, and stores the results in the built-in storage unit. Preferably, redundant storage technology is used to ensure data security and reliability.

[0122] Step 5: Data Transfer

[0123] The communication module (model: WT-400) transmits processed data to the monitoring center in real time via wireless transmission technology (preferably Wi-Fi or LoRa communication). The monitoring center views the real-time monitoring results through the data monitoring platform to understand the transformer's operating status. Preferably, the communication module should be configured with multiple encryption algorithms to ensure the security of data transmission.

[0124] Step 6: System Maintenance and Calibration

[0125] The system periodically performs a self-test procedure to check the operating status of each component and ensure fault-free operation. Preferably, the system is equipped with an automatic alarm function, which automatically notifies maintenance personnel to inspect and handle any abnormalities detected. The automatic self-test includes detecting the connection status of components, power supply status, and sensor operating status.

[0126] Preferably, the following operations are performed periodically to ensure the long-term stable operation of the system:

[0127] The spectral sensor (model: SP-300) and conductivity sensor (model: EC-100) are calibrated to ensure detection accuracy. Preferably, a standard oil sample with known properties is used for calibration.

[0128] Check the working condition of the micro pump (MP-100) and the adjustable flow valve (FV-200), and perform maintenance and replacement as necessary to ensure the continuity and stability of oil sample collection.

[0129] Update the microprocessor's algorithms to ensure the accuracy and efficiency of data processing. Preferably, update the software regularly to optimize algorithm performance.

Claims

1. An online monitoring device for transformer oil particles based on spectral and conductivity detection, characterized in that, Includes the following components: Sampling module: includes a micro pump and an adjustable flow valve. The micro pump is used to achieve continuous oil sample collection, and the adjustable flow valve is used to ensure the stability of the oil sample. Sensor module: includes a spectral sensor and a conductivity sensor. The spectral sensor is used to detect the optical properties of the oil sample, and the conductivity sensor is used to monitor the conductivity changes of the oil sample and detect the electrical properties of different impurities in the oil. Analysis and processing unit: Integrated microprocessor method, wherein the microprocessor is used to process sensor data and determine the type and concentration of impurities; Communication module: Uses wireless transmission technology to send the impurity types and concentrations obtained from the analysis and processing unit to the monitoring center in real time; The integrated microprocessor includes an impurity type and concentration discrimination method, which includes a discrimination calculation model. The discrimination calculation model includes a classification module and a concentration fitting module. The classification module is used to determine the type of impurity in the oil sample, and the concentration fitting module is used to calculate the impurity concentration corresponding to the impurity under the determined impurity type. The discrimination calculation model includes an offline training phase and an online detection phase; Offline training phase: Collect a number of offline data, including spectral data obtained by detecting different impurities in transformer oil through a spectral sensor and conductivity data obtained by detecting transformer oil under different impurity concentration conditions through a conductivity sensor. Label the spectral data and conductivity data of each impurity. Multiple linear regression and principal component analysis algorithms were used to process the spectral data to calculate the concentration of each different impurity, denoted as impurity concentration one. Gradient descent algorithm was used to process the conductivity data to calculate the concentration of each different impurity, denoted as impurity concentration two. Subsequently, impurity concentration one and impurity concentration two for each impurity were fused to obtain the final concentration of each impurity. Several final concentrations of each impurity were used as training set one, and the types of impurities in training set one were used as labels for the type training samples. Use several conductivity data for each impurity as training set two, and use the actual concentration of each impurity in training set two as the label of the corresponding concentration training sample. The classification module is trained using the first training set. When the loss no longer decreases, the well-trained classification module is obtained. The concentration fitting module is trained using the second training set. When the loss no longer decreases, the well-trained concentration fitting module is obtained. After the classification module and concentration fitting module are trained, the well-trained discriminant calculation model is obtained. Online detection phase: Data collected by spectral and conductivity sensors is input into the trained discrimination calculation model, which outputs the type and concentration of different particles.

2. The online particle monitoring device in transformer oil according to claim 1, characterized in that: During the offline training phase, the classification module uses the Support Vector Machine (SVM) algorithm to construct a hyperplane to separate impurity data of different categories and minimize the hinge loss function; the concentration fitting module uses a regression analysis algorithm to optimize model parameters through least squares or gradient descent to ensure accurate fitting of the relationship between conductivity data and impurity concentration. The training samples are input into the classification module. The Support Vector Machine (SVM) algorithm separates the data of different categories by constructing a hyperplane, which is represented as: w·x+b=0 (1) Where w is the normal vector, b is the bias term, and x is the input data; SVM uses the hinge loss function, defined as: in It is the training sample of the i-th class in training set 1. The predicted value, This represents the category label of the i-th training sample; The goal of training the classification module is to minimize the following loss function: Where C is the regularization parameter, used to control the complexity of the classification module and the degree of fit to the training set, and n represents the number of training samples of each class in the training set. The concentration fitting module uses a regression analysis algorithm to calculate the concentration of each impurity: Where Y is the concentration of each impurity. Let βj be the j-th training sample with concentration in training set 2, where β0, β1, ..., βj. m These are the regression coefficients, and ε is the error term; The goal of training the concentration fitting module is to minimize the loss function J(θ): Where m is the total number of concentration training samples in training set two. Let j be the actual impurity concentration of the j-th concentration training sample. These are the predicted values ​​from the concentration fitting module; During training, the gradient descent algorithm is used to optimize the parameters of the classification module and the concentration fitting module, updating the SVM weight vector and the coefficients of the regression analysis algorithm, as shown in the following formula: Where θ represents the model parameters, which are the weight vector in SVM and the regression coefficients in regression analysis algorithms; α is the learning rate. It is the gradient of the loss function with respect to the parameters.

3. The online particle monitoring device in transformer oil according to claim 2, characterized in that: During the offline training phase, the method for processing spectral data using multiple linear regression and principal component analysis algorithms to calculate different impurity concentrations is as follows: D concentration2 =X·β+ε1 (7) Among them, D concentration1 Let ε be the impurity concentration, X be the spectral data obtained from the spectral sensor, β be the regression coefficient, and ε1 be the error term.

4. The online particle monitoring device in transformer oil according to claim 3, characterized in that: During the offline training phase, the gradient descent algorithm is used to process the conductivity data to calculate the concentrations of different impurities. The method is as follows: The concentration fitting model uses a linear regression algorithm to fit the relationship between conductivity data and impurity concentration: D concentration2 =β0+β1·x1+β2·x2+…+β n ·x n (8) Among them, D concentration2 Let the impurity concentrations be two, x1, x2, ..., x n The conductivity data are obtained from the conductivity sensor, β0,β1,…,β n is the regression coefficient.

5. The online particle monitoring device in transformer oil according to claim 4, characterized in that: During the offline training phase, the concentration-D obtained from the spectral sensor data of different impurities is calculated. concentration1 Concentration 2D calculated from conductivity sensor data concentration2 The mixture is then fused to obtain the final concentration of each impurity. The fusion formula is as follows: D combined =α·D concentration1 +(1-a)·D concentration2 (9) Among them, D combined This represents the final concentration of each impurity, and α is the fusion weight.

Images