A prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network

Abstract

The present invention relates to an SF6 pressure prediction method based on a BP neural network. By analyzing the influence of internal and external factors such as ambient temperature and its variation, relative humidity, wind speed, weather type, and conductor current on the change of SF6 pressure, a BP neural network prediction model is established to predict the pressure value of the SF6 online monitoring device, providing data reference for the normal operation of the monitoring equipment, and a discrimination strategy for analyzing the reasons for the pressure reduction of GIS equipment is proposed. The gas pressure of a typical SF6 online monitoring device of a UHV substation GIS equipment and the corresponding natural environment and conductor current data are selected for MATLAB simulation. The results show that the accuracy of the SF6 pressure prediction result can reach more than 98.5%, and the correctness of the theoretical analysis is verified.

Description

Technical Field

[0001] The present invention belongs to the technical field of power operation and maintenance equipment, and particularly relates to a method for predicting the pressure of an SF6 on-line monitoring device based on a BP neural network. Background Art

[0002] Due to its good electrical insulation performance and excellent arc extinguishing ability, SF6 gas is widely used in equipment such as SF6 circuit breakers and gas-insulated switchgears (GIS). During the operation of the equipment, the density of SF6 gas is one of the important factors determining the arc extinguishing and insulation capabilities of the circuit breaker. The SF6 pressure of the corresponding equipment can be read through an SF6 density relay, or the real-time tracking and recording of the SF6 pressure can be achieved by installing an SF6 on-line monitoring device.

[0003] Affected by external natural environments and internal conductor currents and other factors, the SF6 pressure data shown by the above two devices have non-linear fluctuations. If the displayed SF6 pressure is lower than the preset value, the system will misissue alarm or blocking signals, affecting the normal operation of the equipment. Therefore, it is necessary to study and predict the variation law of the SF6 pressure of the equipment, so as to analyze and judge whether the change of the SF6 pressure of the equipment at a certain moment is a normal phenomenon, and improve the reliability of the equipment operation.

[0004] Zhang Licheng et al. studied an SF6 low-pressure alarm event of an SF6 porcelain column circuit breaker, and pointed out that it is necessary to ensure that the SF6 density relay is consistent with or close to the environment of the main gas chamber, so that the compensation function of the density relay can be correctly realized; Zhen Li et al. analyzed the reasons for the misissued SF6 gas low-pressure alarm signal of a 220 kV circuit breaker, and proposed to adopt the method of a temperature sensing element so that the density relay can sense the ambient temperature of the circuit breaker body, thereby avoiding misissuing pressure blocking signals. Wang Zhibin studied the problem of circuit breaker blocking caused by SF6 liquefaction in severely cold areas in winter, and solved it by taking the method of temporarily heating the circuit breaker; Li Haibo summarized the structural characteristics of various types of SF6 density controllers, qualitatively analyzed the influence of factors such as temperature compensation methods, altitude, temperature rise of electrical equipment, and meter oil leakage on accurate gas monitoring, and put forward relevant precautions in combination with product design, operation and maintenance. Chen Yuanming studied the compensation method for reducing the measurement error of the SF6 on-line monitoring device, and verified it with test data. The existing research mainly focuses on the qualitative judgment of SF6 pressure changes and reducing measurement errors, etc., lacking quantitative analysis and prediction of the pressure changes of the actual operating SF6 gas chamber. Summary of the Invention

[0005] The object of the present invention is to provide an SF6 on-line monitoring pressure prediction method based on a BP neural network. Taking the pressure of a certain model of SF6 on-line monitoring device of a 1000 kV GIS device as the research object, analyzing the influence of external natural environment and internal current and other factors on the change of SF6 pressure, establishing a BP neural network prediction model to predict the SF6 pressure, and proposing a discrimination strategy for analyzing the reasons for the reduction of the GIS device pressure, so as to provide a reference for monitoring the operation state of the device and identifying the reasons for the pressure reduction.

[0006] The present invention adopts the following technical solutions:

[0007] A prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network, which includes the following steps:

[0008] (1) Data selection and processing: Select the pressure data collected by the on-line monitoring device of the gas chamber to be measured of the interval circuit breaker, the internal current data of the interval circuit breaker at the corresponding moment, and the environmental data at the corresponding moment;

[0009] (2) Design of the BP neural network model;

[0010] (3) Training of the BP neural network

[0011] (4) Prediction and analysis of the pressure of the SF6 on-line monitoring device by the BP neural network.

[0012] In step (1), the mapminmax function is used to normalize various types of data, and the value range of each variable is [-1, 1].

[0013] In step (1), the environmental data includes environmental temperature, relative humidity, weather type, wind speed, and temperature change rate.

[0014] In step (1), the sampling interval is 15 minutes.

[0015] In step (2), the BP neural network model adopts a 3-layer topological structure BP neural network, the number of input layer nodes is 6, which are environmental temperature, relative humidity, weather type, wind speed, temperature change rate, and conductor current; the number of output layer nodes is 1.

[0016] In step (2), the transfer function of the hidden layer is set to tansig, the transfer function of the output layer is set to logsig, and the training function is set to trainlm; the number of hidden layer nodes is 13.

[0017] In step (3), the BP neural network toolbox in MATLAB is used for simulation training.

[0018] In step (4), the pressure error data predicted by the BP neural network model is used to set the error threshold P0 to measure the effectiveness of the measurement error, and then the actual state of the device is analyzed.

[0019] In step (4), when the device issues a low air pressure alarm signal, the cause of the decrease in the SF6 pressure in the gas chamber can be identified by continuously observing the difference relationship between the predicted result P' and the actual pressure P.

[0020] In step (4), the result P' predicted by the BP neural network is compared with the actual pressure value P: If the difference between P' and P is basically less than P0 within 2 consecutive hours, it can be determined that the gas chamber pressure data is normal, and the change in the gas chamber pressure is caused by the change in the external environment, rather than a leakage fault in the device; If within 2 consecutive hours, the difference between P' and P is basically greater than P0, and the value of P shows a downward trend over time, it can be judged that the device has a leakage.

[0021] The beneficial effects of the present invention are as follows: The present invention mainly studies the pressure prediction method of the SF6 on-line monitoring device based on the BP neural network. By analyzing the influence of factors such as the external natural environment and the internal conductor current on the change of the SF6 pressure, a BP neural network prediction model is established to predict the pressure value of the SF6 on-line monitoring device, providing data reference for the normal operation of the monitoring device. Selecting the gas pressure data of a typical SF6 on-line monitoring device of the GIS equipment in a certain UHV substation and the corresponding natural environment and conductor current data for simulation analysis, the results show that: compared with the true value, the prediction accuracy of the pressure prediction value can reach more than 98.5%, verifying the effectiveness of the prediction model of the present invention. On this basis, a discrimination strategy for analyzing the cause of the decrease in the SF6 pressure of the GIS equipment is proposed, providing strong support for quickly analyzing the cause of the decrease in the device pressure and improving the operation reliability of the device. Description of the Drawings

[0022] Figure 1 It is a typical SF6 pressure curve of the on-line monitoring device for GIS circuit breakers.

[0023] Figure 2 It is an analysis diagram of the influencing factors of the SF6 pressure change.

[0024] Figure 3 It is a correlation result diagram of each influencing factor and the SF6 on-line monitoring pressure value.

[0025] Figure 4 It is a topological structure diagram of the BP neural network.

[0026] Figure 5 It is a BP neural network prediction model.

[0027] Figure 6 It is the training result of the BP network.

[0028] Figure 7 It is a regression analysis chart.

[0029] Figure 8 It is the overall prediction error chart of the BP neural network.

[0030] Figure 9 It is a normal distribution Q-Q chart.

[0031] Figure 10 It is a histogram of the prediction error distribution.

[0032] Figure 11 It is the prediction chart of the BP neural network. Specific implementation manners

[0033] The technical solutions will be clearly and completely described below in conjunction with the embodiments and drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0034] 1. SF6 density monitoring device

[0035] The existing SF6 density monitoring devices mainly include two types: SF6 density relays and SF6 on-line monitoring devices, which adopt different temperature compensation methods respectively.

[0036] The SF6 temperature compensation principle is as follows: The SF6 gas in high-voltage electrical equipment is sealed in a fixed container and has a certain density under the rated pressure at 20°C. In the case of no air leakage in the equipment, the gas pressure changes with the temperature, but its density remains unchanged all the time. In order to effectively monitor whether there is air leakage, the measured real-time pressure is converted into the corresponding pressure at 20°C, and this converted pressure can be used as the SF6 gas density. In essence, it uses the gas pressure at 20°C to represent the gas content in the fixed volume.

[0037] 1.1 SF6 density relay

[0038] Currently, most of the SF6 gas density relays used in high-voltage substations are mechanical pointer type.

[0039] The pointer-type SF6 density relay realizes temperature compensation through the expansion and contraction of the bimetallic strip inside the dial. Due to the small volume of the dial, although the density relay is at the same ambient temperature as the equipment, the temperature change rate of its dial is much higher than that of the main body. In addition, the influence of sunlight radiation change on the dial temperature cannot be ignored. The pressure data of this type of device needs to be read manually, with low data accuracy and difficult continuous data acquisition. It can be used for long-term data recording and comparison, but is not suitable for short-term prediction analysis.

[0040] 1.2 SF6 On-line Density Monitoring Device

[0041] With the development of intelligent monitoring technology, SF6 on-line monitoring devices with high data accuracy and the functions of real-time monitoring and data recording are becoming more and more common in the application of UHV and extra-UHV substation equipment.

[0042] The SF6 on-line monitoring device usually consists of signal acquisition, data processing and communication, and display parts. The signal acquisition module is mainly composed of a pressure sensor, a temperature sensor or a density sensor, and various sensors are usually installed at the three-way valve port on the extension of the circuit breaker or combined electrical appliance tank.

[0043] The data processing unit obtains the SF6 gas pressure at 20°C through the SF6 gas equation according to the collected physical quantities. The commonly used gas equations are shown in formulas (1)-(3):

[0044] P = 0.57×10 -4 ρT(1 + B) - ρ 2 A (1)

[0045] A = 0.75×10 -4 (1 - 0.727×10 -3 ρ) (2)

[0046] B = 2.51×10 -3 ρ(1 - 0.84×10 -3 ρ) (3)

[0047] Among them, P is the SF6 pressure, ρ is the SF6 density, and T is the ambient temperature. Combining formulas (1)-(3), the relationship between the pressure P and the density ρ can be obtained as formula (4).

[0048] P = (54.525 - 0.121T)ρ 3 + (0.143T - 75)ρ 2 + 0.057Tρ (4)

[0049] If the physical quantity measured by the detector is ρ, then the measured ρ and T = 293.2K (20°C) can be substituted into formula (4) to obtain the GIS chamber pressure value P at 20°C 20 ;

[0050] If the physical quantities measured by the detector are P and T, then iterative calculations need to be carried out according to formulas (1)-(3) to obtain the gas density at the corresponding temperature, and then ρ and T = 293.2K (20°C) are substituted into formula (4) to obtain the pressure value of the GIS equipment converted to 20°C.

[0051] Theoretically, temperature compensation can offset the influence of external environmental temperature changes on the SF6 pressure changes. That is, under normal conditions, the pressure curve displayed on the online monitoring background should be a straight line. In fact, since the sensor of the online monitoring device is installed at the three-way valve port on the outer extension of the tank, although the sensor is inside the tank, the physical quantities such as pressure or density collected by it are the data at the three-way valve interface inside the tank. The pressure or density at this local position is greatly affected by the temperature changes at the location. Even the data obtained after conversion through the gas equation is not an approximately horizontal curve, but shows a certain fluctuation pattern, especially with large fluctuations during periods of large temperature differences in spring and autumn or drastic changes in weather conditions.

[0052] Typical SF6 online monitoring data of a GIS device in a UHV substation are as Figure 1 shown. Since the pressure data sampling interval of the online monitoring device is continuous, the data accuracy is high, and it shows a certain fluctuation regularity, which is suitable for predictive analysis.

[0053] 2. Analysis of influencing factors for SF6 pressure changes

[0054] Based on the above analysis, since the sensor of the SF6 online monitoring device is installed at the three-way valve interface of the GIS device shell, the local temperature change at the sampling point will affect the sensor measurement result. The present invention starts from the internal and external factors that affect the temperature change at the sampling point, studies the influence of each factor on the SF6 pressure change of the online monitoring device, and the analysis process is as Figure 2 shown.

[0055] 2.1 External factors

[0056] The influence of the external environment on the temperature at the sampling point of the SF6 online monitoring device sensor is relatively direct, such as sunlight exposure, cloudy, foggy, rainy or drastic temperature change conditions, etc. Therefore, the present invention first analyzes the influence of factors such as environmental temperature, relative humidity, and weather type on the SF6 pressure measurement result of the online monitoring device.

[0057] (1) Environmental temperature

[0058] The level of environmental temperature directly affects the sampling point temperature, and further affects the sensor measurement result of the online monitoring device. Therefore, environmental temperature is first selected as the influencing factor for analysis, and is represented by the symbol T.

[0059] (2) Relative humidity

[0060] Relative humidity refers to the percentage of water vapor pressure in the air to the saturated water vapor pressure, and is represented by the symbol RH. There are differences in the local temperature change of an object under different humidities, that is, humidity affects the temperature change at the location where the sensor is located. Therefore, relative humidity is selected as the factor affecting SF6 pressure.

[0061] (3) Wind speed

[0062] Wind speed refers to the flow velocity of air, denoted by the symbol F. Considering the influence of wind speed on the local tank temperature change, this paper also studies wind speed as one of the factors affecting the pressure change of the SF6 on-line monitoring device.

[0063] (4) Weather type

[0064] Since the radiation received by the sensor at the location is different under different weather types (sunny, cloudy, rainy, snowy), which affects the temperature sampling effect of the sensor, the present invention studies the weather type as one of the factors affecting the pressure change of the SF6 on-line monitoring device. For quantitative analysis, the fuzzy set theory method is adopted for the treatment of weather type, and the fuzzy function is used to represent the characteristic values of weather. The membership degree of each weather characteristic value is denoted by the symbol X, and each weather characteristic value is shown in Table 1.

[0065] Table 1 Membership degree of weather characteristic values

[0066]

[0067] (5) Temperature change rate

[0068] The temperature change rate described in the present invention refers to the temperature change within a unit time interval. Since the severity of temperature change also cannot be ignored in affecting the temperature sampling value of the sensor, therefore, the present invention separately analyzes the environmental temperature change as an influencing factor, denoted by the symbol Δθ, and the calculation formula is as follows:

[0069]

[0070] Among them, T1 is the temperature value at the current sampling moment, T0 is the temperature value at the previous sampling moment, and t is the sampling time interval, with the unit of min.

[0071] 2.2 Internal factors

[0072] The internal conductor of the electrical equipment generates temperature rise through current, which causes the internal temperature of the GIS gas chamber to increase (the operating temperature rise can reach 65K), and then affects the temperature at the location of the SF6 on-line monitoring device sensor. Therefore, the present invention selects the magnitude of the current in the internal conductor of the GIS equipment as the influencing factor for the SF6 pressure change, denoted by the symbol I.

[0073] 2.3 Correlation analysis

[0074] In order to calculate the influence degree of each influencing factor on the SF6 pressure change, this paper selects the on-line monitoring pressure values in the typical seasons of the operating equipment in a certain UHV substation and the corresponding external natural environment data and internal conductor current data for correlation calculation. The data sampling interval is 15 minutes, with a total of 1800 groups, and some sample data are shown in Table 2.

[0075] Table 2 Partial sample data

[0076]

[0077] The correlation results between each influencing factor and the SF6 on-line monitoring pressure value are as follows Figure 3 shown

[0078] In practical applications, if the obtained fuzzy correlation coefficient is within a certain small range, such as within [-0.1, 0.1], it can be considered that the correlation is weak and can be ignored. The absolute values of the correlation coefficients between the six influencing factors selected in the present invention and the pressure are all greater than 0.1, and it can be considered that there is a correlation between the six influencing factors proposed in the present invention and the pressure of the SF6 on-line monitoring device

[0079] Among them, the absolute values of the correlation coefficients between the ambient temperature T and the conductor current I and the SF6 on-line monitoring pressure P both exceed 0.5, and the signs are negative. Therefore, the ambient temperature T and the conductor current I have a significant negative correlation with the pressure P. This is because according to formula (4), when the temperature T at the sensor of the on-line monitoring device changes, there is a negative correlation between the SF6 on-line monitoring pressure P and the temperature T (the measured density of the SF6 gas fluctuates around 6.088 kg / m 3 nearby, and this density value can be substituted into the relevant formula to calculate the relationship between P and T). The temperature change rate Δθ and P are actually correlated, and the remaining influencing factors and P are slightly correlated

[0080] 3. BP neural network algorithm

[0081] The BP neural network is one of the most widely used artificial neural networks, also called the error backpropagation network, with advantages such as strong learning ability, good non-linear mapping ability, and good fault tolerance. During the operation of the algorithm, information travels forward and the error propagates backward to correct the network. The core of the algorithm of this network is the first-order gradient method (the steepest descent method), and by optimizing the connection weights between layers, the sum of the squared errors between the actual output value and the ideal output of the neural network is minimized

[0082] A typical BP neural network consists of an input layer, a hidden layer, and an output layer, and its network model is as follows Figure 4 shown. X is the input layer of the network, the number of network nodes is N, w i,j is the connection weight between the input layer and the hidden layer, b ,j is the threshold of the hidden layer, the number of network nodes in the hidden layer is M, w j,k is the connection weight between the hidden layer and the output layer, b ,k is the threshold of the output layer, and Y is the output layer of the network, with the number of nodes being Q

[0083] The mathematical model of the neuron output in the network is

[0084]

[0085] Among them, x is the neuron input, u is the neuron output, w is the weight value, and b is the threshold of the neuron.

[0086] If the sigmoid function is selected as the activation function, the mathematical model is:

[0087]

[0088] Therefore, the output of the j-th neuron in the hidden layer is obtained as:

[0089]

[0090] Similarly, the output of the k-th neuron in the output layer can be obtained as:

[0091]

[0092] The error between the actual output value and the expected value after network training is:

[0093]

[0094] Among them, O k is the expected value of the k-th output sample.

[0095] Substituting the output relationships of each layer into the error formula (10), the relationship between the error and the weights of each layer can be obtained. The weights between each layer are adjusted step by step in the direction of the error gradient descent until the error E meets the requirements.

[0096] 4. SF6 Pressure Prediction and Analysis Based on BP Neural Network

[0097] 4.1 Data Selection and Processing

[0098] The SF6 pressure data selected in the present invention are typical seasonal pressure data collected by the on-line monitoring device of the A-phase gas chamber of the circuit breaker in a certain interval of 1000 kV of a certain UHV substation during normal operation (the manufacturer of the on-line monitoring device is Simatex of Switzerland, and the sensor product model is: trafag8774). The internal current data of the circuit breaker at the corresponding moment are from the monitoring records in the substation. The environmental data at the corresponding moment are obtained through the official website of the "China Meteorological Administration".

[0099] The sampling interval of all data is 15 minutes, and the total number of data is 1800 groups. To improve the convergence speed, the mapminmax function is used to normalize all types of data, and the value range of each variable is [-1, 1].

[0100] 4.2 BP Neural Network Model Design

[0101] The present invention adopts a three-layer topology BP neural network. The number of nodes in the input layer is 6, that is, N = 6, which are the ambient temperature, relative humidity, weather type, wind speed, temperature change rate, and conductor current affecting the pressure of the SF6 on-line monitoring device respectively; the number of nodes in the output layer is M = 1. The transfer function of the hidden layer is set to tansig, the transfer function of the output layer is set to logsig, the training function is set to trainlm, the learning rate is 0.1, and the target accuracy is 0.00001. The number of nodes in the hidden layer is selected according to the empirical formula (11).

[0102]

[0103] Among them, a is an adjustment constant, and its value ranges from 1 to 10. After multiple training comparisons, it is finally determined that the number of nodes in the hidden layer of this network is 13, that is, M = 13.

[0104] The BP neural network prediction model is as Figure 5 shown.

[0105] 4.3 BP Neural Network Training and Testing

[0106] The present invention uses the BP neural network toolbox in MATLAB for simulation. 75% of the total number of samples, that is, 1260 groups of data, are selected as the training samples of the network; 15% of the total number of samples, that is, 270 groups of data, are selected as the test samples; the remaining 15%, that is, 270 groups of data, are used for verification. When iterating to the 26th time, the network prediction result reaches the best state, as Figure 6 shown.

[0107] Through regression analysis( Figure 7 ) it can be seen that the training, verification and testing of the BP neural network obtained by simulation are in good condition, and the R value of the overall data is about 0.94. The overall prediction error (training data, verification and testing data) of the obtained BP neural network is as Figure 8 shown. The maximum error value of 1800 groups of samples does not exceed ±0.008 MPa, the prediction accuracy is above 98.5%, the prediction result accuracy is relatively high, and the correctness of the theoretical analysis is verified.

[0108] The statistical Q-Q diagram of the prediction error of the BP neural network is as Figure 9 shown. Most of the data points in this diagram tend to fall on a straight line in the first quadrant. Therefore, it can be approximately considered that the prediction error of the SF6 on-line monitoring device pressure obeys the normal distribution.

[0109] The prediction error distribution histogram of this BP network is as Figure 10 shown.

[0110] The estimated value of the prediction error mean is -5.53×10 -6MPa, the 95% confidence interval of the mean error is [-1.52×10 -5 , 4.14×10 -6 MPa.

[0111] The trained BP neural network model is used for simulation, and the results are as Figure 11 shown. The obtained prediction model can better track the actual change trend and effectively predict the pressure fluctuations caused by external natural environment and internal current changes.

[0112] 4.4 Prediction Application

[0113] According to the usage instructions provided by the manufacturer and the actual operation and maintenance of the operation and maintenance personnel, during the daily operation of this type of SF6 on-line monitoring device, false alarms of SF6 pressure often occur due to changes in the external environment. When this kind of defect occurs, the operation and maintenance personnel often cannot determine whether the pressure change is caused by gas chamber leakage or external environmental changes through on-line monitoring data. According to the previous analysis, the pressure error data predicted by the BP neural network model established in the present invention approximately follows a normal distribution, and within the 95% confidence interval, the error expectation value is -5.53×10 -6 MPa. Therefore, an appropriate error threshold P0 can be set to measure the effectiveness of the measurement error, and then the actual state of the equipment can be accurately analyzed. According to the analysis of the actual pressure data change range of the on-line monitoring device and the recording accuracy of the equipment itself, the prediction error threshold P0 can be set to 2×10 -3 MPa. When the SF6 pressure P in the gas chamber of the GIS equipment is lower than the alarm value P W , and the "low pressure alarm" signal is issued, the equipment operation and maintenance personnel can analyze the leakage defect through the following strategy:

[0114] First, compare the result P' predicted by the BP neural network with the actual pressure value P.

[0115] 1) If the difference between P' and P is basically less than P0 within 2 consecutive hours (15 minutes as a sampling point, a total of 8 sampling points), it can be determined that the gas chamber pressure data is normal, and the gas chamber pressure change is caused by external environmental changes, rather than a leakage fault of the equipment.

[0116] 2) If within 2 consecutive hours, the difference between P' and P is basically greater than P0, and the P value shows a downward trend over time, it can be judged that the equipment has a leakage.

[0117] The above discriminant strategy for the cause of SF6 pressure reduction is shown in Table 3.

[0118] Table 3 Discriminant Strategy for SF6 Pressure Reduction of GIS Equipment

[0119]

[0120] For GIS devices of different types and different phases, appropriate BP neural networks can be trained according to their own external environment and internal conductor current data for pressure prediction, and an appropriate pressure prediction error threshold P0 can be set. When the "low air pressure alarm" signal occurs in the device, the cause of the reduction of the SF6 pressure in the gas chamber can be identified by continuously observing the difference relationship between the predicted result P' and the actual pressure P.

[0121] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principle and practical application of the present invention, so that those skilled in the art can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network, characterized in that, It includes the following steps: (1) Data selection and processing: Select the pressure data collected by the on-line monitoring device of the gas chamber to be measured of the disconnector, the internal current data of the disconnector at the corresponding moment, and the environmental data at the corresponding moment; use the mapminmax function to normalize various types of data, and the value range of each variable is [-1, 1]; (2) Design of BP neural network model; (3) Training of BP neural network; (4) Prediction and analysis of the pressure of the SF6 on-line monitoring device by the BP neural network; Set the error threshold P0 by using the pressure error data predicted by the BP neural network model; when the low air pressure alarm signal occurs in the equipment, the cause of the decrease in the SF6 pressure of the gas chamber can be identified by continuously observing the difference relationship between the predicted result P' and the actual pressure P; If the difference between P' and P is basically less than P0 within 2 consecutive hours, it can be determined that the gas chamber pressure data is normal, and the change in the gas chamber pressure is caused by the change in the external environment, rather than the leakage failure of the equipment; If the difference between P' and P is basically greater than P0 within 2 consecutive hours, and the value of P shows a downward trend over time, it can be judged that the equipment has a leakage.

2. The prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network according to claim 1, characterized in that, The environmental data includes environmental temperature, relative humidity, weather type, wind speed, and temperature change rate.

3. The prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network according to claim 1, characterized in that, In step (1), the sampling interval is 15 minutes.

4. The prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network according to claim 1, characterized in that, In step (2), the BP neural network model adopts a 3-layer topological structure BP neural network, and the number of input layer nodes is 6, namely environmental temperature, relative humidity, weather type, wind speed, temperature change rate, and conductor current; the number of output layer nodes is 1.

5. The prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network according to claim 1, characterized in that, In step (2), the number of hidden layer nodes is 13.

6. The prediction method for the pressure of an SF6 on-line monitoring device based on a BP neural network according to claim 1, characterized in that, In step (3), use the BP neural network toolbox in MATLAB for simulation training.

Images