Transformer overheat state recognition method based on electronic nose

Abstract

The application discloses a transformer overheat state recognition method based on an electronic nose, which comprises the following steps: first, building an electronic nose system suitable for detecting pyrolysis gas demand; second, obtaining smell information of transformer oil pyrolysis gas and dividing sample data sets; third, obtaining transformer pyrolysis gas sample data H0 and H1 according to the power transformer operation regulation and experiments; fourth, processing the gas sample H1 in the overheat state; beneficial effects: the overheat temperature can be recognized, the precision is high, the recognition speed is fast, and the method is suitable for real-time monitoring of transformers in transformer rooms and box-type transformer stations; the risk of transformer fire and explosion can be effectively reduced; compared with the prior art, the electronic nose system is low in cost, convenient to deploy, easy to operate, strong in universality, and convenient for monitoring and recognizing the overheat state of the transformer.

Description

Technical Field

[0001] This invention relates to a method for identifying transformer overheating conditions, and more particularly to a method for identifying transformer overheating conditions based on an electronic nose. Background Technology

[0002] Currently, oil-immersed power transformers are a key component of power distribution systems, responsible for converting voltage levels of different grades. Their stable operation is a crucial guarantee for economic development and social progress. A transformer explosion or fire can lead to widespread power outages, seriously endangering lives and property. Prolonged operation of transformers in an overheated state is a significant factor contributing to explosions and fires; therefore, monitoring transformer overheating is essential to reducing the frequency of such accidents.

[0003] Existing methods for identifying transformer overheating conditions mainly fall into three categories: installing fiber optic temperature sensors, establishing data-driven hot spot temperature (HST) prediction models, and transformer condition detection methods based on dissolved gas analysis in oil. All three methods have several drawbacks: fiber optic temperature sensors are expensive and cannot be installed on already operational equipment; data-driven hot spot temperature prediction models are difficult to apply to equipment with significant structural differences, resulting in poor versatility; and transformer condition detection based on dissolved gas analysis in oil requires expensive gas chromatography equipment and highly trained professionals. Therefore, it is necessary to develop a transformer overheating condition identification method that overcomes these shortcomings, is cost-effective, and has strong versatility.

[0004] An electronic nose, a device that mimics the olfactory sensory mechanism of mammals, is designed to detect odors like an animal's nose. The sensor array in an electronic nose system is equivalent to olfactory receptor cells in a biological system, capable of recognizing the "fingerprint" information of different odors and performing appropriate data processing. With its advantages of being economical, portable, and fast in recognition, the electronic nose has been applied in fields such as food, agriculture, and aerospace. By combining it with efficient machine learning methods, it can achieve odor recognition of specific substances.

[0005] Existing methods for detecting and identifying overheated conditions suffer from high cost, poor versatility, and incomplete detection. Previous research has shown that transformer oil releases pyrolysis gases when it is in an overheated state. Therefore, developing a method for identifying overheated conditions based on detecting these pyrolysis gases in transformer oil is a major technical challenge. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of high cost, poor versatility and incomplete detection of existing overheating conditions detection and identification methods, and to provide a transformer overheating condition identification method based on an electronic nose.

[0007] The transformer overheating state identification method based on an electronic nose provided by this invention includes the following steps:

[0008] The first step is to build an electronic nose system suitable for detecting pyrolysis gases: The main structure of the electronic nose includes a gas chamber, a signal conditioning circuit, a data acquisition card, and a gas sensor. The gas flow rate of the electronic nose gas chamber is 800-1200 ml / min.

[0009] The second step involves obtaining odor information from transformer oil pyrolysis gases and dividing the sample dataset. The specific steps are as follows:

[0010] Step 1: Place 80ml of transformer oil sample in a covered alumina crucible, and then use a heating platform to heat the crucible at a constant temperature from room temperature to 95℃, 120℃, 160℃, 200℃, 250℃, 300℃, 350℃ and 400℃ for 40min. After heating, open the crucible lid, collect the pyrolysis gas using a 3L gas collection bag, and use an electronic nose to obtain the odor information of the pyrolysis gas and convert it into a data format that is easy to process and store it.

[0011] Step 2: The single acquisition time of the electronic nose is set to 60s, and the sampling frequency is 100Hz. Each acquisition is considered to generate one data point. In addition, the heating experiment is carried out in a fume hood. A new sample is used for each heating experiment. After each test, the electronic nose system is cleaned with clean air for 10 minutes to ensure that the sensor signal is at baseline for the next acquisition.

[0012] Step 3: According to the document DL / T 572-2021 Power Transformer Operation Procedure, the maximum oil temperature of an oil-immersed transformer should not exceed 95℃. Therefore, the obtained gas sample data is divided into different datasets: gas data from room temperature to 95℃ is denoted as H0, representing data under normal temperature; gas data from 120℃ to 400℃ is denoted as H1, representing data under superheated temperature. Within the H1 dataset, several smaller datasets are divided according to different superheated temperatures: 120℃, 160℃, 200℃, 250℃, 300℃, 350℃, and 400℃ are denoted as t1, t2, t3, t4, t5, t6, and t7, respectively, representing data at different superheated temperatures.

[0013] The third step, based on the power transformer operation regulations and the experimentally obtained transformer pyrolysis gas sample data H0 and H1, is to first establish a transformer overheating state identification model to distinguish whether the transformer is in an overheating state. The specific steps are as follows:

[0014] Step 1: Extract features from the two gas sample datasets H0 and H1, which represent "normal" and "overheated" respectively. The feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN).

[0015] Step 2: Using the feature values ​​of the gas sample data and the label category corresponding to each gas sample, train the PSO-SVM (Particle Swarm Optimized Support Vector Machine) for normal or overheating conditions to obtain the overheating state identification model of the transformer, which is used to distinguish whether the transformer is in an overheating state.

[0016] Step 4: For the gas sample H1 in a superheated state, a temperature identification and prediction model is established using the random forest method. The specific steps are as follows:

[0017] Step 1: Extract features from the gas sample data t1, t2, t3, t4, t5, t6, and t7. The feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN).

[0018] Step 2: Use the feature values ​​of the gas sample data and the actual experimental temperature value corresponding to each gas sample to train the random forest model and establish an overheating temperature identification model for transformers to predict the overheating temperature of transformers.

[0019] The beneficial effects of this invention are:

[0020] The transformer overheating identification method based on an electronic nose provided by this invention can not only accurately identify whether a transformer is overheating through gas detection, but also identify the overheating temperature. It boasts high accuracy and fast identification speed, making it suitable for real-time transformer monitoring in transformer rooms and prefabricated substations. The electronic nose system provided by this invention offers a fast response time; when the electronic nose system issues a transformer overheating alarm signal, maintenance personnel can respond quickly, effectively reducing the risk of transformer fires and explosions. Compared with existing technologies, the electronic nose system provided by this invention is low-cost, easy to deploy, easy to operate, and highly versatile, making the monitoring and identification of transformer overheating conditions extremely convenient. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall framework for identifying the overheating temperature of the transformer using the electronic nose as described in this invention.

[0022] Figure 2This is a schematic diagram of the overheating state identification model described in this invention.

[0023] Figure 3 This is a schematic diagram of the overheating temperature identification model process described in this invention.

[0024] Figure 4 This is a schematic diagram of the overheating state prediction confusion matrix of the training data (MAX features) described in this invention.

[0025] Figure 5 This is a schematic diagram of the overheating state prediction confusion matrix for the test data (MAX features) described in this invention.

[0026] Figure 6 This is a schematic diagram of the confusion matrix for overheating temperature prediction based on the training data (MEAN features) described in this invention.

[0027] Figure 7 This is a schematic diagram of the overheating temperature prediction confusion matrix for the test data (MEAN features) described in this invention. Detailed Implementation

[0028] Please see Figures 1 to 7 As shown:

[0029] The transformer overheating state identification method based on an electronic nose provided by this invention includes the following steps:

[0030] The first step is to build an electronic nose system suitable for detecting pyrolysis gases. The main structure of the electronic nose includes a gas chamber, a signal conditioning circuit, a data acquisition card, and a gas sensor array. The sensor array uses 32 sensors of the following models: TGS2612, TGS2611, TGS2620, TGS2603, TGS2602, TGS2610, TGS2600, GSBT11, MS1100, MP135, MP901, MP-9, MP-3B, MP-4, MP-5, MP-2, MP503, MP801, MP905, MP402, WSP1110, WSP2110, WSP7110, MP-7, TGS2612, TGS2611, TGS2620, MP-3B, MP702, TGS2610, TGS2600, and TGS2618-COO.

[0031] Step 2: Obtaining data on transformer oil pyrolysis gases generated during transformer overheating: 80ml of a sample of 25# transformer oil produced by Dongguan Luosheng Lubricating Oil Co., Ltd. was placed in a covered alumina crucible. The heating platform was heated to a constant temperature (room temperature to 95℃, 120℃, 160℃, 200℃, 250℃, 300℃, 350℃, and 400℃) for 40 minutes. The crucible and heating platform were then covered with a square glass cover with dimensions of 50cm x ...

[0032] Third, the electronic nose's single-sampling time was set to 60 seconds, and the sampling frequency was 100 Hz. Each acquisition was considered as obtaining one data point. After each test, the electronic nose system was cleaned with clean air for 10 minutes to ensure the sensor signal was at baseline for the next acquisition. Table 1 shows the number of data acquired at different experimental temperatures.

[0033] Table 1. Number of data points obtained at different temperatures

[0034]

[0035] The fourth step involves dividing the acquired gas sample data into different datasets: gas data from room temperature to 95°C is denoted as H0, representing data at normal temperatures; gas data from 120°C to 400°C is denoted as H1, representing data at superheated temperatures. Within the H1 dataset, multiple smaller datasets are further subdivided based on different superheated temperatures: 120°C, 160°C, 200°C, 250°C, 300°C, 350°C, and 400°C are denoted as t1, t2, t3, t4, t5, t6, and t7, respectively, representing data at different superheated temperatures.

[0036] Step 5: Extract features from the two gas sample datasets H0 and H1, representing "normal" and "overheated" respectively. Feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN).

[0037] Step 6: Build an overheating state identification model using the feature values ​​of the gas sample data and the corresponding label category (normal or overheated) for each gas sample. 70% of the data is used for model training, and 30% is used for model testing. This invention uses the Particle Swarm Optimization (PSO) algorithm to find the optimal hyperparameters (penalty parameter c and kernel function parameter gamma) in the Support Vector Machine (SVM). The found optimal hyperparameters are then fed into the SVM to finally obtain the Particle Swarm Optimization Support Vector Machine model (PSO-SVM). The Particle Swarm Optimization algorithm includes the following steps:

[0038] Step 1: Set the population size to 5, that is, initialize the population X = {X1, X2, X3, X4, X5};

[0039] Step 2: Set the maximum number of iterations to 10, the inertia weight W to 0.8, the acceleration constants c1 and c2 to 1.5, and select the error rate of the training prediction results as the fitness function;

[0040] Step 3: Calculate the fitness value for each particle;

[0041] Step 4: Calculate the individual historical best position and global historical best position of each particle based on the fitness value, and update the velocity and position of each particle.

[0042] speed:

[0043] Location:

[0044] W represents the inertial weight, c1 and c2 are the acceleration constants, rand is a random number between [0,1], and V... id t and X id t Represents particle X i At time t, Pbest's velocity and position in d-dimensional space are given. id t and Gbest id t Represents particle X i The historical best position of an individual in dimension d at time t is the global historical best position of the population.

[0045] Step 5: If the maximum number of iterations reaches 10, output the global historical best position vector. The vector parameters are the optimized hyperparameter values, namely the penalty parameter c and kernel function parameter gamma in SVM. If the number of iterations is less than 10, return to step 6.3) to continue iterating.

[0046] Step 7: The trained PSO-SVM model is the transformer overheating state identification model. Table 2 shows the identification results corresponding to each feature extraction method:

[0047] Table 2 Results of the Overheating State Identification Model

[0048]

[0049] The overheating state identification model achieved an accuracy rate of 85.48% to 99.19%, with the highest accuracy rate of 99.19% observed under the MAX feature. Furthermore, this overheating state identification model had a false alarm rate of 0 and a false negative rate of less than 1%.

[0050] Step 8: Extract features from the gas sample data t1, t2, t3, t4, t5, t6, and t7 representing the superheated state. Feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN).

[0051] Step 9: Using the feature values ​​of the gas sample data from Step 8 and the actual experimental temperature values ​​corresponding to each gas sample, an overheating temperature identification model is established. 70% of the data is used for training the random forest model (containing 50 decision trees), and 30% of the data is used for testing to predict the transformer's overheating temperature. Table 3 shows the identification results corresponding to each feature extraction method:

[0052] Table 3 Results of the overheat temperature identification model

[0053]

[0054] The overheating temperature identification model achieved an accuracy rate of 86.79% to 99.05%, with the highest prediction accuracy of 99.06% observed under the MEAN feature. Clearly, the trained random forest model possesses excellent temperature identification capabilities.

Claims

1. A method for identifying transformer overheating status based on an electronic nose, characterized in that: The method includes the following steps: The first step is to build an electronic nose system suitable for detecting pyrolysis gases: The main structure of the electronic nose includes a gas chamber, a signal conditioning circuit, a data acquisition card, and a gas sensor. The gas flow rate of the electronic nose gas chamber is 800-1200 ml / min. The second step involves obtaining odor information from transformer oil pyrolysis gases and dividing the sample dataset. The specific steps are as follows: Step 1: Place 80ml of transformer oil sample in a covered alumina crucible, and then use a heating platform to heat the crucible at a constant temperature from room temperature to 95℃, 120℃, 160℃, 200℃, 250℃, 300℃, 350℃ and 400℃ for 40min. After heating, open the crucible lid, collect the pyrolysis gas using a 3L gas collection bag, and use an electronic nose to obtain the odor information of the pyrolysis gas and convert it into a data format that is easy to process and store it. Step 2: The single acquisition time of the electronic nose is set to 60s, and the sampling frequency is 100Hz. Each acquisition is considered to generate one data point. In addition, the heating experiment is carried out in a fume hood. A new sample is used for each heating experiment. After each test, the electronic nose system is cleaned with clean air for 10 minutes to ensure that the sensor signal is at baseline for the next acquisition. Step 3: According to the document DL / T 572-2021 Power Transformer Operation Procedure, the maximum oil temperature of an oil-immersed transformer should not exceed 95℃. Therefore, the obtained gas sample data is divided into different datasets: gas data from room temperature to 95℃ is denoted as H0, representing data under normal temperature; gas data from 120℃ to 400℃ is denoted as H1, representing data under superheated temperature. Within the H1 dataset, several smaller datasets are divided according to different superheated temperatures: 120℃, 160℃, 200℃, 250℃, 300℃, 350℃, and 400℃ are denoted as t1, t2, t3, t4, t5, t6, and t7, respectively, representing data at different superheated temperatures. The third step, based on the power transformer operation regulations and the experimentally obtained transformer pyrolysis gas sample data H0 and H1, is to first establish a transformer overheating state identification model to distinguish whether the transformer is in an overheating state. The specific steps are as follows: Step 1: Extract features from the two gas sample datasets H0 and H1, which represent "normal" and "overheated" respectively. The feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN). Step 2: Using the feature values ​​of the gas sample data and the label category corresponding to each gas sample, train the PSO-SVM (Particle Swarm Optimized Support Vector Machine) for normal or overheating conditions to obtain the overheating state identification model of the transformer, which is used to distinguish whether the transformer is in an overheating state. Step 4: For the gas sample H1 in a superheated state, a temperature identification and prediction model is established using the random forest method. The specific steps are as follows: Step 1: Extract features from the gas sample data t1, t2, t3, t4, t5, t6, and t7. The feature extraction methods include transient feature extraction methods such as Fourier transform (FT), exponential curve fitting (ECF), Gaussian curve fitting (GCF), polynomial curve fitting (PCF), and wavelet transform (WT), as well as steady-state feature extraction methods such as integral value (IV), maximum value (MAX), and average value (MEAN). Step 2: Use the feature values ​​of the gas sample data and the actual experimental temperature value corresponding to each gas sample to train the random forest model and establish an overheating temperature identification model for transformers to predict the overheating temperature of transformers.

Images