GIS equipment multi-air-chamber SF6 automatic air supply system and method based on deep learning
The deep learning-based GIS equipment multi-chamber SF6 automatic gas replenishment system solves the problems of high cost, long cycle and high safety risk of SF6 gas replenishment in the existing technology, realizes an efficient and safe automatic gas replenishment process, and ensures the stable operation of equipment and power grid.
Patent Information
- Application Number
- CN202511703597.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-19
- Publication Date
- 2026-01-23
AI Technical Summary
Existing SF6 gas replenishment methods suffer from high costs, long cycles, and high safety risks, especially equipment failures and poisoning risks caused by manual operation.
A multi-chamber SF6 automatic gas replenishment system based on deep learning is adopted for GIS equipment. It includes a gas pipeline module, valve assembly, intelligent monitoring module, control module and human-machine interface. Through the collaborative optimization of deep learning model and hardware, intelligent monitoring and automatic gas replenishment of SF6 status in each chamber are realized.
It enables accurate monitoring and efficient automatic gas replenishment of SF6 status in each gas chamber of GIS equipment, reduces costs, improves gas replenishment efficiency, avoids safety risks caused by human operation, and ensures the reliability of equipment and the safe and stable operation of the power grid.
Smart Images
Figure CN121383083A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of automatic control of power equipment, and particularly relates to a GIS device multi-gas chamber SF6 automatic gas supplementing system and method based on deep learning. BACKGROUND
[0002] A gas insulated switchgear (GIS) is composed of circuit breakers, disconnectors, grounding switches, current transformers, voltage transformers, arresters, busbars, connecting pieces and outgoing terminal ends, etc. All the devices or components are enclosed in a metal grounded shell, and a certain pressure of SF6 insulation gas is filled in the shell, so the GIS is also called SF6 fully enclosed combined electric appliance. Among them, SF6 gas is decomposed into fluorides, hydrofluoric acid and other toxic and harmful gases after the circuit breaker is charged, opened or closed, or in the case of fault. Currently, the gas pressure supplementing is mostly in a manual gas supplementing mode. During the gas supplementing process, a person needs to be close to the gas filling port, and there is a risk of poisoning and safety due to the contact between the person and the gas. Meanwhile, improper operation in the manual gas supplementing mode can cause device failure, and it is time-consuming and laborious to supplement the gas in each gas chamber. The existing SF6 gas supplementing has the following shortcomings:
[0003] 1. High cost: the traditional gas supplementing mode is manual gas supplementing in each gas chamber. Each time the gas supplementing needs to be carried out, a gas cylinder needs to be taken to a substation, and high human resource cost is consumed.
[0004] 2. Long supplementing period: in the traditional gas supplementing mode, there are pressure confirmation, gas cylinder pulling and other work between the discovery of low pressure in a gas chamber and the completion of the gas supplementing, the supplementing period is long, and the operation risk of a power grid is high.
[0005] 3. High safety risk: in the traditional gas supplementing mode, a person needs to be close to the gas filling port, and SF6 gas is decomposed into fluorides, hydrofluoric acid and other toxic and harmful gases in the case of fault. There is a risk of poisoning and safety due to the contact between the person and the gas.
[0006] Therefore, a multi-gas chamber SF6 automatic gas supplementing device based on deep learning is developed, so as to save the gas supplementing cost, improve the gas supplementing efficiency, avoid the safety and operation risk caused by manual exhaust, and has positive significance. SUMMARY
[0007] The technical problem to be solved by the present application is that, in view of the deficiencies in the prior art, a GIS device multi-gas chamber SF6 automatic gas supplementing system and method based on deep learning are developed. Through the cooperative optimization of a deep learning model and hardware, intelligent monitoring of SF6 states in each gas chamber, supplementing trend prediction and high-efficiency automatic gas supplementing full-process automation are realized. The SF6 states in each gas chamber of the GIS device can be accurately monitored and high-efficiency automatic gas supplementing can be realized, so that the reliability of the device and the safe and stable operation of a power grid are ensured.
[0008] The technical scheme of the present application is realized by the following measures:
[0009] A GIS device multi-gas chamber SF6 automatic gas filling system based on deep learning, comprising: a gas pipeline module, a valve assembly, an intelligent monitoring module, a control module, and a man-machine interface;
[0010] The gas pipeline module includes a weighing sensor, a multi-way valve, and a connecting pipeline, the weighing sensor is installed on the gas cylinder transport trolley tray, and the weight of the gas cylinder during the entire gas filling process can be measured, and the multi-way valve and the connecting pipeline are used to complete the gas filling of one gas cylinder for multiple gas chambers in the same interval;
[0011] The structure of the valve assembly includes a plurality of electromagnetic valves, a double one-way valve, and an exhaust valve; the electromagnetic valves and the double one-way valve are installed at the outlets of a plurality of SF6 gas chamber joints, and the exhaust valve is arranged at the exhaust port of the multi-way valve;
[0012] The intelligent monitoring module includes a pressure sensor, a temperature sensor, a humidity sensor, and a transmitter, which are used for data acquisition and transmission to the control module;
[0013] The control module receives data information transmitted by the intelligent monitoring module, controls the electromagnetic valve, records time, pressure, temperature, humidity, and gas filling frequency data information through a storage unit, performs edge computing and big data analysis using an NVIDIA Jetson chip, accurately outputs the required gas filling rate of the gas chamber, predicts the next gas filling time, triggers early gas filling, avoids sudden changes in gas chamber pressure, and ensures equipment safety;
[0014] The man-machine interface: uses an Android platform tablet computer or mobile phone to exchange data with the main control chip assembly, reads fluid flow direction and controls flow rate, and provides man-machine interaction display, working state display, parameter setting, and operation history record.
[0015] As a preferred, the control module uses an STM32H7 chip to collect data information of the pressure sensor, the temperature sensor, and the humidity sensor, and controls the electromagnetic valve, and uploads the data to the mobile phone APP end through Bluetooth for recording.
[0016] As a preferred, the control module imports an LSTM-ATTENTION model package into an NVIDIA Jetson chip, the input dimension of the LSTM-ATTENTION model includes four-dimensional data of temperature, humidity, pressure, and gas flow rate within a 60-second window, and the output is the leakage rate in the next 5 minutes; wherein, after obtaining the input pressure, temperature, and humidity data, the LSTM uses the intermediate state of the data obtained by the hidden layer calculation, and then uses the function to solve the characteristic vector h t,j and the hidden state ht Similarity. The calculation method is as follows:
[0017]
[0018] In the formula: W s b is the weight matrix of the fully connected layer. s The bias vector value; For the hidden state h t Transpose; for h i,j with h t The correlation between the hidden layer vector attention weights α and the attention weights is typically calculated using the Softmax function. i After interacting with h i,j The output of the attention layer can be obtained by weighted summation. α i , The calculation methods are as follows:
[0019]
[0020] In the formula: t is the output node of the fully connected layer, and τ is the total length of the sequence.
[0021] Preferably, the control module preprocesses the collected data, uses a Kalman filter algorithm to dynamically weight the temperature, humidity, and pressure data, corrects errors in the gas state equation, and improves data reliability.
[0022] Preferably, the human-machine interface displays relevant information about the gas cylinders, including the cylinder number, the most recently recorded cylinder weight, the pressure value of each chamber, the most recent gas replenishment time and amount. A yellow alarm is triggered when the cylinder weight is less than 60 kg, and a red alarm is triggered when it is less than 50 kg. The interface also displays the real-time parameters and historical data of the corresponding cylinders in detail.
[0023] Preferably, the weighing sensor is a cantilever beam type weighing sensor.
[0024] An application method for a multi-chamber SF6 automatic gas replenishment system for GIS equipment based on deep learning includes the following steps:
[0025] (1) Prepare a gas cylinder with sufficient and qualified gas, connect it firmly to the pipeline, and make sure to secure the transport trolley to prevent it from tipping over;
[0026] (2) Install gas pipeline system and valves, and at the same time install pressure sensor, temperature sensor and humidity sensor at each gas chamber and gas cylinder to collect pressure, temperature and humidity data in real time;
[0027] (3) In actual operation, when there is an abnormal power outage or data acquisition equipment failure, resulting in missing or abnormal data in the original data, the acquired data will affect the establishment of the prediction model. It is necessary to preprocess the acquired data and use the Kalman filter algorithm to dynamically weight the temperature, humidity and pressure data, correct the error of the gas state equation, and improve the reliability of the data.
[0028] (4) Predict the future SF6 leakage rate trend of each gas chamber based on deep learning and trigger gas replenishment in advance: The deep learning model LSTM-ATTENTION is used to capture the time series dependency relationship by using a 4-layer LSTM to address the problem that the traditional PID algorithm cannot handle nonlinear leakage. The model has 256 hidden units and the attention mechanism dynamically weights key features, including the sudden pressure difference signal. The input dimensions include four-dimensional data of temperature, humidity, pressure and gas flow rate within a 60-second window, and the output is the leakage rate g / s for the next 5 minutes.
[0029] (5) Before replenishing gas, exhaust the pipeline to prevent dust and water vapor accumulated in the pipeline from entering the gas chamber, which may cause short circuit faults in the GIS during operation and cause the internal gas to deteriorate.
[0030] (6) Adjust the status of each valve dynamically according to the predicted leakage rate to achieve dynamic matching between the gas supply rate and the leakage rate, and avoid equipment damage caused by sudden pressure changes.
[0031] (7) After the gas replenishment is completed, the storage unit records data, which mainly includes the gas cylinder number, the most recently recorded gas cylinder weight, the pressure value of each gas chamber, and the most recent gas replenishment time and replenishment amount.
[0032] Compared with existing technologies, the advantages of this invention are as follows: Addressing the shortcomings of existing technologies, a deep learning-based multi-chamber SF6 automatic gas replenishment system for GIS equipment has been developed. This system comprises a gas pipeline module, valve components, an intelligent monitoring module, a control module, and a human-machine interface as its main core components. Through collaborative optimization of the deep learning model and hardware, the entire process of intelligent monitoring of SF6 status in each chamber, prediction of replenishment trends, and efficient automatic gas replenishment is fully automated. This invention successfully achieves accurate monitoring of the SF6 status in each chamber of GIS equipment and efficient automatic gas replenishment, saving replenishment costs, improving replenishment efficiency, avoiding safety and operational risks associated with manual venting, and ensuring the reliability of the equipment and the safe and stable operation of the power grid. Attached Figure Description
[0033] Appendix Figure 1 This is a schematic diagram of the hardware structure of the multi-chamber SF6 automatic gas replenishment device for GIS equipment according to an embodiment of the present invention.
[0034] Appendix Figure 2 This is a schematic diagram illustrating the principle of the control process of the present invention.
[0035] Appendix Figure 3 This is a schematic diagram of the automatic SF6 replenishment process for a multi-chamber GIS device according to an embodiment of the present invention.
[0036] Appendix Figure 4 This is a schematic diagram of the LSTM-Attention model structure of the present invention.
[0037] In the diagram: 1 is a gas cylinder, 2 is a pressure sensor, 3 is a humidity sensor, 4 is a temperature sensor, 5 is a gas cylinder transport trolley, 6 is a weighing sensor, 7 is an exhaust valve, 8 is a multi-way valve, 9 is a solenoid valve, and 10 is a double check valve. Detailed Implementation
[0038] The present invention is not limited to the following embodiments, and the specific implementation can be determined according to the technical solution of the present invention and the actual situation.
[0039] The present invention will be further described below with reference to embodiments:
[0040] Example 1: As shown in the attached document Figure 1 , 2 As shown in Figures 3 and 4, a deep learning-based GIS equipment multi-chamber SF6 automatic gas replenishment system includes: a gas pipeline module, a valve assembly, an intelligent monitoring module, a control module, and a human-machine interface.
[0041] The gas pipeline module includes a weighing sensor 6, a multi-way valve 8 and its connecting pipelines. The weighing sensor 6 is installed on the pallet of the gas cylinder transport trolley 5 and can measure the weight of the gas cylinder 1 throughout the gas replenishment process. The multi-way valve 8 and its connecting pipelines are used to replenish gas from one gas cylinder to multiple gas chambers in the same interval. The weighing sensor 6 is a cantilever beam type weighing sensor.
[0042] The valve assembly includes multiple solenoid valves 9, double check valves 10, and exhaust valves 7; the solenoid valves 9 and double check valves 10 are installed at the outlets of multiple SF6 gas chamber connectors, and the exhaust valves 7 are located at the exhaust port of the multi-way valve.
[0043] The intelligent monitoring module includes a pressure sensor 2, a temperature sensor 4, a humidity sensor 3, and a transmitter, which are used to collect data and send it to the control module.
[0044] The control module receives data from the intelligent monitoring module and controls the solenoid valve. Simultaneously, it records data such as time, pressure, temperature, humidity, and number of gas replenishments through the storage unit. Utilizing the NVIDIA Jetson chip for edge computing and big data analysis, it accurately outputs the rate at which the gas chamber needs to be replenished and predicts the time of the next gas replenishment, triggering early gas replenishment to avoid sudden pressure changes in the gas chamber and ensure equipment safety.
[0045] The human-machine interface uses an Android tablet or mobile phone to exchange data with the main control chip component, read the fluid flow direction and perform flow rate control, and provides human-machine interaction display, working status display, parameter setting and operation history.
[0046] The control module uses an STM32H7 chip to acquire data from pressure sensor 2, temperature sensor 4, and humidity sensor 3, and controls the solenoid valves. The data is then uploaded to a mobile app via Bluetooth for recording. The LSTM-ATTENTION model package is imported into the NVIDIA Jetson chip. The LSTM-ATTENTION model takes four dimensions as input: temperature, humidity, pressure, and gas flow rate within a 60-second window, and outputs the leakage rate for the next 5 minutes. The LSTM, after obtaining the input pressure, temperature, and humidity data, uses its hidden layers to calculate intermediate data states, and then uses a function... Solving for the eigenvector h t,j With hidden state h t Similarity. The calculation method is as follows:
[0047]
[0048] In the formula: W s b is the weight matrix of the fully connected layer. s The bias vector value; For the hidden state h t Transpose; for h i,j with h t The correlation between the hidden layer vector attention weights α and the attention weights is typically calculated using the Softmax function. i After interacting with h i,j The output of the attention layer can be obtained by weighted summation. α i , The calculation methods are as follows:
[0049]
[0050] In the formula: t is the output node of the fully connected layer, and τ is the total length of the sequence.
[0051] The control module preprocesses the collected data and uses a Kalman filter algorithm to dynamically weight the temperature, humidity, and pressure data, correcting errors in the gas state equation and improving data reliability.
[0052] The human-machine interface displays information related to the gas cylinders, including the cylinder number, the most recently recorded cylinder weight, the pressure value of each chamber, the most recent gas replenishment time and amount. A yellow alarm is triggered when the cylinder weight is less than 60kg, and a red alarm is triggered when it is less than 50kg. The interface also displays the real-time parameters and historical data of the corresponding cylinders in detail.
[0053] This invention, through deep learning models and hardware co-optimization, achieves full automation of the process from intelligent monitoring of SF6 status in each gas chamber to prediction of gas replenishment trends and efficient automatic gas replenishment. It successfully realizes accurate monitoring of SF6 status in each gas chamber of GIS equipment and efficient automatic gas replenishment, saving gas replenishment costs, improving gas replenishment efficiency, avoiding safety and operational risks caused by manual venting, and ensuring the reliability of the equipment and the safe and stable operation of the power grid.
[0054] Example 2:
[0055] An application method for a multi-chamber SF6 automatic gas replenishment system for GIS equipment based on deep learning includes the following steps:
[0056] (1) Prepare a gas cylinder with sufficient and qualified gas, connect it firmly to the pipeline, and make sure to secure the transport trolley to prevent it from tipping over;
[0057] (2) Install gas pipeline system and valves, and at the same time install pressure sensor, temperature sensor and humidity sensor at each gas chamber and gas cylinder to collect pressure, temperature and humidity data in real time;
[0058] (3) In actual operation, when there is an abnormal power outage or data acquisition equipment failure, resulting in missing or abnormal data in the original data, the acquired data will affect the establishment of the prediction model. It is necessary to preprocess the acquired data and use the Kalman filter algorithm to dynamically weight the temperature, humidity and pressure data, correct the error of the gas state equation, and improve the reliability of the data.
[0059] (4) Predict the future SF6 leakage rate trend of each gas chamber based on deep learning and trigger gas replenishment in advance: The deep learning model LSTM-ATTENTION is used to capture the time series dependency relationship by using a 4-layer LSTM to address the problem that the traditional PID algorithm cannot handle nonlinear leakage. The model has 256 hidden units and the attention mechanism dynamically weights key features, including the sudden pressure difference signal. The input dimensions include four-dimensional data of temperature, humidity, pressure and gas flow rate within a 60-second window, and the output is the leakage rate g / s for the next 5 minutes.
[0060] (5) Before replenishing gas, exhaust the pipeline to prevent dust and water vapor accumulated in the pipeline from entering the gas chamber, which may cause short circuit faults in the GIS during operation and cause the internal gas to deteriorate.
[0061] (6) Adjust the status of each valve dynamically according to the predicted leakage rate to achieve dynamic matching between the gas supply rate and the leakage rate, and avoid equipment damage caused by sudden pressure changes.
[0062] (7) After the gas replenishment is completed, the storage unit records data, which mainly includes the gas cylinder number, the most recently recorded gas cylinder weight, the pressure value of each gas chamber, and the most recent gas replenishment time and replenishment amount.
[0063] The above technical features constitute the embodiments of the present invention, which have strong adaptability and implementation effect. Unnecessary technical features can be added or removed according to actual needs to meet the needs of different situations.
Claims
1. A deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment, characterized in that... include: Gas pipeline module, valve assembly, intelligent monitoring module, control module, human-machine interface; The gas pipeline module includes a weighing sensor, a multi-way valve and its connecting pipeline. The weighing sensor is installed on the pallet of the gas cylinder transport trolley and can measure the weight of the gas cylinder throughout the gas replenishment process. The multi-way valve and its connecting pipeline can be used to replenish gas from one gas cylinder to multiple gas chambers in the same interval. The valve assembly includes multiple solenoid valves, double check valves, and exhaust valves; the solenoid valves and double check valves are installed at the outlets of multiple SF6 gas chamber connectors, and the exhaust valve is located at the exhaust port of the multi-way valve. The intelligent monitoring module includes a pressure sensor, a temperature sensor, a humidity sensor, and a transmitter, which are used to collect data and send it to the control module. The control module receives data from the intelligent monitoring module and controls the solenoid valve. Simultaneously, it records data on time, pressure, temperature, humidity, and number of gas replenishments through the storage unit. It utilizes the NVIDIA Jetson chip for edge computing and big data analysis to accurately output the rate at which the gas chamber needs to be replenished and predict the time of the next gas replenishment, triggering early gas replenishment to avoid sudden pressure changes in the gas chamber and ensure equipment safety. The human-machine interface uses an Android tablet or mobile phone to exchange data with the main control chip component, read the fluid flow direction and perform flow rate control, and provides human-machine interaction display, working status display, parameter setting and operation history.
2. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 1, characterized in that... The control module uses an STM32H7 chip to collect data from pressure, temperature, and humidity sensors, and controls the solenoid valves. The data is then uploaded to a mobile app via Bluetooth for recording.
3. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 2, characterized in that... The control module imports the LSTM-ATTENTION model package into the NVIDIA Jetson chip. The LSTM-ATTENTION model input dimensions include four-dimensional data: temperature, humidity, pressure, and gas flow rate within a 60-second window, and outputs the leakage rate for the next 5 minutes. Specifically, after obtaining the input pressure, temperature, and humidity data, the LSTM uses its hidden layers to calculate intermediate data states, and then employs a function... Solving for the eigenvector h t,j With hidden state h t Similarity. The calculation method is as follows: In the formula: W s b is the weight matrix of the fully connected layer. s The bias vector value; For the hidden state h t Transpose; for h i,j with h t The correlation between the hidden layer vector attention weights α and the attention weights is typically calculated using the Softmax function. i After interacting with h i,j The output of the attention layer can be obtained by weighted summation. α i , The calculation methods are as follows: In the formula: t is the output node of the fully connected layer, and τ is the total length of the sequence.
4. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 1, 2, or 3, characterized in that: The control module preprocesses the collected data and uses a Kalman filter algorithm to dynamically weight the temperature, humidity, and pressure data, correcting errors in the gas state equation and improving data reliability.
5. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 1, 2, or 3, characterized in that: The human-machine interface displays information related to the gas cylinders, including the cylinder number, the most recently recorded cylinder weight, the pressure value of each chamber, the most recent gas replenishment time and amount. A yellow alarm is triggered when the cylinder weight is less than 60kg, and a red alarm is triggered when it is less than 50kg. The interface also displays the real-time parameters and historical data of the corresponding cylinders in detail.
6. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 4, characterized in that: The human-machine interface displays information related to the gas cylinders, including the cylinder number, the most recently recorded cylinder weight, the pressure value of each chamber, the most recent gas replenishment time and amount. A yellow alarm is triggered when the cylinder weight is less than 60kg, and a red alarm is triggered when it is less than 50kg. The interface also displays the real-time parameters and historical data of the corresponding cylinders in detail.
7. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 1, 2, or 3, characterized in that... The weighing sensor is a cantilever beam type weighing sensor.
8. The deep learning-based SF6 automatic gas replenishment system for multi-chamber GIS equipment according to claim 6, characterized in that... The weighing sensor is a cantilever beam type weighing sensor.
9. An application method for a multi-chamber SF6 automatic gas replenishment system for GIS equipment based on deep learning, characterized in that... Includes the following steps: (1) Prepare a gas cylinder with sufficient and qualified gas, connect it firmly to the pipeline, and make sure to secure the transport trolley to prevent it from tipping over; (2) Install gas pipeline system and valves, and at the same time install pressure sensor, temperature sensor and humidity sensor at each gas chamber and gas cylinder to collect pressure, temperature and humidity data in real time; (3) In actual operation, when there is an abnormal power outage or data acquisition equipment failure, resulting in missing or abnormal data in the original data, the acquired data will affect the establishment of the prediction model. It is necessary to preprocess the acquired data and use the Kalman filter algorithm to dynamically weight the temperature, humidity and pressure data, correct the error of the gas state equation, and improve the reliability of the data. (4) Predict the future SF6 leakage rate trend of each gas chamber based on deep learning and trigger gas replenishment in advance: The deep learning model LSTM-ATTENTION is used to capture the time series dependency relationship by using a 4-layer LSTM to address the problem that the traditional PID algorithm cannot handle nonlinear leakage. The model has 256 hidden units and the attention mechanism dynamically weights key features, including the sudden pressure difference signal. The input dimensions include four-dimensional data of temperature, humidity, pressure and gas flow rate within a 60-second window, and the output is the leakage rate g / s for the next 5 minutes. (5) Before replenishing gas, exhaust the pipeline to prevent dust and water vapor accumulated in the pipeline from entering the gas chamber, which may cause short circuit faults in the GIS during operation and cause the internal gas to deteriorate. (6) Adjust the status of each valve dynamically according to the predicted leakage rate to achieve dynamic matching between the gas supply rate and the leakage rate, and avoid equipment damage caused by sudden pressure changes. (7) After the gas replenishment is completed, the storage unit records data, which mainly includes the gas cylinder number, the most recently recorded gas cylinder weight, the pressure value of each gas chamber, and the most recent gas replenishment time and replenishment amount.