Disassembly-free intelligent maintenance method and device for transformer substation respirator

Abstract

The invention discloses a transformer substation respirator disassembly-free intelligent maintenance method and device, and belongs to the technical field of transformer substation maintenance, and the transformer substation respirator disassembly-free intelligent maintenance device comprises a dynamic sensing system, a maintenance execution mechanism and an intelligent decision center which are interconnected through an industrial bus; the dynamic sensing system comprises an optical detection unit and an environment sensing unit; the optical detection unit adopts a multispectral imaging module for scanning a silica gel tank body in real time, and recognizes the proportion of a silica gel color-changing area through a convolutional neural network algorithm; the environment sensing unit is provided with a miniature temperature and humidity sensor, an SF6 concentration sensor and an oil cup liquid level sensor in an integrated mode and used for monitoring gas components and oil liquid states in the respirator in real time. Multi-dimensional state monitoring, all-parameter coverage and integrated control are achieved through optical and environment sensing, the problem that a traditional system is discrete is solved, the color change rate is quantized through multi-spectral imaging, multi-parameter synchronous collection is achieved, and the detection precision is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of substation maintenance, and particularly relates to a substation breather intelligent maintenance method and device without disassembly. BACKGROUND

[0002] The breather of a transformer has two functions: one is to make the air in the transformer body communicate with the outside air through the breather, so that the air pressure inside and outside the transformer body is equal. In high-temperature weather or full-load operation, the transformer oil expands due to temperature rise, the internal pressure rises and cannot be released, causing the transformer to spray oil. The second is to absorb the moisture in the air entering the transformer through the desiccant in the breather, so that the insulating oil in the transformer maintains good electrical performance, prevents humid air from directly entering the transformer oil pillow, and prevents the transformer oil from being damp, reducing or damaging the insulation strength of the transformer, so that the winding is burned. The maintenance of the substation breather mainly relies on artificial periodic inspection and operation, which specifically includes: artificial inspection mechanism, through fixed cycle (usually once a month) on-site inspection of the degree of silicone discoloration, relying on visible light visual judgment of the silicone state; single threshold trigger maintenance, when the humidity sensor detects that the relative humidity inside the breather is >80%, the silicone replacement or regeneration operation is triggered; high-temperature regeneration process, constant temperature heating (>=80℃) is used to regenerate the silicone, which takes about 4-6 hours; discrete system architecture, the monitoring system, the regeneration device and the oil injection module are independently operated, and data interaction relies on manual recording and transmission. The defects of the prior art are: low detection accuracy, single parameter, visible light visual inspection cannot quantify the degree of silicone discoloration, and is significantly affected by the light environment, which is easy to cause misjudgment, the single humidity threshold mechanism does not fuse temperature, SF6 concentration and other parameters, the false positive rate is high in high-humidity areas, the artificial inspection cycle is long, and the small leakage hidden danger in the early stage of silicone discoloration cannot be captured. SUMMARY

[0003] The purpose of the present application is to overcome the shortcomings of the prior art, and to provide a substation breather intelligent maintenance method and device without disassembly, which solves the problems in the background art.

[0004] The purpose of the present application is achieved as follows:

[0005] A substation breather intelligent maintenance method without disassembly, comprising: state perception, calculating the silicone discoloration rate in real time through multispectral imaging, and synchronously collecting the temperature, humidity, SF6 concentration and oil cup liquid level inside the breather; dynamically deciding to build an environmental state matrix; regenerating the silicone at 50±2℃, synchronously starting vacuum pumping, and the mechanical arm disassembling the silicone tank; and performing closed-loop feedback.

[0006] The implementation process calculates the discoloration rate of silica gel in real time through multispectral imaging (wavelength 380-2500 nm), synchronously collects the temperature, humidity, SF6 concentration and oil level in the respirator, collects the surface image of the silica gel tank in real time, synchronously reads the humidity, temperature and SF6 concentration sensor data, triggers the mechanical arm disassembly instruction when the discoloration rate of silica gel is greater than the dynamic threshold (combined with temperature and humidity / automatic correction of load current), the mechanical arm transfers the silica gel tank to the regeneration chamber, the vacuum pump removes SF6 to a concentration less than 300 ppm, the far infrared heater maintains 50±2℃ for 100±5 minutes, the prediction model is established based on the oil level sensor data to control the oil supplementing accuracy of the electromagnetic valve, the data is uploaded to the operation and maintenance platform through 5G communication, and the gas protection system is linked in real time. Realize multispectral imaging quantitative discoloration rate (error ≤3%) + multi-parameter synchronous collection, improve detection accuracy, solve the problem of silica gel carbonization caused by high temperature regeneration through 50±2℃ low temperature regeneration process, realize quantitative silica gel discoloration degree to avoid missed judgment, and reduce the false alarm rate in high humidity areas by combining temperature, SF6 concentration and other parameters.

[0007] The step of dynamic decision-making includes:

[0008] The weight is dynamically improved to w H = w H0 + 0.15·(H-80%), in the formula, w H is the humidity weight after dynamic adjustment, w H0 is the initial value of the humidity weight, and H is the real-time relative humidity of the environment.

[0009] The implementation process constructs a four-dimensional weight matrix, assigns values according to climate types, continuously optimizes the weight coefficients through a machine learning model (LSTM network), updates the decision matrix every 24 hours, further reduces the false alarm rate from >60% to <5%, eliminates false alarms in heavy rain weather in high humidity areas, and has strong adaptability: supports full climate environment from -40℃ to 60℃. The higher the humidity, the greater the weight. The influence of the humidity parameter in the decision matrix Q is enhanced, which is more likely to trigger maintenance, adapts to high humidity areas, and avoids missed detection in special working conditions such as heavy rain.

[0010] The safety interlocking step automatically detects the gas protection state before maintenance, generates an alarm signal and suspends the operation if it has not been exited, the implementation process controls the temperature in three zones by far infrared heater (upper zone 55℃, middle zone 50℃, lower zone 45℃), initial discoloration rate 10%-30%: regeneration 80±3 minutes, initial discoloration rate 30%-50%: regeneration 100±5 minutes, vacuum synchronization, and the vacuum degree is maintained at -0.08MPa throughout the regeneration process, which saves 37% energy compared with traditional 80℃ high temperature regeneration, greatly reducing energy consumption. Before maintenance, the gas protection state is automatically detected, an alarm signal is generated and the operation is suspended if the gas protection has not been exited, and the regeneration process is started if the gas protection has been exited, and SF6 is synchronously pumped to <300ppm, effectively eliminating the risk of gas misoperation during maintenance.

[0011] The substation breather disassembly-free intelligent maintenance device is used for realizing the method of any one of claims 1-3, and comprises a dynamic sensing system, a maintenance execution mechanism and an intelligent decision center, which are interconnected through an industrial bus; the dynamic sensing system comprises an optical detection unit and an environment sensing unit; the optical detection unit adopts a multi-spectrum imaging module for real-time scanning of a silica gel tank body, and identifies a proportion of a silica gel discoloration area through a convolutional neural network algorithm; the environment sensing unit is integrally provided with a micro temperature and humidity sensor, an SF6 concentration sensor and an oil cup liquid level sensor, and is used for real-time monitoring of internal gas components and oil liquid states of the breather. The implementation process is: sensing data, generation of instructions by the decision center, action of the execution mechanism, 5G feedback to the platform, multi-dimensional state monitoring by the optical and environment sensing, full-parameter coverage, integrated control, and solution of the traditional system separation problem.

[0012] The maintenance execution mechanism comprises a dual-mode regeneration device and a closed-loop oil injection system; the dual-mode regeneration device comprises a silica gel regeneration module and a replacement module, the regeneration module regenerates silica gel at a low temperature of 45-55 DEG C through a uniform heating assembly, and the replacement module adopts a six-degree-of-freedom mechanical arm to quickly disassemble and assemble the silica gel tank; the closed-loop oil injection system comprises a precision metering pump, a high-precision liquid level controller and a self-adaptive reversing valve, and adjusts the oil injection amount in real time based on an oil cup liquid level prediction model,

[0013] The oil cup liquid level prediction model is,

[0014] Delta L = alpha * T + beta * P

[0015] Wherein, Delta L is the liquid level change, alpha is the transformer load parameter, T is the real-time load of the transformer, beta is the oil evaporation coefficient, and P is the environmental temperature change.

[0016] During work, the low-temperature uniform heating (45-55 DEG C) of the regeneration module, the function control of the six-degree-of-freedom mechanical arm clamping force of the replacement module, the closed-loop oil injection system, the model adjustment of the oil injection amount, the switching of the self-cleaning module to the ethanol flushing oil way after oil injection, the model control liquid level precision, and the realization of zero error of oil injection.

[0017] The operations performed by the intelligent decision center include: generating a state evaluation matrix through the data of the dynamic sensing system; outputting a maintenance instruction through a pre-trained decision tree model: starting the regeneration module when theta belongs to (0.5, 0.7) and Q * w is greater than or equal to 0.6; triggering the replacement module when theta is greater than or equal to 0.7 or Q * w is less than 0.6; wherein, theta is the real-time discoloration rate of silica gel, Q is the environment state evaluation matrix, and w is the environment weight vector; and the input layer of the pre-trained decision tree model introduces an attention mechanism and synchronously activates a warning unit; sending an encrypted work order to the operation and maintenance platform through a 5G communication module.

[0018] The implementation process generates a state evaluation matrix, 5G sends an encrypted work order, pre-trains a decision tree, introduces an attention mechanism, performs real-time calculation, links to an execution mechanism, intelligently switches modes to avoid over-regeneration, and prolongs the service life of the silica gel.

[0019] The multispectral imaging module is provided with a polarization filter for enhancing the image signal-to-noise ratio in a strong electromagnetic interference environment; the convolutional neural network adopts a transfer learning framework to pre-train a data set, which contains two thousand groups of silica gel discoloration state samples; the uniform heating assembly includes a nanometer carbon fiber heating layer for adhering to the outer wall of the silica gel tank and a PID temperature controller for maintaining a temperature curve with an accuracy of 0.1℃; and the regeneration process starts a vacuum pump simultaneously to extract volatile gases.

[0020] During implementation, the polarization filter suppresses electromagnetic interference, the CNN transfer learning trains and identifies 2000 samples in the discoloration area, the nanometer carbon fiber heating layer of the regeneration module adheres to the tank, the PID temperature control accuracy, and the vacuum pump is started simultaneously. The polarization filter improves the image signal-to-noise ratio and the CNN recognition accuracy, the partition PID temperature control maintains a constant temperature of 50±2℃, and the silica gel activity is preserved. Through polarization filtering and CNN, the recognition accuracy far exceeds visual detection, and the detection error is reduced.

[0021] The end of the six-degree-of-freedom mechanical arm is provided with a flexible gripper and a torque sensor, and the clamping force dynamically satisfies:

[0022]

[0023] Wherein, F clamp is the clamping force, e -0.2t is an exponential decay function, and N is a unit Newton;

[0024] The closed-loop oil injection system is internally provided with a self-cleaning module, which is used to switch the oil circuit to the ethanol storage tank after performing the oil injection operation, and flush the oil circuit.

[0025] The working principle is that the force control function, high clamping force in the initial disassembly and assembly stage to prevent falling off, exponential decay in the later stage to prevent fracturing, ethanol flushing, and elimination of cross-contamination of the oil circuit. The force control model eliminates the risk of artificial disassembly and assembly cracks, reduces the failure rate, avoids oil quality degradation through self-cleaning, and prolongs the service life of the oil circuit.

[0026] The dynamic perception system includes a control unit execution module that collects the discoloration rate of the silica tank based on an optical sensor,

[0027] When the trigger condition is met, maintenance is started, where S c is the area of the discoloration region, S t is the total area, k t is the temperature compensation coefficient, and k his the humidity supplement coefficient; the relative humidity H inside the respirator is monitored by a micro humidity sensor r , when H r When ≥80%, the maintenance mechanism is activated synchronously;

[0028] The optical sensor is a multispectral imager with an operating wavelength range of 850-1550nm. The error formula for detecting the color change rate of silica gel is:

[0029]

[0030] Where δ is the detection error, S c-real is the actual color change area, S c-det The device detects discoloration area, and 3% is the upper limit of the allowable error.

[0031] Implement process start maintenance conditions, S i is the color change area, S t is the total area, k T is the temperature coefficient, k H is the humidity coefficient; multispectral imaging (850-1550nm), quantification The threshold is compensated according to the environmental coefficient, and the threshold is dynamically corrected by the formula, so that high humidity / low temperature environments are not missed, and adaptive triggering is achieved.

[0032] The dynamic perception system also includes an intelligent early warning module, which transmits encrypted data frames via the 5G protocol.

[0033] The encrypted data frame is: Frame=[Header|ID|θ|H|||CRC],

[0034] Among them, the CRC check code is generated by the polynomial G(x)=x 8 +x 2 +1 calculation, where Header is the frame header, θ is the color change rate, H is the humidity value, G(x) is the generating polynomial, and x is the variable symbol;

[0035] The fault diagnosis model of the intelligent early warning module is as follows: input vector X = [θ, T, H, C]T; decision output y = sign(ω·X+b); where θ is the discoloration rate of silica gel, T is the ambient temperature, H is the relative humidity, C is the SF6 gas concentration, ω is the SVM weight vector, and b is the decision hyperplane bias term.

[0036] During the implementation process, from multi-parameters to SVM classification, to fault code output, and finally 5G encrypted transmission and platform alarm, CRC verification ensures data integrity and achieves zero tampering in transmission.

[0037] The beneficial effects of the present application: realize multispectral imaging quantitative discoloration rate (error ≤3%) + multi-parameter synchronous acquisition, improve detection accuracy, solve the problem of carbonization of silica gel caused by high temperature regeneration through 50±2℃ low temperature regeneration process, realize quantitative silica gel discoloration degree to avoid missed judgment, and reduce the false positive rate in high humidity areas by fusing temperature, SF6 concentration and other parameters. Through machine learning model (LSTM network), the weight coefficient is continuously optimized, the decision matrix is updated every 24 hours, and the false positive rate is further reduced from >60% to <5%, such as eliminating false positives in high humidity areas during heavy rain weather, and strong adaptability: supporting full climate environment from-40℃ to 60℃. The higher the humidity, the greater the weight. The influence of the humidity parameter in the decision matrix Q is enhanced, which is more likely to trigger maintenance, and is suitable for high humidity areas to avoid missed detection in special working conditions such as heavy rain. The gas protection state is automatically detected before maintenance, and if the gas protection has not exited, an alarm signal is generated and paused; if it has exited, the regeneration process is started, and vacuum pumping is started to SF6<300ppm, which effectively eliminates the risk of gas misoperation during maintenance. The low temperature uniform heating (45-55℃) of the regeneration module, the six-degree-of-freedom mechanical arm clamping force of the replacement module controlled according to the function, the closed-loop oil injection system, the oil injection amount adjusted according to the model, the ethanol flushing oil way switched after the self-cleaning module injects oil, the model controls the liquid level accuracy, and realizes zero error of oil injection. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a schematic diagram of the device of the present application. DETAILED DESCRIPTION

[0039] The present application will be further described in detail below with reference to the accompanying drawings, it should be pointed out that it is only for more clearly illustrating and explaining the present application.

[0040] Example 1

[0041] As shown in Figure 1 , the present embodiment discloses a transformer substation respirator disassembly-free intelligent maintenance method, which comprises: state sensing, real-time calculation of silica gel discoloration rate through multispectral imaging, and synchronous acquisition of internal temperature, humidity, SF6 concentration and oil cup liquid level of the respirator; dynamic decision-making to construct an environmental state matrix; regenerating silica gel at 50±2℃, synchronously starting vacuum pumping, and mechanical arm disassembling silica gel tank; and performing closed-loop feedback.

[0042] The implementation process calculates the discoloration rate of silica gel in real time through multispectral imaging (wavelength 380-2500 nm), synchronously collects the temperature, humidity, SF6 concentration and oil level in the respirator, collects the surface image of the silica gel tank in real time, synchronously reads the humidity, temperature and SF6 concentration sensor data, triggers the mechanical arm disassembly instruction when the discoloration rate of silica gel is greater than the dynamic threshold (combined with temperature and humidity / automatic correction of load current), the mechanical arm transfers the silica gel tank to the regeneration chamber, the vacuum pump removes SF6 to a concentration less than 300 ppm, the far infrared heater maintains 50±2℃ for 100±5 minutes, the prediction model is established based on the oil level sensor data to control the oil supplementing accuracy of the electromagnetic valve, the data is uploaded to the operation and maintenance platform through 5G communication, and the gas protection system is linked in real time. Realize multispectral imaging quantitative discoloration rate (error ≤3%) + multi-parameter synchronous collection, improve detection accuracy, solve the problem of silica gel carbonization caused by high temperature regeneration through 50±2℃ low temperature regeneration process, realize quantitative silica gel discoloration degree to avoid missed judgment, and reduce the false alarm rate in high humidity areas by combining temperature, SF6 concentration and other parameters.

[0043] The step of dynamic decision-making includes:

[0044] The weight is dynamically improved to w H = w H0 + 0.15·(H-80%), in the formula, w H is the humidity weight after dynamic adjustment, w H0 is the initial value of the humidity weight, and H is the real-time relative humidity of the environment.

[0045] The implementation process constructs a four-dimensional weight matrix, assigns values according to climate types, continuously optimizes the weight coefficients through a machine learning model (LSTM network), updates the decision matrix every 24 hours, further reduces the false alarm rate from >60% to <5%, eliminates false alarms in heavy rain weather in high humidity areas, and is highly adaptable: supports full climate environment from -40℃ to 60℃. The higher the humidity, the greater the weight. The influence of the humidity parameter in the decision matrix Q is enhanced, which is more likely to trigger maintenance, adapts to high humidity areas, and avoids missed detection in special working conditions such as heavy rain.

[0046] The safety interlocking step automatically detects the gas protection state before maintenance, generates an alarm signal and suspends the operation if it has not been exited, the implementation process controls the temperature in three zones by far infrared heater (upper zone 55℃, middle zone 50℃, lower zone 45℃), initial discoloration rate 10%-30%: regeneration 80±3 minutes, initial discoloration rate 30%-50%: regeneration 100±5 minutes, vacuum synchronization, and the vacuum degree is maintained at -0.08MPa throughout the regeneration process, which saves 37% energy compared with traditional 80℃ high temperature regeneration, greatly reducing energy consumption. Before maintenance, the gas protection state is automatically detected, an alarm signal is generated and the operation is suspended if the gas protection has not been exited, and the regeneration process is started if the gas protection has been exited, and SF6 is synchronously pumped to <300ppm, effectively eliminating the risk of gas misoperation during maintenance.

[0047] A disassembly-free intelligent maintenance device for substation respirators, used to implement the method of any one of claims 1-3, comprises: a dynamic sensing system, a maintenance actuator, and an intelligent decision-making center, all interconnected via an industrial bus. The dynamic sensing system includes an optical detection unit and an environmental sensing unit. The optical detection unit uses a multispectral imaging module to scan the silicone tank in real time and identifies the percentage of discolored silicone areas using a convolutional neural network algorithm. The environmental sensing unit integrates a micro temperature and humidity sensor, an SF6 concentration sensor, and an oil cup level sensor for real-time monitoring of the respirator's internal gas composition and oil status. The implementation process involves sensing data, the decision center generating instructions, the actuator operating, and 5G feedback to the platform. Optical and environmental sensing enable multi-dimensional status monitoring, full parameter coverage, and integrated control, resolving the issues of traditional discrete systems.

[0048] The maintenance actuator includes a dual-mode regeneration device and a closed-loop oil injection system; the dual-mode regeneration device includes a silicone regeneration module and a replacement module. The regeneration module regenerates silicone at a low temperature of 45-55°C by uniformly heating the components, and the replacement module uses a six-degree-of-freedom robotic arm to quickly disassemble and assemble the silicone tank; the closed-loop oil injection system includes a precision metering pump, a high-precision liquid level controller and an adaptive reversing valve, which adjusts the oil injection amount in real time based on the oil cup liquid level prediction model.

[0049] The oil cup level prediction model is:

[0050] ΔL=α·T+β·P

[0051] Among them, ΔL is the change in liquid level, α is the transformer load parameter, T is the real-time load of the transformer, β is the oil volatility coefficient, and P is the change in ambient temperature.

[0052] During operation, the regeneration module is uniformly heated at low temperature (45-55°C), the clamping force of the six-degree-of-freedom robotic arm that replaces the module is controlled by a function, the closed-loop oil filling system, the oil filling amount is adjusted according to the model, and the self-cleaning module switches to ethanol flushing oil circuit after oiling. The model controls the liquid level accuracy to achieve zero error in oil filling.

[0053] The operations performed by the intelligent decision center include: generating a status assessment matrix through data from the dynamic perception system; outputting maintenance instructions through a pre-trained decision tree model: starting the regeneration module when θ∈(0.5,0.7) and Q·w≥0.6; triggering module replacement when θ≥0.7 or Q·w<0.6; where θ is the real-time color change rate of silicone, Q is the environmental status assessment matrix, and w is the environmental weight vector; and introducing an attention mechanism into the input layer of the pre-trained decision tree model to synchronously activate the early warning unit; and sending an encrypted work order to the operation and maintenance platform via the 5G communication module.

[0054] The implementation process generates a state evaluation matrix, 5G sends encrypted work orders, pre-trains decision trees, introduces an attention mechanism, performs real-time calculations, and links to the execution mechanism to avoid over-regeneration and extend the life of the silicone.

[0055] The multispectral imaging module is provided with a polarization filter to enhance the image signal-to-noise ratio in a strong electromagnetic interference environment; the convolutional neural network uses a transfer learning framework to pre-train a data set containing two thousand sets of silicone discoloration state samples; the uniform heating component includes a nanocarbon fiber heating layer for adhering to the outer wall of the silicone tank; a PID temperature controller is used to maintain a temperature curve with an accuracy of 0.1℃; and a vacuum pump is started simultaneously to extract volatile gases during the regeneration process.

[0056] During implementation, the polarization filter suppresses electromagnetic interference, the CNN transfer learning uses 2000 samples to train and identify discolored areas, the nanocarbon fiber heating layer of the regeneration module adheres to the tank, the PID temperature control accuracy, and the vacuum pump is started simultaneously. Polarization filtering, improved image signal-to-noise ratio, and improved CNN recognition accuracy, partitioned PID temperature control, maintained a constant temperature of 50±2℃, and preserved the activity of the silicone. Through polarization filtering and CNN, the recognition accuracy far exceeds visual detection, and the detection error is reduced.

[0057] The end of the six-degree-of-freedom mechanical arm is provided with a flexible gripper and a torque sensor, and the clamping force dynamically satisfies:

[0058]

[0059] Wherein, F clamp is the clamping force, e -0.2t is an exponential decay function, and N is a unit Newton;

[0060] The closed-loop oil injection system is internally provided with a self-cleaning module, which is used to switch the oil circuit to the ethanol storage tank after performing the oil injection operation, and flush the oil circuit.

[0061] The working principle is that the force control function, high clamping force in the initial disassembly and assembly stage to prevent falling off, exponential decay in the later stage to prevent fracturing, ethanol flushing, and elimination of cross-contamination of the oil circuit. The force control model eliminates the risk of artificial disassembly and assembly cracks, reduces the failure rate, self-cleaning avoids oil quality degradation, and prolongs the service life of the oil circuit.

[0062] Example 2

[0063] As Figure 1 shown, the embodiment discloses a transformer breather free disassembly and assembly intelligent maintenance method, which includes: state perception, real-time calculation of silicone discoloration rate through multispectral imaging, and synchronous collection of internal temperature, humidity, SF6 concentration, and oil cup liquid level of the breather; dynamic decision-making to build an environmental state matrix; regenerating silicone at 50±2℃, starting a vacuum air pump simultaneously, and a mechanical arm disassembling and assembling the silicone tank; and performing closed-loop feedback.

[0064] The implementation process calculates the discoloration rate of silica gel in real time through multispectral imaging (wavelength 380-2500 nm), synchronously collects the temperature, humidity, SF6 concentration and oil cup liquid level inside the respirator, collects the surface image of the silica gel tank in real time, synchronously reads the humidity, temperature and SF6 concentration sensor data, when the discoloration rate of silica gel is greater than the dynamic threshold (combined with temperature and humidity / automatic correction of load current), triggers the mechanical arm disassembly instruction, the mechanical arm transfers the silica gel tank to the regeneration chamber, the vacuum pump removes SF6 to a concentration less than 300 ppm, the far infrared heater maintains 50±2℃ for 100±5 minutes, the prediction model is established based on the oil level sensor data to control the oil supplementing precision of the electromagnetic valve, the data is uploaded to the operation and maintenance platform through 5G communication, and the gas protection system is linked in real time. Realize multispectral imaging quantitative discoloration rate (error ≤3%) + multi-parameter synchronous collection, improve detection accuracy, solve the problem of silica gel carbonization caused by high temperature regeneration through 50±2℃ low temperature regeneration process, realize quantitative silica gel discoloring degree to avoid missed judgment, and reduce the false alarm rate in high humidity areas by combining temperature, SF6 concentration and other parameters.

[0065] The step of dynamic decision-making includes:

[0066] The weight is dynamically improved to w H = w H0 + 0.15·(H-80%), in the formula, w H is the humidity weight after dynamic adjustment, w H0 is the initial value of the humidity weight, and H is the real-time relative humidity of the environment.

[0067] The implementation process constructs a four-dimensional weight matrix, assigns values according to climate types, continuously optimizes the weight coefficients through a machine learning model (LSTM network), updates the decision matrix every 24 hours, further reduces the false alarm rate from >60% to <5%, eliminates false alarms in heavy rain weather in high humidity areas, and is highly adaptable: supports full climate environment from -40℃ to 60℃. The higher the humidity, the greater the weight. The influence of the humidity parameter in the decision matrix Q is enhanced, which is more likely to trigger maintenance, is suitable for high humidity areas, and avoids missed detection in special working conditions such as heavy rain.

[0068] Safety interlocking step, automatic detection of gas protection state before maintenance, if not exit, generate alarm signal and suspend operation, implement process far infrared heater three zone temperature control (upper zone 55℃, middle zone 50℃, lower zone 45℃), initial color change rate 10%-30%: regeneration 80±3 minutes, initial color change rate 30%-50%: regeneration 100±5 minutes, vacuum synchronization, maintain vacuum degree-0.08MPa during regeneration, 37% energy saving than traditional 80℃ high temperature regeneration, greatly reduce energy consumption. Automatic detection of gas protection state before maintenance, if gas protection has not exited, generate alarm signal and suspend; if has exited, start regeneration process, synchronize vacuum pumping to SF6<300ppm, effectively eliminate the risk of gas misoperation during maintenance.

[0069] A substation breather disassembly-free intelligent maintenance device for realizing the method of any one of claims 1-3, comprising a dynamic sensing system, a maintenance execution mechanism and an intelligent decision center, which are interconnected through an industrial bus; the dynamic sensing system comprises an optical detection unit and an environmental sensing unit; the optical detection unit adopts a multispectral imaging module for real-time scanning of the silica gel tank, and identifies the proportion of the silica gel color change area through a convolutional neural network algorithm; the environmental sensing unit is integrally provided with a micro temperature and humidity sensor, an SF6 concentration sensor and an oil cup liquid level sensor for real-time monitoring of the internal gas composition and oil state of the breather. The implementation process is: sensing data, decision center generated instructions, execution mechanism action, 5G feedback to the platform, optical and environmental sensing realize multi-dimensional state monitoring, full parameter coverage, integrated control, solve the problem of traditional system separation.

[0070] The maintenance execution mechanism comprises a dual-mode regeneration device and a closed-loop oil injection system; the dual-mode regeneration device comprises a silica gel regeneration module and a replacement module, the regeneration module regenerates silica gel at a low temperature of 45-55℃ through a uniform heating assembly, and the replacement module adopts a six-degree-of-freedom mechanical arm to quickly disassemble and assemble the silica gel tank; the closed-loop oil injection system comprises a precision metering pump, a high-precision liquid level controller and a self-adaptive reversing valve, and adjusts the oil injection amount in real time based on an oil cup liquid level prediction model,

[0071] The oil cup liquid level prediction model is,

[0072] ΔL=α·T+β·P

[0073] Wherein, ΔL is the liquid level change, α is the transformer load parameter, T is the real-time load of the transformer, β is the oil evaporation coefficient, and P is the environmental temperature change.

[0074] During work, the low-temperature uniform heating (45-55℃) of the regeneration module, the six-degree-of-freedom mechanical arm clamping force of the replacement module controlled according to a function, the closed-loop oil injection system, the oil injection amount adjusted according to the model, the self-cleaning module switching the ethanol flushing oil way after oil injection, the model controlling the liquid level precision, and realizing zero error of oil injection.

[0075] The operation performed by the intelligent decision center includes: generating a state evaluation matrix through the dynamic sensing system; outputting maintenance instructions through a pre-trained decision tree model: starting a regeneration module when θ∈(0.5, 0.7) and Q·w≥0.6; triggering a replacement module when θ≥0.7 or Q·w<0.6; wherein θ is the real-time color change rate of silica gel, Q is the environmental state evaluation matrix, and w is the environmental weight vector; and the input layer of the pre-trained decision tree model introduces an attention mechanism and synchronously activates a warning unit; and sending an encrypted work order to the operation and maintenance platform through a 5G communication module.

[0076] The implementation process generates a state evaluation matrix, sends an encrypted work order through 5G, pre-trains a decision tree, introduces an attention mechanism, calculates in real time, links to an executive mechanism, intelligently switches modes to avoid over-regeneration, and prolongs the service life of silica gel.

[0077] The multispectral imaging module is provided with a polarization filter for enhancing the image signal-to-noise ratio in a strong electromagnetic interference environment; the convolutional neural network adopts a transfer learning framework to pre-train a data set, which contains two thousand groups of silica gel color change state samples; the uniform heating assembly includes: a nanometer carbon fiber heating layer for adhering to the outer wall of the silica gel tank; a PID temperature controller for maintaining a temperature curve with an accuracy of 0.1℃; and a vacuum pump started synchronously during the regeneration process to extract volatile gases.

[0078] During the implementation process, the polarization filter suppresses electromagnetic interference, the CNN transfer learning trains and identifies 2000 samples to identify the color change area, the nanometer carbon fiber heating layer of the regeneration module adheres to the tank, the PID temperature control accuracy, and the vacuum pump is started synchronously. Polarization filtering, improving image signal-to-noise ratio, CNN recognition accuracy, partition PID temperature control, maintaining a constant temperature of 50±2℃, and retaining silica gel activity. Through polarization filtering and CNN, the recognition accuracy far exceeds visual detection, and the detection error is reduced.

[0079] The end of the six-degree-of-freedom mechanical arm is provided with a flexible gripper and a torque sensor, and the clamping force dynamically satisfies:

[0080]

[0081] Wherein, F clamp is the clamping force, e -0.2t is an exponential decay function, and N is a unit Newton;

[0082] The closed-loop oil injection system is internally provided with a self-cleaning module, which is used to switch the oil circuit to the ethanol storage tank after performing the oil injection operation, and flush the oil circuit.

[0083] The working principle is that the force control function, the high clamping force in the initial disassembly stage prevents falling off, the exponential decay in the later stage prevents fracturing, ethanol flushing, and elimination of oil cross contamination, the force control model eliminates the risk of artificial disassembly cracks, reduces the failure rate, self-cleaning avoids oil quality deterioration, and prolongs the service life of the oil circuit.

[0084] The dynamic sensing system comprises a control unit execution module, the control unit execution module acquires the discoloration rate of the silica tank based on an optical sensor,

[0085] When the trigger condition is met maintenance is started, in the formula, S c is the discoloration area, S t is the total area, k t is the temperature compensation coefficient, k h is the humidity compensation coefficient; the relative humidity H r inside the respirator is monitored by a micro humidity sensor, r when H

[0086] The optical sensor is a multispectral imager, and the working wavelength range is 850-1550nm, and the error formula for detecting the discoloration rate of the silica gel is:

[0087]

[0088] In the formula, δ is the detection error, S c-real is the actual discoloration area, S c-det is the device detection discoloration area, and 3% is the error upper limit.

[0089] The implementation process starts the maintenance condition, S i is the discoloration area, S t is the total area, k T is the temperature coefficient, k H is the humidity coefficient; multispectral imaging (850-1550nm), quantitative compensation threshold according to environmental coefficient, formula dynamic correction threshold, high humidity / low temperature environment without leakage, and self-adaptive triggering is realized.

[0090] The dynamic sensing system further comprises an intelligent early warning module, and the intelligent early warning module transmits an encrypted data frame through a 5G protocol,

[0091] The encrypted data frame is: Frame=[Header|ID|θ|H|||CRC],

[0092] wherein, the CRC check code is generated by a polynomial G(x)=x 8 +x 2+1 calculation, wherein Header is a frame header, θ is a color change rate, H is a humidity value, G(x) is a generating polynomial, and x is a variable symbol;

[0093] The fault diagnosis model of the intelligent early warning module is: input vector X = [θ, T, H, C]T; decision output y = sign(ω·X+b); wherein, θ is the color change rate of silica gel, T is the environmental temperature, H is the relative humidity, C is the SF6 gas concentration, ω is the SVM weight vector, and b is the decision hyperplane bias term.

[0094] In the implementation process, from multiple parameters to SVM classification, to output fault code, finally 5G encrypted transmission, platform alarm, CRC check guarantees data integrity, realizes zero tampering transmission.

[0095] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art in the technical range disclosed by the present application, according to the technical scheme and the concept of the present application, equivalent replacement or change, should be covered in the protection scope of the present application.

Claims

1. A disassembly-free intelligent maintenance method for a substation respirator, characterized in that: include: State perception: multispectral imaging is used to calculate the silicone discoloration rate in real time, and simultaneously collect the respirator's internal temperature, humidity, SF6 concentration, and oil cup level; Dynamic decision-making builds an environment state matrix; Regenerate the silica gel at 50±2℃, start vacuum extraction simultaneously, and the robotic arm disassembles and assembles the silica gel tank; Perform closed-loop feedback.

2. The method for intelligent maintenance of substation respirator without disassembly according to claim 1 is characterized in that: The steps of dynamic decision-making include: when the ambient humidity is greater than 80%, The weight is dynamically increased to w H =w H0 +0.15·(H-80%), Where w H is the humidity weight after dynamic adjustment, w H0 is the initial value of humidity weight, and H is the real-time relative humidity of the environment.

3. The method for intelligent maintenance of substation respirator without disassembly according to claim 1 is characterized in that: It also includes a safety interlock step, which automatically detects the gas protection status before performing maintenance. If it has not exited, an alarm signal will be generated and the operation will be suspended.

4. A substation respirator disassembly-free intelligent maintenance device for implementing the method of any one of claims 1 to 3, characterized in that: include: The dynamic perception system, maintenance execution agency and intelligent decision-making center are interconnected through the industrial bus; The dynamic perception system includes an optical detection unit and an environment sensing unit; The optical detection unit uses a multispectral imaging module to scan the silicone tank in real time and identifies the proportion of the silicone discoloration area through a convolutional neural network algorithm; The environmental sensing unit is integrated with a micro temperature and humidity sensor, an SF6 concentration sensor and an oil cup level sensor, which are used to monitor the gas composition and oil status inside the respirator in real time.

5. The substation respirator disassembly-free intelligent maintenance device according to claim 4 is characterized in that: The maintenance actuator includes a dual-mode regeneration device and a closed-loop oil injection system; The dual-mode regeneration device includes a silica gel regeneration module and a replacement module. The regeneration module regenerates silica gel at a low temperature of 45-55°C by uniformly heating the components. The replacement module uses a six-degree-of-freedom robotic arm to quickly disassemble and assemble the silica gel tank. The closed-loop oil injection system includes a precision metering pump, a high-precision liquid level controller and an adaptive reversing valve, which adjusts the oil injection amount in real time based on the oil cup liquid level prediction model. The oil cup level prediction model is: ΔL=α·T+β·P Among them, ΔL is the change in liquid level, α is the transformer load parameter, T is the real-time load of the transformer, β is the oil volatility coefficient, and P is the change in ambient temperature.

6. The substation respirator disassembly-free intelligent maintenance device according to claim 5 is characterized in that: The operations performed by the intelligent decision-making center include: Generate a state assessment matrix through data from the dynamic perception system; A pre-trained decision tree model outputs maintenance instructions: when θ∈(0.5,0.7) and Q·w≥0.6, the regeneration module is activated; when θ≥0.7 or Q·w<0.6, the module replacement is triggered. Here, θ is the real-time color change rate of the silicone, Q is the environmental state assessment matrix, and w is the environmental weight vector. The input layer of the pre-trained decision tree model introduces an attention mechanism to simultaneously activate the early warning unit. Send encrypted work orders to the operation and maintenance platform through the 5G communication module.

7. The substation respirator disassembly-free intelligent maintenance device according to claim 6 is characterized in that: The multispectral imaging module is equipped with a polarization filter to enhance the image signal-to-noise ratio in a strong electromagnetic interference environment. The convolutional neural network uses a transfer learning framework to pre-train a data set containing 2,000 sets of silicone color change state samples. The uniform heating assembly includes: a nano-carbon fiber heating layer for conforming to the outer wall distribution of the silicone tank; a PID temperature controller for maintaining the temperature curve with an accuracy of 0.1°C; During the regeneration process, the vacuum pump is started synchronously to extract volatile gases.

8. The substation respirator disassembly-free intelligent maintenance device according to claim 7 is characterized in that: The end of the six-degree-of-freedom robotic arm is provided with a flexible gripper and a torque sensor, and the clamping force dynamically satisfies: Among them, F clamp is the clamping force, e -0.2t is an exponential decay function, N is the unit Newton; The closed-loop oil filling system is internally provided with a self-cleaning module, which is used to switch the oil circuit to the ethanol storage tank after performing the oil filling operation and flush the oil circuit.

9. The substation respirator disassembly-free intelligent maintenance device according to claim 8 is characterized in that: The dynamic perception system includes a control unit execution module, which collects the discoloration rate of the silicon can based on the optical sensor. When the trigger condition is met Maintenance is started when c is the area of ​​the discoloration region, S t is the total area, k t is the temperature supplement coefficient, k h is the humidity supplement coefficient; Monitor the relative humidity H inside the respirator using a micro humidity sensor r , when H r When ≥80%, the maintenance mechanism is activated synchronously; The optical sensor is a multispectral imager with an operating wavelength range of 850-1550nm. The error formula for detecting the color change rate of silica gel is: Where δ is the detection error, S c-real is the actual color change area, S c-det The device detects discoloration area, and 3% is the upper limit of the allowable error.

10. The substation respirator non-disassembly intelligent maintenance device according to claim 9 is characterized in that: The dynamic perception system also includes an intelligent early warning module. The intelligent warning module transmits encrypted data frames via the 5G protocol. The encrypted data frame is: Frame=[Header|ID|θ|H|||CRC], Among them, the CRC check code is generated by the polynomial G(x)=x 8 +x 2 +1 calculation, where Header is the frame header, θ is the color change rate, H is the humidity value, G(x) is the generating polynomial, and x is the variable symbol; The fault diagnosis model of the intelligent early warning module is: Input vector X = [θ, T, H, C]T; decision output y = sign(ω·X+b); where θ is the discoloration rate of silica gel, T is the ambient temperature, H is the relative humidity, C is the SF6 gas concentration, ω is the SVM weight vector, and b is the decision hyperplane bias term.

Images