A method of testing the water content of a casing

Abstract

This invention relates to a method for testing the water content of casing. First, a dielectric response curve test is performed on casings with different water contents within a frequency range of 1000Hz–0.01Hz using a Field-Sensitive Spectrum (FDS) method, obtaining FDS-based dielectric spectrum curves for different water contents. Then, feature parameters are extracted from the FDS-based dielectric spectrum curves. Next, using the extracted feature parameters and frequency as input layers and the water content as the output layer, a backpropagation neural network is used to fit the target model function, denoted as y. (FDS) Finally, the particle swarm optimization algorithm is used to optimize the weighting factors for each frequency band, resulting in the most accurate target model for precise testing of the moisture content of the transformer's solid insulation. This invention is not easily affected by leakage current, temperature, etc., and can accurately assess the bushing insulation condition; it can obtain specific moisture content values, and can quantitatively analyze and track changes in bushing moisture content over months or years; it can accurately measure the moisture content in the insulating paper of oil-paper capacitor bushings.

Description

Technical Field

[0001] This invention belongs to the field of power technology, and in particular relates to a method for testing the water content of bushings. Background Technology

[0002] Bushings are crucial insulation devices in power systems, and their insulation condition directly affects the safe and stable operation of the power system. Bushings generally consist of three parts: a conductor, an insulation layer, and a flange. Based on the different insulation materials, bushings can be classified into pure porcelain bushings, resin bushings, oil-filled bushings, gas-filled bushings, oil-paper capacitor bushings, etc. Currently, the most commonly used bushings in the industry are oil-filled bushings (110kV and below) and oil-paper capacitor bushings (above 110kV). The moisture content of the bushing insulation material (insulating paper, oil) reflects the insulation condition of the bushing and is of great significance for predicting the remaining life of the bushing and diagnosing faults.

[0003] Currently, the assessment of bushing insulation condition relies solely on conventional preventative tests. However, these tests, such as insulation resistance testing, dissolved gas chromatography analysis, dielectric loss factor testing, and capacitance testing, are limited by their susceptibility to surface leakage current, temperature, and air humidity. This makes accurate assessment of bushing insulation condition difficult, especially for oil-paper capacitor bushings, where moisture is mostly concentrated within the insulating paper, with very little water content in the oil. Therefore, conventional testing methods cannot accurately evaluate the bushing's insulation condition. Consequently, a new method for testing bushing moisture content that overcomes the limitations of conventional preventative tests and provides more accurate results is urgently needed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for testing the water content of casing. Based on the frequency domain spectrum method (FDS), the FDS dielectric spectrum curves of casings with different water contents are first obtained. Then, a characteristic parameter function model of the dielectric spectrum curve is established. The target model function is fitted and optimized using the BP neural network algorithm and the particle swarm algorithm to obtain the most accurate target model, so as to achieve the purpose of testing the water content of casing most accurately. At the same time, it can overcome the limitations of the existing test methods.

[0005] This invention provides a method for testing the water content of casing, comprising the following steps:

[0006] Step 1: Based on the FDS method, conduct dielectric spectrum response tests on the casing with different water contents in the frequency range of 1000Hz-0.01Hz to obtain the dielectric spectrum curves of the casing under different water contents;

[0007] Step 2: Extract characteristic parameters from the spectral response curves of each medium in the frequency range of 1000Hz-0.01Hz based on the FDS method. This includes dividing the 1000Hz-0.01Hz range into 5 frequency bands: band ① 1000Hz~100Hz, band ② 100Hz~10Hz, band ③ 10Hz~1Hz, band ④ 1Hz~0.1mHz, and band ⑤ 0.1Hz~0.01Hz. The characteristic parameter is the conductivity amplitude A within each of the 5 frequency bands. n Conductivity slope a n Polarization amplitude X n Polarization time constant τ n n = 1 to 5, that is, within the frequency band ① 1000Hz to 100Hz, they are respectively denoted as: conductivity amplitude A1, conductivity slope a1, polarization amplitude X1, and polarization time constant τ1. Similarly, the characteristic parameters of the other four frequency bands ②, ③, ④, and ⑤ are deduced in turn.

[0008] Step 3: Using the feature parameters extracted from the spectral curves of each medium based on the FDS method as the input layer and the water content corresponding to each spectral curve as the output layer, the target model function is trained using a BP neural network model.

[0009] Step 4, for the target model under different water contents Each frequency band is assigned a weighting factor Q. n The corresponding frequencies are: ① 1000Hz~100Hz for Q1, ② 100Hz~10Hz for Q2, ③ 10Hz~1Hz for Q3, ④ 1Hz~0.1mHz for Q4, and ⑤ 0.1Hz~0.01Hz for Q5. After adding weighting factors, it becomes Using the particle swarm optimization algorithm to adjust the weight factor Q n Optimization is performed to obtain the most accurate target model;

[0010] Step 5: Obtain an accurate test model for casing water content based on FDS based on the optimized target model. The water content obtained based on this model is the specific water content of the tested casing.

[0011] Furthermore, the BP neural network model described in step 3 consists of two modules: a forward propagation network for information and a backward propagation network for error. Various information from the outside is transmitted through the input layer of the BP neural network into its hidden layer for network operation and processing, and the final processing result is obtained through the output layer. When the error between the output result of the BP neural network and its preset input value is large, the backward propagation stage of the BP neural network is entered, and the network weights are updated until the error between the output result and the expected result meets the set conditions.

[0012] The forward propagation process of the signal is as follows:

[0013] 1) The input variable net of the i-th node in the hidden layer of the neural network i :

[0014]

[0015] 2) The output variable y of the i-th node in the hidden layer of the neural network i :

[0016]

[0017] 3) The input variable net of the k-th node in the output layer of the neural network k :

[0018]

[0019] 4) The output variable O of the kth node in the output layer of the neural network k :

[0020]

[0021] Wherein, variable x j Let x represent the input parameters of the j-th node in the input layer of the BP neural network. j The conductivity amplitude A is respectively n Conductivity slope a n Polarization amplitude X n Polarization time constant τ n n = 1 to 5; w ij The variable θ represents the neural network weight parameters between the i-th node of the hidden layer and the j-th node of the input layer in a BP neural network. i The threshold parameter represents the i-th node in the hidden layer of a BP neural network; variable This represents the activation function of the hidden layer in a BP neural network; variable w ki The weights between the k-th node of the output layer and the i-th node of the hidden layer in a BP neural network are represented by the variable α, where i = 1 to q.k The threshold parameter of the k-th node in the output layer of the BP neural network is represented by Ψ(x), where k = 1 to L; the activation function of the output layer of the BP neural network is represented by Ψ(x); the variable o k This represents the output of the k-th node in the output layer of a BP neural network, with the output variable o. k This refers to the water content.

[0022] By employing the above-described scheme and the casing moisture content testing method, the following technical effects are achieved:

[0023] 1) The testing method proposed in this invention is not easily affected by leakage current, temperature, etc., and can accurately assess the insulation status of bushings. It overcomes the limitations of conventional preventive tests, such as insulation resistance testing, dissolved gas chromatography analysis in oil, dielectric loss factor, and capacitance testing, which are easily affected by surface leakage current, temperature, and humidity, and are prone to deviation.

[0024] 2) The testing method proposed in this invention can obtain specific moisture content values ​​for specific bushings, and can quantitatively analyze and track the changes in moisture content of bushings over several months or years. It solves the problems of conventional testing methods being limited by their specific methodological principles, greatly affected by on-site factors, and the possibility of large deviations between adjacent test results, and the inability to quantitatively analyze and track the degree of insulation aging of bushings over several years or months.

[0025] 3) The bushing moisture content testing method based on frequency domain spectrum method (FDS) proposed in this invention can accurately measure the moisture content in the insulating paper of oil-paper capacitor bushings. This solves the problem that current testing methods for oil-paper capacitor bushings, where most of the moisture inside the bushing is concentrated in the insulating paper and the moisture content in the oil is very small, make it impossible to accurately assess the moisture content inside the oil-paper capacitor bushing.

[0026] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0027] Figure 1 This is a flowchart of the casing moisture content testing method of the present invention;

[0028] Figure 2 This is a diagram of the BP neural network structure of the present invention.

[0029] Figure 3 The particle swarm optimization algorithm of this invention applies weight factor Q. n Optimization flowchart;

[0030] Figure 4This is a schematic diagram of the circuit wiring for the method of testing the water content of the casing in the present invention. Specific embodiments

[0031] The following will further describe the specific embodiments of the present invention in detail in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0032] Refer Figure 1 As shown, this embodiment provides a method for testing the water content of the casing, including the following steps:

[0033] S1. Based on the FDS method, the dielectric spectrum response test of the casing with different water contents is carried out within the frequency range of 1000 Hz - 0.01 Hz to obtain the dielectric spectrum curves of the casing under different water contents. Further, for the convenience of description, the specific wiring schematic diagram is as Figure 4 shown, where 1 is a dielectric spectrum analyzer based on FDS; 2 is the L1 line of the test voltage output terminal of the dielectric spectrum analyzer; 3 is the L2 line of the test current input terminal of the dielectric spectrum analyzer; 4 is the L3 line of the working grounding terminal of the dielectric spectrum analyzer; 5 is the casing. The specific wiring is that the L1 line of the voltage output terminal is connected to the top of the casing; the L2 line of the current input terminal is connected to the end screen of the casing; the L3 line of the working grounding terminal is grounded, and then the dielectric spectrum response test of the casing with different water contents is carried out.

[0034] S2. Extract the characteristic parameters of each dielectric spectrum response curve based on the FDS method within the frequency range of 1000 Hz - 0.01 Hz. First, divide 1000 Hz - 0.01 Hz into 5 frequency bands, namely frequency band ① 1000 Hz - 100 Hz, frequency band ② 100 Hz - 10 Hz, frequency band ③ 10 Hz - 1 Hz, frequency band ④ 1 Hz - 0.1 mHz, and ⑤ 0.1 Hz - 0.01 Hz. Further, the characteristic parameters are respectively taken as the conductance amplitude A n , the conductance slope a n , the polarization amplitude X n , the polarization time constant τ n , n = 1 - 5, that is, within the frequency band ① 1000 Hz - 100 Hz, they are respectively recorded as: conductance amplitude A1, conductance slope a1, polarization amplitude X1, and polarization time constant τ1. Similarly, for the other 4 frequency bands of ②, ③, ④, and ⑤, the characteristic parameters are analogized in turn.

[0035] S3. Take the characteristic parameters extracted from each dielectric spectrum curve based on the FDS method as the input layer, and the water content corresponding to each dielectric spectrum curve as the output layer, and train the target model function with the BP neural network model Furthermore, the neural network function described in this step is the error feedback neural network algorithm. Structurally, a BP neural network consists of two modules: a forward propagation network for information and a backward propagation network for error. The basic structure of a BP neural network is as follows: Figure 2 As shown.

[0036] from Figure 2 As can be seen from its structure, a backpropagation (BP) neural network mainly consists of three layers: the input layer, the hidden layers, and the output layer. Various information from the outside is transmitted through the input layer to the hidden layers for processing, and then output through the output layer to obtain the final result. When the error between the output and the pre-set input value is large, the network enters the backpropagation phase, updating the network weights until the error between the output and the expected result meets a certain condition.

[0037] The main steps in the forward propagation of the signal are as follows:

[0038] First, the input variable net of the i-th node in the hidden layer of the neural network. i :

[0039]

[0040] Second, the output variable y of the i-th node in the hidden layer of the neural network. i :

[0041]

[0042] Third, the input variable net of the k-th node in the output layer of the neural network. k :

[0043]

[0044] Fourth, the output variable O of the k-th node in the output layer of the neural network. k :

[0045]

[0046] Where the variable x j The meaning of represents the input parameters of the j-th node in the input layer of the BP neural network. It should be noted that here, x j The conductivity amplitude A is respectively n Conductivity slope a n Polarization amplitude X n Polarization time constant τ n n = 1 to 5; w ijThe variable θ represents the neural network weight parameters between the i-th node of the hidden layer and the j-th node of the input layer in a BP neural network. i The meaning of represents the threshold parameter of the i-th node in the hidden layer of a BP neural network; variable The meaning of represents the activation function of the hidden layer in a BP neural network; the variable w ki The meaning represents the weight parameters between the k-th node of the output layer and the i-th node of the hidden layer in a BP neural network, where i = 1 to q; variable α k The meaning of represents the threshold parameter of the k-th node in the output layer of the BP neural network, k = 1 to L; the meaning of variable Ψ(x) represents the activation function of the output layer of the BP neural network; the meaning of variable o k The meaning of represents the output of the k-th node in the output layer of the BP neural network. It should be noted that the output variable o here k This refers to the water content.

[0047] S4. Target model under different moisture contents Each frequency band is assigned a weighting factor Q. n That is, Q1 corresponds to frequency band ① 1000Hz~100Hz, Q2 corresponds to frequency band ② 100Hz~10Hz, Q3 corresponds to frequency band ③ 10Hz~1Hz, Q4 corresponds to frequency band ④ 1Hz~0.1mHz, and Q5 corresponds to frequency band ⑤ 0.1Hz~0.01Hz. After adding weighting factors, it becomes Then, the particle swarm optimization (PSO) algorithm is used to adjust the weighting factor Q. n Optimization is performed to obtain the most accurate target model. Furthermore, the particle swarm optimization algorithm process described in this step is as follows: Figure 3 As shown:

[0048] S5. Finally, obtain the accurate test model for casing water content based on FDS. The water content obtained based on this model is the specific water content of the tested casing.

[0049] This invention first uses a Fibre Channel Analyzer (FDS) to test the dielectric response curves of casings with different water contents within a frequency range of 1000Hz-0.01Hz, obtaining FDS-based dielectric spectrum curves for different water contents. Then, feature parameters are extracted from the FDS-based dielectric spectrum curves. Next, using the extracted feature parameters and frequency as the input layer and water content as the output layer, a backpropagation (BP) neural network is used to fit (train) the target model function, denoted as y. (FDS) Finally, the particle swarm optimization algorithm is used to optimize the weighting factors for each frequency band, resulting in the most accurate target model. This allows for precise testing of the moisture content of the transformer's solid insulation, achieving the following technical benefits:

[0050] 1) The testing method proposed in this invention is not easily affected by leakage current, temperature, etc., and can accurately assess the insulation status of bushings. It overcomes the limitations of conventional preventive tests, such as insulation resistance testing, dissolved gas chromatography analysis in oil, dielectric loss factor, and capacitance testing, which are easily affected by surface leakage current, temperature, and humidity, and are prone to deviation.

[0051] 2) The testing method proposed in this invention can obtain specific moisture content values ​​for specific bushings, and can quantitatively analyze and track the changes in moisture content of bushings over several months or years. It solves the problems of conventional testing methods being limited by their specific methodological principles, greatly affected by on-site factors, and the possibility of large deviations between adjacent test results, and the inability to quantitatively analyze and track the degree of insulation aging of bushings over several years or months.

[0052] 3) The bushing moisture content testing method based on frequency domain spectrum method (FDS) proposed in this invention can accurately measure the moisture content in the insulating paper of oil-paper capacitor bushings. This solves the problem that current testing methods for oil-paper capacitor bushings, where most of the moisture inside the bushing is concentrated in the insulating paper and the moisture content in the oil is very small, make it impossible to accurately assess the moisture content inside the oil-paper capacitor bushing.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of testing the water content of a casing, characterized by, Includes the following steps: Step 1: Based on the FDS method, conduct dielectric spectrum response tests on the casing with different water contents in the frequency range of 1000Hz-0.01Hz to obtain the dielectric spectrum curves of the casing under different water contents; Step 2: Extract characteristic parameters from the spectral response curves of each medium in the frequency range of 1000Hz-0.01Hz based on the FDS method. This includes dividing the 1000Hz-0.01Hz range into 5 frequency bands: band ① 1000Hz~100Hz, band ② 100Hz~10Hz, band ③ 10Hz~1Hz, band ④ 1Hz~0.1mHz, and band ⑤ 0.1Hz~0.01Hz. The characteristic parameter is the conductivity amplitude A in each of the 5 frequency bands. n Conductivity slope a n Polarization amplitude X n Polarization time constant τ n n=1~5, that is, within the frequency band ① 1000Hz~100Hz, they are respectively denoted as: conductivity amplitude A1, conductivity slope a1, polarization amplitude X1, polarization time constant τ1. Similarly, the characteristic parameters of the other 4 frequency bands ②, ③, ④, and ⑤ are deduced in turn. Step 3: Using the feature parameters extracted from the spectral curves of each medium based on the FDS method as the input layer and the water content corresponding to each spectral curve as the output layer, the target model function is trained using a BP neural network model. , n∈1~5; Step 4, for the target model under different water contents Each frequency band is assigned a weighting factor Q. n The frequencies are as follows: ① 1000Hz~100Hz corresponds to Q1; ② 100Hz~10Hz corresponds to Q2; ③ 10Hz~1Hz corresponds to Q3; ④ 1Hz~0.1mHz corresponds to Q4; and ⑤ 0.1Hz~0.01Hz corresponds to Q5. After adding weighting factors, it becomes The particle swarm optimization algorithm is used to adjust the weight factor Q. n Optimization is performed to obtain the most accurate target model; Step 5: Obtain an accurate test model for casing water content based on FDS based on the optimized target model. The water content obtained based on this model is the specific water content of the tested casing. The BP neural network model described in step 3 consists of two modules: a forward propagation network for information and a backward propagation network for error. Various information from the outside is transmitted through the input layer of the BP neural network into its hidden layer for network operation and processing, and the final processing result is obtained through the output layer. When the error between the output result of the BP neural network and its preset input value is large, the backward propagation stage of the BP neural network is entered, and the network weights are updated until the error between the output result and the expected result meets the set conditions. The forward propagation process of the signal is as follows: 1) The input variable net of the i-th node in the hidden layer of the neural network i : 2) The output variable y of the i-th node in the hidden layer of the neural network i : 3) The input variable net of the k-th node in the output layer of the neural network k : 4) The output variable O of the k-th node in the output layer of the neural network k : Wherein, variable x j Let x represent the input parameters of the j-th node in the input layer of the BP neural network. j The conductivity amplitude A is respectively n Conductivity slope a n Polarization amplitude X n Polarization time constant τ n n=1~5; w ij The variable θ represents the neural network weight parameters between the i-th node of the hidden layer and the j-th node of the input layer in a BP neural network. i Let represent the threshold parameter of the i-th node in the hidden layer of the BP neural network; let Ø(x) represent the activation function of the hidden layer of the BP neural network; let w represent the threshold parameter of the i-th node in the hidden layer of the BP neural network. ki The weights between the k-th node of the output layer and the i-th node of the hidden layer in a BP neural network are represented, i=1~q; variable α k The variable Ψ(x) represents the threshold parameter of the k-th node in the output layer of the BP neural network, k=1~L; the variable o represents the activation function of the output layer of the BP neural network; k This represents the output of the k-th node in the output layer of a BP neural network, with the output variable o. k This refers to the water content.

Images