A transformer oil online monitoring jet diffusion type oil-gas separation device and method

Abstract

The present application relates to oil dissolved gas oil-gas separation technical field, disclose a kind of transformer oil online monitoring injection diffusion type oil-gas separation device and method, the device includes horizontally arranged degassing chamber, buffer tube, vertically arranged injection cylinder and gas sampling module, wherein, the top side of injection cylinder is sequentially connected with pressure sensor, three-way electromagnetic valve, pressure relief valve, proportional valve and carrier gas interface, the bottom side of injection cylinder is sequentially connected with electromagnetic valve and unpowered multi-stage injection nozzle, gas sampling module is composed of standard gas interface, carrier gas interface, dosing tube, chromatographic column interface, pressure sensor, electromagnetic valve and three-way electromagnetic valve.The present application does not need oil circulating pump, and the function of oil sample circulation can be realized by degassing chamber-stepping motor;In negative pressure environment, using high pressure, high speed injection diffusion, oil dissolved gas can be quickly removed, the degassing time is shortened, and the degassing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of dissolved gas separation technology, and more specifically, to a jet diffusion type oil-gas separation device and method for online monitoring of transformer oil. Background Technology

[0002] In the vast architecture of the power system, transformers are one of the core electrical devices ensuring stable power transmission. Their stable operation is not only the cornerstone of reliable power supply but also has a profound impact on society's production and daily life. Therefore, ensuring the stable operation of transformers is of paramount importance for enhancing the reliability of power supply from the power system.

[0003] Dissolved gas analysis (DGA) in oil has been widely recognized in the power industry as an important means of preventing faults in oil-filled electrical equipment. In the standard DL / T596-2021 "Preventive Testing Procedures for Power Equipment," DGA is listed as the second test item, demonstrating its importance in power equipment maintenance. Currently, laboratory gas chromatography is one of the main methods for DGA. This method, with its high precision and accuracy, provides reliable data support for fault diagnosis of power equipment. However, with the rapid development of technology, online DGA detection technology is also gradually being promoted and applied. This technology, by installing an online DGA analyzer on the transformer, enables real-time monitoring of the composition and content of dissolved gases in the oil, providing more timely and accurate information on the transformer's operating status.

[0004] Oil-gas separation is an indispensable and crucial step in the entire process of dissolved gas analysis in oil. Currently, commonly used oil-gas separation methods mainly include membrane degassing, dynamic headspace degassing, and vacuum degassing. However, each of these methods has certain limitations. While membrane degassing is relatively simple in principle, it suffers from long equilibrium times and low degassing efficiency. In practical applications, this method struggles to meet timeliness requirements, especially when rapid results are needed, where its disadvantages become more pronounced. Dynamic headspace degassing requires gas replenishment during the degassing process, and the waste oil must undergo treatment before being pumped back to the main transformer, increasing both operational complexity and operating costs. Furthermore, vacuum degassing involves complex equipment structures, long degassing times, and its degassing rate is easily affected by the gas content in the oil, resulting in poor repeatability. This makes it difficult to guarantee the consistency and reliability of test results in actual testing.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] To address the problems in related technologies, this invention proposes a transformer oil online monitoring jet diffusion type oil-gas separation device and method to overcome the aforementioned technical problems existing in the prior art.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a transformer oil online monitoring jet diffusion type oil-gas separation device is provided, comprising:

[0009] The degassing chamber is horizontally positioned. Inside the degassing chamber is a piston, which is connected to a stepper motor on one side of the degassing chamber via a screw-slider structure to drive the piston to move within the degassing chamber. The bottom side of the degassing chamber is connected to the solenoid valve V1 and the oil inlet of the transformer via an oil inlet pipe. The bottom side of the degassing chamber is also connected to the solenoid valve V2 and the oil return port of the transformer via a return oil pipe. A pressure sensor P1 is located in the middle of one side of the degassing chamber.

[0010] The buffer tube is connected to the top side of the degassing chamber via a pipeline and a solenoid valve V5 at its bottom end. Liquid level sensors L2 and L1 are installed on the top and bottom sides of the buffer tube, respectively. The top of the buffer tube is connected to the solenoid valve V4 and the gas injection module in sequence via a pipeline.

[0011] The vertically arranged injection cylinder has its top side connected to pressure sensor P3, three-way solenoid valve V9, pressure relief valve, proportional valve and carrier gas interface in sequence via pipelines. Its bottom side is connected to solenoid valve V12 and non-powered multi-stage injection nozzle in sequence via pipelines. The non-powered multi-stage injection nozzle is located at the top end of the degassing chamber. The bottom side of the injection cylinder is also connected to solenoid valve V3 and the bottom side of the degassing chamber in sequence via pipelines.

[0012] The gas injection module consists of a standard gas interface, a carrier gas interface, a quantitative tube, a chromatographic column interface, a pressure sensor P2, solenoid valves V6, V8, V10, and V11, and a three-way solenoid valve V7. The standard gas interface is connected to solenoid valve V11 via a pipeline, the carrier gas interface is connected to three-way solenoid valve V7 via a pipeline, the top of the quantitative tube is connected to solenoid valves V8 and V10 via a pipeline, the bottom of the quantitative tube is connected to three-way solenoid valve V7, pressure sensor P2, solenoid valves V6 and V11 via a pipeline, and the chromatographic column interface is connected to three-way solenoid valves V7 and V8 via a pipeline.

[0013] Among them, the proportional valve and the pressure relief valve are both piezoelectric ceramic valves, and the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis backfeedforward compensation voltage after dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage.

[0014] Furthermore, in order to better limit the movement of the piston in the degassing chamber, a first limit sensor, a second limit sensor, and a third limit sensor are arranged sequentially from left to right on the top of the degassing chamber, and a heating control module is arranged on the outer side of the degassing chamber and the outer side of the injection cylinder.

[0015] Furthermore, to expand the contact area between the gas and liquid phases, improve oil-gas separation efficiency, and shorten oil-gas separation time, the non-powered multi-stage nozzle includes a base. A nozzle is located on one side of the base, and a primary contraction section is formed inside the nozzle. The other end of the primary contraction section is connected to a secondary contraction section. A vacuum suction chamber is formed between the nozzle and the base, with one end of the vacuum suction chamber connected to the secondary contraction section and the other end sequentially connected to a tertiary contraction section, a throat section, and a diffusion zone located inside the base. Gas suction channels and liquid suction channels connected to the vacuum suction chamber are located on both the top and bottom sides of the base. One end of the nozzle is a high-pressure liquid inlet connected to a solenoid valve V12. One end of the gas suction channel is connected to the upper space of the degassing chamber, and the other end is connected to the top of the vacuum suction chamber. One end of the diffusion zone is a diffusion outlet connected to the upper space of the degassing chamber. One end of the liquid suction channel is connected to the lower space of the degassing chamber, and the other end is connected to the bottom of the vacuum suction chamber.

[0016] Furthermore, to better facilitate gas injection, when in degassing mode, the carrier gas passes through the carrier gas interface, then through the inlet of the three-way solenoid valve V7 to the outlet of the three-way solenoid valve V7, and then through the column interface into the chromatographic column. When the gas injection module is in injection mode, the carrier gas passes through the carrier gas interface, then sequentially through the three-way solenoid valve V7, the metering tube, and the solenoid valve V8 to the column interface and into the chromatographic column. When the gas injection module is in standard gas mode, the standard gas passes through the standard gas interface, then sequentially through the solenoid valve V11, the metering tube, and the solenoid valve V10 before being purged. When the gas injection module is in purging mode, the carrier gas passes through the carrier gas interface, then through the three-way solenoid valve V7 to the column interface, then through the carrier gas interface, then through the three-way solenoid valve V7, then through the solenoid valve V8 and the solenoid valve V10 before being purged. Finally, the carrier gas passes through the carrier gas interface to the three-way solenoid valve V7, then through the solenoid valve V8, the metering tube, the solenoid valve V6, and the solenoid valve V4 before being purged.

[0017] Furthermore, the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis feedforward compensation voltage (after dual-valve output consistency calibration compensation) and the reinforcement learning fine-tuning voltage, including:

[0018] The pressure feedback signal and target pressure command of the controlled object are acquired in real time. The pressure deviation is calculated based on the pressure feedback signal and the target pressure. The target displacement required by the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve is calculated by combining the target pressure difference curve.

[0019] The target displacement commands (derived from the target displacement amount) of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are input into the corresponding pre-trained hysteresis inverse model to generate the feedforward compensation voltage of the corresponding valve.

[0020] The reinforcement learning controller is optimized using a proximal policy. The state space is the hidden state vector of the pressure deviation and the hysteresis inverse model and the actual displacement of the corresponding piezoelectric ceramic valve. The output is the fine-tuning correction voltage of the corresponding valve.

[0021] The feedforward compensation voltage and the fine-tuning correction voltage of the corresponding valve are superimposed to generate the initial drive voltages of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve, respectively.

[0022] By utilizing the gain deviation coefficient and hysteresis deviation coefficient of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve, the initial driving voltage of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are corrected respectively to obtain the final driving voltage of the corresponding valve.

[0023] Based on the direction of pressure deviation, the final drive voltage of the corresponding valve is applied to the corresponding piezoelectric ceramic valve to change the valve opening, thereby controlling the flow rate and pressure of the carrier gas.

[0024] Furthermore, the input to the hysteresis inverse model includes a target displacement sequence consisting of the target displacement commands of the past M time steps, and the first-order rate of change feature of the current target displacement command. The model is trained by applying multi-frequency, multi-amplitude broadband excitation signals to the piezoelectric valve in the offline stage and collecting the mapping data of its input voltage and output displacement to learn the rate-dependent hysteresis nonlinearity of the piezoelectric ceramic and its inverse mapping.

[0025] Furthermore, the expression for the reward function of the proximal policy optimization reinforcement learning controller is as follows:

[0026] ;

[0027] In the formula, R t Let e(t) be the total reward value at time t, and e(t) be the pressure deviation. Energy consumption weighting coefficient Let be the fine-tuning correction voltage at time t. For smoothness weighting coefficients, For degradation protection weighting coefficients, For degradation protection functions, Let t be the final driving voltage at time t.

[0028] Furthermore, based on the direction of the pressure deviation, the final driving voltage of the corresponding valve is selected and applied to the corresponding piezoelectric ceramic valve, including:

[0029] When the pressure deviation is positive, the valve opening of the piezoelectric ceramic pressure relief valve is adjusted by the final drive voltage of the piezoelectric ceramic pressure relief valve to increase the opening of the piezoelectric ceramic pressure relief valve.

[0030] When the pressure deviation is negative, the valve opening of the piezoelectric ceramic proportional valve is adjusted by the final drive voltage of the piezoelectric ceramic proportional valve to increase the opening of the piezoelectric ceramic proportional valve.

[0031] When the pressure deviation is zero, the valve opening of the electro-ceramic proportional valve and the piezoelectric ceramic pressure relief valve remains unchanged, and the update of the drive voltage signal is stopped.

[0032] According to another aspect of the present invention, a jet diffusion type oil-gas separation method for a transformer oil online monitoring jet diffusion type oil-gas separation device is provided, comprising the following steps:

[0033] Start the oil-gas separation process, and sequentially execute the temperature recognition, initial self-test, gas path cleaning and air tightness test processes. After the test is passed, execute the oil circulation and quantitative process.

[0034] After the oil sample is quantified, the stepper motor is first controlled to transfer the transformer oil in the degassing chamber to the injection cylinder. Then, the stepper motor is controlled to pull the piston to the right limit, creating a vacuum inside the degassing chamber. Next, the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis feedforward compensation voltage after the dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage. The injection cylinder is then controlled by driving the proportional valve or pressure relief valve to quickly inject the transformer oil in the chamber into the degassing chamber. During the injection process, the dissolved gas in the oil is transferred from the liquid phase to the gas phase and accumulates at the top of the degassing chamber and in the buffer tube. After the injection degassing is completed, the piston position is adjusted by controlling the stepper motor to push the piston rod, transferring the sample gas accumulated in the degassing chamber to the buffer tube and the quantitative tube in the gas injection module.

[0035] After degassing is completed, the gas injection module is controlled to the injection state, and the degassed gas is sent into the chromatographic column for gas separation. The concentration of each component in the gas sample is obtained by chromatographic analysis.

[0036] The beneficial effects of this invention are as follows:

[0037] 1) This invention eliminates the need for an oil circulation pump, achieving oil sample circulation through a degassing chamber and a stepper motor; under negative pressure, high-pressure, high-speed jet diffusion is used to quickly remove dissolved gases from the oil, shortening the degassing time and improving degassing efficiency; the structure of the degassing system is optimized, reducing investment costs.

[0038] 2) This invention employs multi-stage jet series technology, utilizing the Venturi effect to induce, mix, atomize, and disperse gas and liquid in the degassing chamber by using the low pressure of the vacuum suction chamber. This achieves induction and mixing without additional power, and atomizes and disperses the liquid into tiny droplets, thereby expanding the contact area between the gas and liquid phases, improving oil-gas separation efficiency, and shortening oil-gas separation time.

[0039] 3) This invention accelerates oil-gas separation through jet diffusion, thereby achieving rapid oil-gas balance. It does not require consideration of the influence of degassing environment pressure and degassing rate on the analysis results, and has the advantages of simple structure and good repeatability.

[0040] 4) This invention achieves rapid jet diffusion by driving the piston of the jet cylinder with high-pressure gas. It consists of a jet cylinder, a diffusion nozzle, and valve components. It features simple structure, no mechanical transmission parts, reliable operation, no need for operation and maintenance, and long service life.

[0041] 5) This invention utilizes a composite architecture of LSTM hysteresis feedforward compensation and proximal strategy optimization reinforcement learning fine-tuning, combined with a dual-valve output consistency symmetrical compensation mechanism, to significantly improve the injection differential pressure control accuracy and substantially improve the consistency of degassing. Simultaneously, it can automatically correct manufacturing tolerances and long-term aging deviations of the dual valves, extending the service life of the piezoelectric ceramic valves and effectively improving the repeatability and data reliability of online monitoring and chromatographic analysis of transformer oil.

[0042] 6) This invention indirectly calculates the volume of degassing gas by collecting the pressure during the degassing process. Combined with the gas component concentration given by chromatographic analysis, the concentration of dissolved gas in the oil can be calculated by referring to relevant formulas using the gas partition theorem (Ostwald coefficient). It does not need to consider the influence of degassing environmental pressure and degassing rate on the gas analysis results. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the structure of a transformer oil online monitoring jet diffusion type oil-gas separator according to an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a non-powered multi-stage spray nozzle in a transformer oil online monitoring jet diffusion type oil-gas separator according to an embodiment of the present invention.

[0046] In the picture:

[0047] 1. Transformer; 2. Degassing chamber; 3. Stepper motor; 4. Buffer tube; 5. Standard gas interface; 6. Carrier gas interface; 7. Quantitative tube; 8. Chromatographic column interface; 9. Injection cylinder; 10. Gas injection module; 11. Heating control module; 12. Proportional valve; 13. Non-powered multi-stage injection nozzle; 1301. Matrix; 1302. Nozzle; 1303. First-stage contraction section; 1304. Second-stage contraction section; 1305. Vacuum suction chamber; 1306. Third-stage contraction section; 1307. Throat section; 1308. Diffusion zone; 1309. Gas suction channel; 1310. Liquid suction channel; 14. Pressure relief valve. Detailed Implementation

[0048] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0049] Example 1

[0050] like Figures 1-2 As shown, a transformer oil online monitoring jet diffusion type oil-gas separation device includes:

[0051] The degassing chamber 2 is horizontally arranged. A piston is installed inside the degassing chamber 2. The piston is connected to a stepper motor 3 on one side of the degassing chamber 2 through a screw and slider structure to drive the piston to move inside the degassing chamber 2. The bottom side of the degassing chamber 2 is connected to the oil inlet of the solenoid valve V1 and the oil inlet of the transformer 1 in sequence through an oil inlet pipe. The bottom side of the degassing chamber 2 is also connected to the oil return of the solenoid valve V2 and the oil return of the transformer 1 in sequence through a return oil pipe. A pressure sensor P1 is installed in the middle of one side of the degassing chamber 2.

[0052] The bottom end of the buffer tube 4 is connected to the top side of the degassing chamber 2 via a pipeline and a solenoid valve V5. A liquid level sensor L2 and a liquid level sensor L1 are respectively installed on the top and bottom sides of the buffer tube 4. The top end of the buffer tube 4 is connected to the solenoid valve V4 and the gas injection module 10 in sequence via a pipeline.

[0053] The vertically arranged injection cylinder 9 has its top side connected to the pressure sensor P3, three-way solenoid valve V9, pressure relief valve 14, proportional valve 12 and carrier gas interface 6 in sequence via pipelines. The bottom side of the injection cylinder 9 is connected to the solenoid valve V12 and the unpowered multi-stage injection nozzle 13 in sequence via pipelines. The unpowered multi-stage injection nozzle 13 is located at the top end of the degassing chamber 2. The bottom side of the injection cylinder 9 is also connected to the solenoid valve V3 and the bottom side of the degassing chamber 2 in sequence via pipelines.

[0054] The gas injection module 10 consists of a standard gas interface 5, a carrier gas interface 6, a quantitative tube 7, a chromatographic column interface 8, a pressure sensor P2, a solenoid valve V6, a solenoid valve V8, a solenoid valve V10, a solenoid valve V11, and a three-way solenoid valve V7. The standard gas interface 5 is connected to the solenoid valve V11 through a pipeline. The carrier gas interface 6 is connected to the three-way solenoid valve V7 through a pipeline. The top end of the quantitative tube 7 is connected to the solenoid valves V8 and V10 through a pipeline. The bottom end of the quantitative tube 7 is connected to the three-way solenoid valve V7, the pressure sensor P2, the solenoid valve V6, and the solenoid valve V11 through a pipeline. The chromatographic column interface 8 is connected to the three-way solenoid valve V7 and the solenoid valve V8 through a pipeline.

[0055] Among them, the proportional valve 12 and the pressure relief valve 14 are both piezoelectric ceramic valves, and the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis backfeedforward compensation voltage after dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage.

[0056] In practical applications, all sensors, solenoid valves, sensors, piezoelectric ceramic valves, stepper motors, and heating control modules in this embodiment are connected to the control terminal via lines. That is, the control terminal can issue control commands to control the operation of each component or obtain real-time data of each component. At the same time, the control terminal can execute a method to determine the driving voltage of the piezoelectric ceramic valve by superimposing the hysteresis backfeedforward compensation voltage after dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage.

[0057] The principle of oil-gas separation in this embodiment is as follows: by adjusting the gas-liquid two-phase volume ratio, oil-gas separation is achieved by pressurizing the injection through a cylinder under negative pressure conditions; negative pressure is not a necessary condition. The main purpose is to increase the contact area between the gas and liquid phases by adjusting the gas-liquid two-phase volume ratio, and accelerate the oil-gas separation process by cylinder injection. Increasing the gas-liquid two-phase volume ratio will inevitably generate negative pressure, but negative pressure is not a necessary condition for oil-gas separation.

[0058] Specifically, the top of the degassing chamber 2 is provided with a first limit sensor (i.e., left limit), a second limit sensor (i.e., middle limit), and a third limit sensor (i.e., right limit) from left to right, so as to better limit the movement of the piston in the degassing chamber. In addition, heating control modules 11 are provided on the outer side of the degassing chamber 2 and the outer side of the injection cylinder 9, so as to heat the oil or environment in the degassing chamber 2 and the injection cylinder 9.

[0059] In this embodiment, a heating control module 11 is provided for the degassing chamber and the injection cylinder. This module can precisely control the heating of the degassing chamber and the injection cylinder chamber and maintain a constant temperature of 50°C via a heating plate or heating coil, temperature sensor, or temperature switch. Through a temperature compensation mechanism, the stability of the oil-gas balance is ensured, maintaining the constant temperature of the degassing chamber and the injection cylinder chamber, thus ensuring the accuracy and stability of the analysis results.

[0060] Specifically, the non-powered multi-stage nozzle 13 includes a base 1301, a nozzle 1302 is provided on one side of the base 1301, a first-stage contraction section 1303 is provided inside the nozzle 1302, and a second-stage contraction section 1304 is connected to the other end of the first-stage contraction section 1303; a vacuum suction chamber 1305 is formed between the nozzle 1302 and the base 1301, and one end of the vacuum suction chamber 1305 is connected to the second-stage contraction section 1304, and the other end of the vacuum suction chamber 1305 is sequentially connected to a third-stage contraction section 1306, a throat section 1307 and a diffusion zone 1308 provided inside the base 1301; a gas suction channel 1309 and a liquid suction channel 1310 communicating with the vacuum suction chamber 1305 are provided on the top side and the bottom side of the base 1301. One end of nozzle 1302 is a high-pressure liquid inlet and is connected to solenoid valve V12; one end of gas intake channel 1309 is connected to the upper space of degassing chamber 2, and the other end of gas intake channel 1309 is connected to the top of vacuum intake chamber 1305; one end of diffusion zone 1308 is a diffusion outlet and is connected to the upper space of degassing chamber 2; one end of liquid intake channel 1310 is connected to the lower space of degassing chamber 2, and the other end of liquid intake channel 1310 is connected to the bottom of vacuum intake chamber 1305.

[0061] In practical applications, the unpowered multi-stage nozzle in this embodiment adopts a Venturi tube variable diameter structure design. Utilizing the Venturi effect, a negative pressure is generated at the nozzle outlet through the high-speed flow of high-pressure fluid, drawing in low-pressure fluid, increasing the total fluid flow rate, and increasing the fluid thrust in the throat section. The oil injection diffusion process applies the Venturi effect's entrainment, mixing, atomization, and dispersion phenomena. By utilizing the low pressure of the vacuum suction chamber to draw in gas and liquid from the degassing chamber, it achieves entrainment and mixing without additional power, and atomizes and disperses the liquid into tiny droplets, thereby expanding the contact area between the gas and liquid phases, improving oil-gas separation efficiency, and shortening oil-gas separation time.

[0062] In addition, the non-powered multi-stage nozzle adopts multi-stage injection series technology. After the transformer oil passes through the nozzle diameter change of the second-stage contraction section, it is sprayed out of the injection port at high speed. At this time, the pressure of the vacuum suction chamber behind the injection port decreases, thereby drawing in gas from the degassing chamber from the B-end suction port and transformer oil from the degassing chamber from the D-end suction port, increasing the total fluid flow of the throat section, increasing the fluid thrust, and improving the oil-gas separation effect in the diffusion zone.

[0063] Meanwhile, dynamic pressure regulation technology is adopted. The power of the high-speed fluid is the pressure generated by the carrier gas acting on the piston of the injection cylinder 9 after passing through the proportional valve 12 and the solenoid valve V9. The proportional valve 12 and the pressure relief valve 14 are both piezoelectric ceramic valves. The micro-flow valve driven by piezoelectric ceramics automatically adjusts the carrier gas pressure according to the real-time feedback of the degassing chamber pressure sensor to ensure sufficient oil pressure in the unpowered multi-stage injection nozzle.

[0064] Specifically, when the gas injection module is in degassing mode, the carrier gas passes through the carrier gas interface 6, then through the inlet of the three-way solenoid valve V7 to the outlet of the three-way solenoid valve V7, and then enters the chromatographic column through the column interface 8; when the gas injection module 10 is in injection mode, the carrier gas passes through the carrier gas interface 6, then sequentially through the three-way solenoid valve V7, the metering tube 7, and the solenoid valve V8 to the column interface 8 and enters the chromatographic column; when the gas injection module 10 is in standard gas mode, the standard gas passes through the standard gas interface 5... The gas is purged after passing through solenoid valve V11, metering tube 7, and solenoid valve V10 in sequence. When the gas injection module 10 is in the purging state, the carrier gas passes through carrier gas interface 6, then through three-way solenoid valve V7 to the column interface 8. Then the carrier gas passes through carrier gas interface 6, then through three-way solenoid valve V7, then through solenoid valve V8 and solenoid valve V10 before being purged. Finally, the carrier gas passes through carrier gas interface 6 to three-way solenoid valve V7, then through solenoid valve V8, metering tube 7, solenoid valve V6, and solenoid valve V4 before being purged.

[0065] Example 2

[0066] A dynamic pressure intelligent control method for a piezoelectric ceramic valve is disclosed. This method determines the driving voltage of the piezoelectric ceramic valve based on the superposition of a hysteresis feedforward compensation voltage (after dual-valve output consistency calibration and compensation) and a reinforcement learning fine-tuning voltage. The valve opening is then adjusted using this driving voltage to achieve dynamic pressure intelligent control. Specifically, the method includes:

[0067] Step 1: Acquire the pressure feedback signal and target pressure command of the controlled object in real time, calculate the pressure deviation based on the pressure feedback signal and the target pressure, and calculate the target displacement required by the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve in combination with the target pressure difference curve.

[0068] Specifically, in the "transformer oil degassing injection" scenario of this embodiment, the path for generating the target differential pressure curve is as follows:

[0069] 1) Process requirements and multimodal sensing target pressure difference It is not a fixed constant, but a time-varying sequence dynamically generated based on the characteristics of the oil being processed and the degassing stage.

[0070] 2) In the pretreatment process for dissolved gas analysis in oil, jet degassing typically involves several different process stages:

[0071] ① Pressure building and pre-purge stage: It is necessary to establish an initial pressure difference in a short period of time to remove residual gas from the pipeline.

[0072] ② Constant pressure extraction stage: When the oil and gas are in full contact, maintain an absolutely constant low pressure difference to ensure the efficient precipitation of characteristic gases.

[0073] ③ Pressure reduction and rinsing stage: After extraction, the pressure difference is slowly reduced, and the mixed gas enriched with the characteristic gas is pushed into the metering tube by the carrier gas.

[0074] 3) The target differential pressure curve is a continuous curve function generated by the workflow within the host computer or control system based on the above-mentioned process. For example, it may be a trapezoidal or S-shaped curve that first rises at an incline, then remains horizontal, and finally declines at an incline.

[0075] Step 2: Input the target displacement commands of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve into the corresponding pre-trained hysteresis inverse model to generate the feedforward compensation voltage of the corresponding valve.

[0076] The input of the hysteresis inverse model includes a target displacement sequence consisting of the target displacement commands of the past M time steps, and the first-order rate of change feature of the current target displacement command. The model is trained by applying a wideband excitation signal with multiple frequencies and amplitudes to the piezoelectric valve in the offline stage and collecting the mapping data between its input voltage and output displacement, so as to learn the rate-dependent hysteresis nonlinearity of the piezoelectric ceramic and its inverse mapping.

[0077] In this embodiment, the structure, parameters, and training process of the hysteresis inverse model are as follows:

[0078] 1. Model Structure: This model adopts a sequence regression architecture consisting of a double-stacked long short-term memory network and a fully connected layer, as shown in Table 1 below:

[0079] Table 1. Structure of the Hysteresis Inverse Model

[0080] Design highlights: A dual-layer LSTM captures the static hysteresis memory effect of piezoelectric ceramics; the displacement rate of change is used as an explicit input, enabling the model to sense the excitation frequency, thus learning rate-dependent hysteresis characteristics.

[0081] 2. The model parameters are shown in Table 2 below:

[0082] Table 2 Model Parameters

[0083] 3. Training process

[0084] 3.1) Data Acquisition

[0085] Excitation signal: a chirped voltage signal with logarithmic sweep frequency superimposed with random amplitude, frequency 0.1Hz~1kHz, amplitude covering the full range of -50V~+800V.

[0086] Synchronous acquisition: The driving voltage V is synchronously recorded at a sampling rate of 100kHz. t Feedback from displacement sensor D t .

[0087] Data volume: Approximately 12 million data points are collected per valve.

[0088] 3.2) Data Preprocessing

[0089] Filtering: 2kHz low-pass filter for noise reduction;

[0090] Normalization: Voltage is mapped to [-1,1], and displacement is mapped to [0,1].

[0091] Window capture: A sliding window (M=50, step size 1 point) is used to generate sequence samples;

[0092] Tag construction: The displacement sequence is used as input and the corresponding voltage value is used as output (inverse mapping);

[0093] Dataset split: 80% training set, 10% validation set, and 10% test set.

[0094] 3.3) Training and Acceptance Standards

[0095] The Adam optimizer was used with a batch size of 256, and training continued until the validation loss converged.

[0096] Acceptance criteria: The residual hysteresis after open-loop inverse compensation on the test set is less than 1.5% of the full scale, which means that more than 90% of the hysteresis nonlinear error has been eliminated by feedforward compensation alone.

[0097] 4. Deployment Instructions

[0098] After training, the model is quantized using INT8 and compiled into an FPGA hardware IP core deployed on an embedded controller. During online inference, M historical target displacement commands are read from the buffer each control cycle, concatenated with the current rate of change feature, and a feedforward voltage is output after one forward propagation. The inference delay is no more than 10 microseconds.

[0099] Step 3: Optimize the reinforcement learning controller using a near-end policy. The state space is the hidden state vector of the pressure deviation and the hysteresis inverse model, and the actual displacement of the corresponding piezoelectric ceramic valve. The output is the fine-tuning correction voltage of the corresponding valve.

[0100] The expression for the reward function of the proximal policy optimization reinforcement learning controller is as follows:

[0101] ;

[0102] In the formula, R t Let e(t) be the total reward value at time t, and e(t) be the pressure deviation. Energy consumption weighting coefficient Let be the fine-tuning correction voltage at time t. For smoothness weighting coefficients, For degradation protection weighting coefficients, For degradation protection functions, Let be the final driving voltage at time t;

[0103] in, This is a penalty item for pressure control accuracy. To calibrate motion energy consumption and internal stress penalty terms; This refers to the high-frequency oscillation and electrode stress suppression term; The penalty value is increased exponentially when the driving voltage approaches the material safety boundary, which is a function term for piezoelectric ceramic degradation protection. This guides the control strategy to achieve an optimal trade-off between actuator life and energy consumption while pursuing pressure accuracy.

[0104] In this embodiment, the near-end policy optimization reinforcement learning controller outputs a fine-tuned correction voltage in a multi-dimensional state space, including:

[0105] 1. Construction of the state space

[0106] In each control cycle, the controller constructs a state vector S in real time from the following three sources. t As shown in Table 3:

[0107] Table 3. Composition and Source of State Vectors

[0108] Total state dimensions: 68. This state vector is constructed in real time on the FPGA each control cycle and fed into the PPO's Actor policy network.

[0109] 2. Network Structure and Inference Process

[0110] The PPO controller employs an Actor-Critic dual-network architecture, with only the Actor network propagating forward during inference: State vector S t(68-dimensional) → Fully connected layer (128, ReLU) → Fully connected layer (128, ReLU) → Output layer (1, Tanh) │ δV ∈ [-1, 1] → Scale to [-50V, +50V] → Fine-tune correction voltage δV.

[0111] Reasoning process:

[0112] 1) In each 100μs control cycle, the hardware accelerator on the FPGA feeds the 68-dimensional state vector into the Actor network at once; 2) After two layers of fully connected forward propagation (total delay <10μs), the output is normalized to the action value [-1,1]; 3) Linearly scaled to the ±50V engineering range, the fine-tuning correction voltage δV is obtained; 4) Compared with the LSTM feedforward voltage V... ff Overlay: The output is sent to the drive power supply.

[0113] 3. Brief description of the training mechanism

[0114] Algorithm: Proximal policy optimization, which limits the policy update magnitude by pruning the objective function to ensure training stability.

[0115] Experience collection: In the digital twin simulation environment, the LSTM inverse model is first used as the feedforward link of the controlled object, and the PPO runs in coordination with it to collect the trajectory (state, action, reward, next state).

[0116] Reward function: Calculated according to the multi-objective reward function of claim 8.

[0117] Policy transfer: After the simulation training converges, the policy is deployed to the physical controller via INT8 quantization; it continues to be fine-tuned online on the physical controller with a very small learning rate (0.0001) so that the policy can smoothly transition from "simulation optimal" to "physical optimal".

[0118] 4. Core Advantages

[0119] Delayed memory vector The key role of PPO is that traditional PID controllers are "blind" to the hysteresis of piezoelectric ceramics, while PPO can accurately sense whether it is currently in the lift or retrace phase and the thickness of the hysteresis loop by reading the hidden state of LSTM, thereby predicting that the same pressure deviation will require completely different correction amounts under different paths.

[0120] Displacement feedback The value of this is to provide actual valve response verification, enabling PPO to detect minor mismatches in the feedforward model online and compensate for them in a timely manner, thus achieving organic synergy between feedforward and feedback.

[0121] Step 4: Superimpose the feedforward compensation voltage and the fine-tuning correction voltage of the corresponding valve to generate the initial drive voltage of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve respectively.

[0122] The initial driving voltage of the piezoelectric ceramic proportional valve is obtained by superimposing the feedforward compensation voltage and the fine-tuning correction voltage of the piezoelectric ceramic proportional valve; the initial driving voltage of the piezoelectric ceramic pressure relief valve is obtained by superimposing the feedforward compensation voltage and the fine-tuning correction voltage of the piezoelectric ceramic pressure relief valve.

[0123] Step 5: Using the gain deviation coefficient and hysteresis deviation coefficient of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve, the initial driving voltage of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are corrected respectively to obtain the final driving voltage of the corresponding valve.

[0124] Specifically, the pre-stored dual-valve symmetric compensation matrix is ​​invoked to multiply the initial drive voltages of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve by the corresponding valve's gain compensation coefficient and hysteresis compensation coefficient, respectively, to complete the real-time amplitude correction of the dual-valve output consistency and obtain the corrected drive voltage. The dual-valve symmetric compensation matrix is ​​generated and periodically updated by the dual-valve characteristic calibration process executed by the system during idle periods, and includes the deviation mapping relationship between the factory standard characteristics and the current measured characteristics.

[0125] The method for determining the dual-valve symmetric compensation matrix is ​​as follows:

[0126] 1) After the system triggers the calibration conditions (idle period after completing a single degassing cycle, continuous operation for 24 hours, or first power-on), close all inlet and outlet valves of the degassing chamber to form a closed calibration cavity between the degassing chamber and the injection cylinder;

[0127] 2) The pressure in the closed cavity is stabilized to the standard calibration pressure by dual-valve coordinated control. After stabilization, the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are driven separately in sequence. A preset multi-amplitude stepped voltage signal sequence (e.g., 0V to 600V, step size 50V, each voltage is held for 20ms) is input, and the actual valve displacement and cavity pressure data corresponding to each voltage point are collected synchronously.

[0128] 3) Fit the measured "drive voltage-output displacement" characteristic curves of the two valves respectively, compare them with the factory standard characteristic curves stored in the system, and calculate the gain deviation coefficient K of each valve. i and hysteresis deviation coefficient H i ;

[0129] Gain deviation coefficient: , where k std k is the slope of the standard characteristic curve. mes,i Let be the slope of the measured characteristic curve of the i-th valve (i=1 corresponds to the piezoelectric ceramic proportional valve, and i=2 corresponds to the piezoelectric ceramic pressure relief valve).

[0130] Hysteresis deviation coefficient: ,in, The maximum width of the standard hysteresis loop. The measured maximum width of the hysteresis loop for the i-th valve (i=1 corresponds to the piezoelectric ceramic proportional valve, and i=2 corresponds to the piezoelectric ceramic pressure relief valve).

[0131] Maximum width of standard hysteresis loop: Before the valve leaves the factory, a complete voltage rise-fall cycle (linearly rising from 0V to the maximum drive voltage, and then linearly falling back to 0V) is applied to the standard calibration sample valves of the same batch. The drive voltage and the corresponding output displacement are collected synchronously, and a "voltage-displacement" hysteresis loop is plotted. The maximum difference between the voltage at the rising edge and the voltage at the falling edge under the same displacement is taken as the standard hysteresis parameter of this model of valve and pre-stored in the non-volatile memory of the control terminal.

[0132] Measured maximum hysteresis loop width: When the system performs the dual-valve consistency calibration process, the same rise-fall voltage cycle as the factory calibration is applied to the current i-th valve (proportional valve or pressure relief valve), the current "voltage-displacement" data of the valve is collected and the measured hysteresis loop is plotted. Similarly, the maximum difference between the voltage at the rising edge and the falling edge under the same displacement is taken as the current measured maximum hysteresis loop width of the valve.

[0133] 4) Generate the dual-valve symmetric compensation matrix:

[0134] ;

[0135] The first row corresponds to the compensation coefficient of the piezoelectric ceramic proportional valve, and the second row corresponds to the compensation coefficient of the piezoelectric ceramic pressure relief valve.

[0136] 5) Write the updated compensation matrix into the non-volatile memory of the control terminal. If the deviation coefficient of any valve exceeds the preset normal range of 0.8~1.2, the system will automatically issue a valve performance degradation warning.

[0137] Step 6: Based on the direction of the pressure deviation, select the final drive voltage of the corresponding valve and apply it to the corresponding piezoelectric ceramic valve to change the valve opening, thereby controlling the carrier gas flow rate and pressure.

[0138] Specifically, based on the direction of the pressure deviation, the final driving voltage of the corresponding valve is selected and applied to the corresponding piezoelectric ceramic valve, including:

[0139] When the pressure deviation is positive (the actual pressure difference is higher than the target pressure difference), the valve opening of the piezoelectric ceramic pressure relief valve is adjusted by using the final drive voltage of the piezoelectric ceramic pressure relief valve, so as to increase the opening of the piezoelectric ceramic pressure relief valve and reduce the system pressure difference.

[0140] When the pressure deviation is negative (the actual pressure difference is lower than the target pressure difference), the valve opening of the piezoelectric ceramic proportional valve is adjusted by using the final drive voltage of the piezoelectric ceramic proportional valve to increase the opening of the piezoelectric ceramic proportional valve and improve the system pressure difference.

[0141] When the pressure deviation is zero, the valve opening of the electro-ceramic proportional valve and the piezoelectric ceramic pressure relief valve remains unchanged, and the update of the drive voltage signal is stopped.

[0142] In addition, this embodiment also includes a digital twin-driven online model drift detection and adaptation mechanism: it continuously calculates the residual error statistical characteristics after compensation by the hysteresis inverse model, and when it detects that the mean or variance of the error has continuously drifted, it automatically triggers a low-amplitude safe frequency scan, and uses the newly collected data to perform local online fine-tuning of the hysteresis inverse model to adapt to the aging or temperature changes of the actuator characteristics.

[0143] Meanwhile, the state space of the near-end policy optimization reinforcement learning controller also contains a load characteristic representation vector identified by the online system in real time; this vector enables the above method for determining the drive voltage of the piezoelectric ceramic valve to automatically adapt to downstream loads of different volumes or different media without the need for manual parameter readjustment.

[0144] To better understand the above-mentioned technical solution for dynamic pressure intelligent control of piezoelectric ceramic valves, the following explanation is provided from the perspective of technical principles:

[0145] In the pretreatment stage of online monitoring of dissolved gases in oil-immersed power transformers, fault characteristic gases need to be efficiently separated from the insulating oil through a jet degassing method. Transformer oil is injected into the degassing chamber at a constant pressure, forming an oil-gas interface for mass transfer. During the injection process, the pressure difference between the input end (injection cylinder) and the outlet end of the degassing chamber must remain absolutely stable. Any minute fluctuation will alter the gas mass transfer efficiency, leading to inconsistent degassing rates and ultimately affecting the reliability of the chromatographic analysis data.

[0146] I. System Configuration and Signal Acquisition for Implementing this Method

[0147] Pressure sensor group: respectively installed at the carrier gas input end (P3) and the degassing chamber outlet end (P2), using resonant high-precision transmitters with an accuracy of 0.01 and a sampling rate of 10kHz;

[0148] Actuation valve assembly: a piezoelectric ceramic driven micro-flow intake proportional valve and a piezoelectric ceramic micro-flow pressure relief valve;

[0149] Control terminal: Based on Xilinx Zynq heterogeneous SoC, the ARM side is responsible for task scheduling, and the FPGA side implements hardware acceleration of all algorithms;

[0150] High-voltage drive power supply: Receives the output signal from the controller, linearly amplifies it, and then drives the piezoelectric ceramic;

[0151] The pressure sensor transmits real-time electrical signals to the control terminal with the goal of stabilizing the input-output pressure difference at a set value (e.g., 300.00 kPa) with a tolerance of ±0.01 kPa.

[0152] II. Multi-stage intelligent control process

[0153] Unlike traditional solutions that rely solely on PID deviation calculation, this system sequentially undergoes four stages of processing within each 100-microsecond control cycle: "deviation perception - feedforward prediction - feedback fine-tuning - safety synthesis and correction".

[0154] Phase 1: Deviation Perception and State Extraction

[0155] The control terminal receives the electrical signal from the pressure sensor and calculates the current pressure deviation. :

[0156] ;

[0157] In the formula, Set the target differential pressure value. This is the actual pressure difference. The current moment;

[0158] Simultaneously extract the integral term of the deviation. and differential terms These constitute the three-dimensional characteristics of the PID.

[0159] Pressure deviation It is mainly used for solving the target displacement command in stage 2; in stage 3, it serves as the core component of the PPO state space.

[0160] Deviation integral term Primarily used in stage 3, it forms the PPO state space and is used to eliminate long-term steady-state errors.

[0161] Deviation differential term Primarily used in stage 3, it forms the PPO state space and is used to predict pressure change trends and prevent overshoot.

[0162] Furthermore, the controller retains the hidden state vector of the LSTM from the hysteresis inverse model inference of the previous cycle and reads the actual displacement x(t) of the piezoelectric valve from the embedded sensor. All of the above information together constitutes the multidimensional state space required by subsequent intelligent algorithms.

[0163] Phase 2, Feedforward Prediction—LSTM Hysteresis Inverse Model Compensation

[0164] This stage corresponds to one of the core innovations of this invention. The control terminal does not directly transmit pressure deviation. Instead of calculating voltage, the system first calculates the target displacement required by the piezoelectric ceramic valve based on the target differential pressure curve. This target displacement sequence and the current displacement change rate are then fed into a pre-trained LSTM hysteresis inverse model.

[0165] The model, trained offline with wideband excitation, has thoroughly grasped the complex nonlinear voltage-displacement mapping relationship of this batch of piezoelectric ceramic valves. During inference, the model outputs a feedforward compensation voltage in real time based on the target displacement and historical path, compensating for over 90% of the hysteresis and rate-dependent nonlinearity of the piezoelectric ceramic valves in one go. This action enables the valve to rapidly approach the correct opening degree in an extremely short time (microseconds), laying a high starting point for subsequent closed-loop fine-tuning.

[0166] Phase 3, Feedback Fine-tuning—PPO Reinforcement Learning Online Calibration

[0167] The residual minute error is handled by the PPO feedback fine-tuning controller deployed on the FPGA. The PPO's Actor policy network receives the complete state vector—including deviation characteristics (PID three-dimensional characteristics: pressure deviation extracted in stage 1). Deviation integral term Differential term of deviation The inference output of the LSTM hidden state (representing hysteresis memory points), actual displacement, etc., is a fine-tuned correction voltage, the amplitude of which is usually limited to ±50V.

[0168] The key distinction at this stage lies in the fact that when the PPO makes corrective decisions, it is not only controlled by the stress bias itself, but also constrained by the multi-objective reward function unique to this invention. This reward function pursues... While maintaining high accuracy, the following behaviors will be explicitly penalized:

[0169] : Corrective actions to reduce excessive penalties, thereby reducing energy consumption and internal stress;

[0170] Suppresses high-frequency oscillations and protects the piezoelectric ceramic electrodes;

[0171] The penalty value increases exponentially when the driving voltage enters the material degradation danger zone.

[0172] Guided by this incentive, PPO learned to eliminate steady-state error with the smallest and smoothest action, rather than continuously generating high-frequency micro-amplitude oscillations like traditional PID controllers.

[0173] Phase 4, Secure Synthesis and Modification

[0174] 1) Determine the initial drive voltage: Initial drive voltage = feedforward compensation voltage + fine-tuning correction voltage;

[0175] 2) Call the pre-stored dual-valve symmetrical compensation matrix, multiply the initial drive voltage of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve by the corresponding valve's gain compensation coefficient and hysteresis compensation coefficient, respectively, to complete the real-time amplitude correction of the dual-valve output consistency and obtain the corrected drive voltage;

[0176] 3) The corrected driving voltage is passed through a hardware safety limiting circuit independent of the AI ​​algorithm (ensuring that the voltage never exceeds the -50V to +800V safe operating range of the piezoelectric ceramic) and then output to the high-voltage driving power supply.

[0177] 4) Based on the direction of deviation, selectively adjust the driving voltage of the intake valve or the pressure relief valve to change the valve opening, thereby accurately controlling the carrier gas flow and pressure, so that the actual pressure difference approaches the target value, and ensure that the pressure difference between the input and output ends is stable during the degassing process.

[0178] 5) Collect the actual displacement feedback and updated pressure feedback signals of the two piezoelectric ceramic valves in real time, and store the target displacement, feedforward compensation voltage, fine-tuning correction voltage, compensation coefficient, actual pressure difference and valve displacement data of the current control cycle into the control buffer, which will serve as the input basis for the hysteresis inverse model hidden state update and reinforcement learning controller state space of the next control cycle.

[0179] III. Comparison of Actual Operation Results and Advantages

[0180] In a continuous 720-hour (30-day) jet degassing aging test, this method exhibits performance that is difficult to achieve with traditional PID solutions (as shown in Table 4):

[0181] Table 4 Performance Comparison of This Method and Traditional PID Scheme

[0182] Furthermore, on day 180 of operation, one device exhibited slow drift due to aging of the piezoelectric material. The system's drift detection module promptly captured this trend and automatically performed a safe frequency sweep to fine-tune and update the LSTM model. After the model update was uploaded and aggregated through federated learning, the PPO controllers of other devices in the same batch gained "pre-immunity" to similar aging, improving response speed by 3 times and ensuring high stability of the degassing injection pressure and long-term reliability of chromatographic analysis data throughout the entire lifecycle.

[0183] This embodiment uses the specific scenario of "jet degassing" to fully demonstrate how the present invention upgrades the traditional single mapping of "deviation-voltage" to a three-level intelligent control closed loop of "feedforward inverse model prediction + feedback lifespan perception fine-tuning + online self-healing collaborative correction". While achieving extreme accuracy, it fundamentally solves the engineering pain points of actuator lifespan loss and long-term maintenance dependence.

[0184] Example 3

[0185] A jet diffusion type oil-gas separation method for a transformer oil online monitoring jet diffusion type oil-gas separation device includes the following steps:

[0186] The process for determining the temperatures of S0, the degassing chamber, and the injection cylinder specifically includes:

[0187] S01. Start the oil-gas separation. First, determine whether the temperature of the degassing chamber and the injection cylinder has reached the preset value. If it has reached the preset value, proceed to the next step. Otherwise, report a heating fault in the degassing chamber and the injection cylinder.

[0188] S1. Initialize the self-test process, which includes:

[0189] S11. Initialize and close all valves and motors;

[0190] S12. Initialize the motor position. The stepper motor moves forward to the right limit. Open the three-way solenoid valves V9 and V12. After the injection cylinder resets, close the three-way solenoid valves V9 and V12. Open the solenoid valves V5 and V4. The stepper motor moves slowly in the reverse direction until the buffer chamber level sensor L2 detects the level signal. Close the stepper motor and solenoid valves V5 and V4. Open the solenoid valve V2. After the stepper motor moves slowly in the reverse direction to the left limit, close the stepper motor and solenoid valve V2.

[0191] S2, the gas path cleaning process specifically includes:

[0192] S21. Open solenoid valves V8, V10, V6, and V4. After purging for a preset time, close solenoid valve V8. Wait for the pressure in the buffer chamber to drop to normal pressure, then close solenoid valves V10, V6, and V4 to complete the cleaning of the gas path, metering tube, and buffer tube.

[0193] S3. The airtightness inspection process specifically includes:

[0194] S31. Control the stepper motor to move forward from the left limit to the right limit, read the initial value of the current pressure sensor P1, and read the secondary value of the pressure sensor P1 after a preset time. Determine whether the change in pressure sensor P1 is less than the preset threshold. If yes, it is determined that the degassing chamber is airtight and the next action is executed. If no, a warning signal is issued to indicate that the degassing chamber is airtight. Replace the degassing chamber sealing ring in time and stop working.

[0195] S4, the oil circulation and metering process, is controlled by a stepper motor that drives the piston in the degassing chamber to move left and right. This, in conjunction with solenoid valves V1 and V2, pressure sensor P1, limit sensor, and level sensor, completes the circulation and metering control of the oil sample. Specifically, this includes:

[0196] S41. Open solenoid valve V1 to control the stepper motor to pull the piston of the degassing chamber from the left limit to the right limit until the pressure sensor P1 is stable and unchanged, then close the stepper motor and solenoid valve V1.

[0197] S42. Open solenoid valve V2 to control the stepper motor to pull the piston of the degassing chamber from the right limit to the left limit until the pressure sensor P1 is stable and unchanged, then close the stepper motor and solenoid valve V2.

[0198] S43. Repeat steps S11 and S12 a preset number of times to complete the cleaning of the oil sample circulation and degassing chamber;

[0199] S44. Open solenoid valve V2 to control the stepper motor to push the degassing chamber piston from the right limit to the middle limit until the pressure sensor P1 is stable. Then close the stepper motor and solenoid valve V2 to complete the oil sample quantification process.

[0200] S5. In the jet diffusion oil-gas separation process, after the oil is metered, the stepper motor is controlled to transfer the transformer oil in the degassing chamber to the jet cylinder; the stepper motor is controlled to pull the piston to the right limit, creating a vacuum inside the degassing chamber; then, the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis backfeedforward compensation voltage after dual-valve output consistency calibration and the reinforcement learning fine-tuning voltage, and the jet cylinder is controlled by driving the proportional valve or pressure relief valve to quickly jet the transformer oil in the cylinder into the degassing chamber. During the jetting process, the dissolved gas in the oil is transferred from the liquid phase to the gas phase and accumulates at the top of the degassing chamber and in the buffer tube; after the jet degassing is completed, the piston position is adjusted by controlling the stepper motor to push the piston rod, transferring the sample gas accumulated in the degassing chamber to the buffer tube and the metering tube in the gas injection module, specifically including:

[0201] S51. The heating control module heats the oil in the degassing chamber to a preset temperature and keeps it constant. At the same time, during the heating period, the pressure in the degassing chamber is judged by the degassing chamber pressure sensor P1, and the piston is pulled by the stepper motor to control the pressure within a predetermined range.

[0202] S52, control the stepper motor to move to the left, and transfer the transformer oil in the degassing chamber to the injection cylinder through the V3 valve;

[0203] S53. Control the stepper motor to pull the piston to the right limit, creating a vacuum inside the degassing chamber;

[0204] S54. The driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis feedforward compensation voltage after the dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage. During the injection process, the three-way solenoid valve V9 and solenoid valve V12 are opened. The proportional valve or pressure relief valve is driven by the driving voltage to quickly inject the transformer oil in the injection cylinder into the degassing chamber. During the injection process, the dissolved gas in the oil is transferred from the liquid phase to the gas phase and accumulates at the top of the degassing chamber and in the buffer tube. After a preset time of settling, the concentration of components in the degassing gas and oil reaches a dynamic equilibrium.

[0205] After S55, injection, and balancing are completed, open solenoid valves V5 and V6. By controlling the stepper motor to push the piston rod to adjust the piston position, the sample gas accumulated in the degassing chamber is transferred to the buffer tube and the metering tube until the buffer tube level sensor L1 detects the level signal. Then, close the stepper motor and solenoid valves V5 and V6.

[0206] S6. Sample injection and analysis process: After degassing, the gas injection module is switched to injection mode, and the extracted gas is sent into the chromatographic column for gas separation. The concentration of each component in the gas sample is obtained through chromatographic analysis, specifically including:

[0207] S61. After the degassing process is completed, switch the gas injection module to the injection state so that the degassing gas enters the chromatographic analysis component and the concentration of each component in the gas sample is obtained by the chromatographic analysis component.

[0208] S62, return oil, control the stepper motor to push the piston to slowly move to the left limit, open the solenoid valve V2 until the pressure sensor P1 is stable, then close the stepper motor and solenoid valve V2.

[0209] The jet diffusion oil-gas separation method in this embodiment can complete headspace degassing without extreme vacuum conditions. By accelerating oil-gas separation through jetting, it achieves the highest degassing efficiency. Similar to conventional offline degassing methods, it does not need to consider the influence of vacuum degree and vacuum degassing rate on the analysis results. It has the advantages of simple structure and good repeatability.

[0210] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "screw connection," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0211] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transformer oil online monitoring jet diffusion type oil-gas separation device, characterized in that, include: A horizontally arranged degassing chamber (2) is provided with a piston inside the degassing chamber (2). The piston is connected to a stepper motor (3) on one side of the degassing chamber (2) through a screw and slider structure to drive the piston to move in the degassing chamber (2). The bottom side of the degassing chamber (2) is connected to the oil inlet of the solenoid valve V1 and the transformer (1) in sequence through an oil inlet pipe. The bottom side of the degassing chamber (2) is also connected to the oil return port of the solenoid valve V2 and the transformer (1) in sequence through a return oil pipe. A pressure sensor P1 is provided in the middle of one side of the degassing chamber (2). The bottom end of the buffer tube (4) is connected to the top side of the degassing chamber (2) through a pipeline and a solenoid valve V5. A liquid level sensor L2 and a liquid level sensor L1 are respectively installed on the top side and the bottom side of the buffer tube (4). The top end of the buffer tube (4) is connected to the solenoid valve V4 and the gas injection module (10) in sequence through a pipeline. A vertically arranged injection cylinder (9) is connected in sequence to a pressure sensor P3, a three-way solenoid valve V9, a pressure relief valve (14), a proportional valve (12), and a carrier gas interface (6) via a pipeline on one side of the top of the injection cylinder (9). The bottom of the injection cylinder (9) is connected in sequence to a solenoid valve V12 and a non-powered multi-stage injection nozzle (13) via a pipeline. The non-powered multi-stage injection nozzle (13) is located at one end of the top of the degassing chamber (2). The bottom of the injection cylinder (9) is also connected in sequence to a solenoid valve V3 and the bottom of the degassing chamber (2) via a pipeline. The gas injection module (10) consists of a standard gas interface (5), a carrier gas interface (6), a quantitative tube (7), a chromatographic column interface (8), a pressure sensor P2, a solenoid valve V6, a solenoid valve V8, a solenoid valve V10, a solenoid valve V11, and a three-way solenoid valve V7. The standard gas interface (5) is connected to the solenoid valve V11 through a pipeline. The carrier gas interface (6) is connected to the three-way solenoid valve V7 through a pipeline. The top end of the quantitative tube (7) is connected to the solenoid valve V8 and the solenoid valve V10 through a pipeline. The bottom end of the quantitative tube (7) is connected to the three-way solenoid valve V7, the pressure sensor P2, the solenoid valve V6, and the solenoid valve V11 through a pipeline. The chromatographic column interface (8) is connected to the three-way solenoid valve V7 and the solenoid valve V8 through a pipeline. The proportional valve (12) and the pressure relief valve (14) are both piezoelectric ceramic valves, and the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis backfeedforward compensation voltage after dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage.

2. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 1, characterized in that, The top of the degassing chamber (2) is provided with a first limit sensor, a second limit sensor and a third limit sensor from left to right, and a heating control module (11) is provided on the outside of the degassing chamber (2) and the outside of the injection cylinder (9).

3. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 1, characterized in that, The non-powered multi-stage spray nozzle (13) includes a base (1301), a nozzle (1302) is provided on one side of the base (1301), a first-stage contraction section (1303) is provided inside the nozzle (1302), and a second-stage contraction section (1304) is connected to the other end of the first-stage contraction section (1303). A vacuum suction chamber (1305) is formed between the nozzle (1302) and the substrate (1301), and one end of the vacuum suction chamber (1305) is connected to the secondary contraction section (1304), and the other end of the vacuum suction chamber (1305) is sequentially connected to the tertiary contraction section (1306), the throat section (1307) and the diffusion zone (1308) opened inside the substrate (1301); The base (1301) has a gas intake channel (1309) and a liquid intake channel (1310) connected to the vacuum intake chamber (1305) on one side of the top and one side of the bottom.

4. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 3, characterized in that, One end of the nozzle (1302) is a high-pressure liquid inlet and is connected to the solenoid valve V12; One end of the gas intake channel (1309) is connected to the upper space of the degassing chamber (2), and the other end of the gas intake channel (1309) is connected to the top of the vacuum intake chamber (1305). One end of the diffusion zone (1308) is a diffusion outlet and is connected to the upper space of the degassing chamber (2); One end of the liquid intake channel (1310) is connected to the lower space of the degassing chamber (2), and the other end of the liquid intake channel (1310) is connected to the bottom of the vacuum intake chamber (1305).

5. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 1, characterized in that, When the gas injection module (10) is in the degassing state, the carrier gas passes through the carrier gas interface (6) and then through the inlet of the three-way solenoid valve V7 to the outlet of the three-way solenoid valve V7, and then enters the chromatographic column through the chromatographic column interface (8); When the gas injection module (10) is in the injection state, the carrier gas passes through the carrier gas interface (6) and then sequentially through the three-way solenoid valve V7, the quantitative tube (7), and the solenoid valve V8 to the chromatographic column interface (8) and enters the chromatographic column; When the gas injection module (10) is in the standard gas state, the standard gas passes through the standard gas interface (5) and then through the solenoid valve V11, the quantitative tube (7), and the solenoid valve V10 in sequence before being discharged. When the gas injection module (10) is in the purging state, the carrier gas passes through the carrier gas interface (6) and then through the three-way solenoid valve V7 to the chromatographic column interface (8). Then, the carrier gas passes through the carrier gas interface (6) and then through the three-way solenoid valve V7, then through the solenoid valve V8 and the solenoid valve V10 before being discharged. Finally, the carrier gas passes through the carrier gas interface (6) and then through the three-way solenoid valve V7, then through the solenoid valve V8, the quantitative tube (7), the solenoid valve V6, and the solenoid valve V4 before being discharged.

6. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 1, characterized in that, The driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis feedforward compensation voltage (after dual-valve output consistency calibration compensation) and the reinforcement learning fine-tuning voltage, including: The pressure feedback signal and target pressure command of the controlled object are acquired in real time. The pressure deviation is calculated based on the pressure feedback signal and the target pressure. The target displacement required by the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve is calculated by combining the target pressure difference curve. The target displacement commands of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are respectively input into the corresponding pre-trained hysteresis inverse model to generate the feedforward compensation voltage of the corresponding valve. The reinforcement learning controller is optimized using a proximal policy. The state space is the hidden state vector of the pressure deviation and the hysteresis inverse model and the actual displacement of the corresponding piezoelectric ceramic valve. The output is the fine-tuning correction voltage of the corresponding valve. The feedforward compensation voltage and the fine-tuning correction voltage of the corresponding valve are superimposed to generate the initial drive voltages of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve, respectively. By utilizing the gain deviation coefficient and hysteresis deviation coefficient of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve, the initial driving voltage of the piezoelectric ceramic proportional valve and the piezoelectric ceramic pressure relief valve are corrected respectively to obtain the final driving voltage of the corresponding valve. Based on the direction of pressure deviation, the final drive voltage of the corresponding valve is applied to the corresponding piezoelectric ceramic valve to change the valve opening, thereby controlling the flow rate and pressure of the carrier gas.

7. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 6, characterized in that, The input of the hysteresis inverse model includes a target displacement sequence consisting of the target displacement commands of the past M time steps, and the first-order rate of change feature of the current target displacement command. The model is trained by applying a wideband excitation signal with multiple frequencies and amplitudes to the piezoelectric valve in the offline stage and collecting the mapping data between its input voltage and output displacement to learn the rate-dependent hysteresis nonlinearity of the piezoelectric ceramic and its inverse mapping.

8. The online monitoring jet diffusion type oil-gas separator for transformer oil according to claim 6, characterized in that, The expression for the reward function of the proximal policy optimization reinforcement learning controller is as follows: ； In the formula, R t Let e(t) be the total reward value at time t, and e(t) be the pressure deviation. Energy consumption weighting coefficient Let be the fine-tuning correction voltage at time t. For smoothness weighting coefficients, For degradation protection weighting coefficients, For degradation protection functions, Let t be the final driving voltage at time t.

9. A transformer oil online monitoring jet diffusion type oil-gas separation device according to claim 6, characterized in that, The step of selecting the final driving voltage of the corresponding valve and applying it to the corresponding piezoelectric ceramic valve according to the direction of pressure deviation includes: When the pressure deviation is positive, the valve opening of the piezoelectric ceramic pressure relief valve is adjusted by the final drive voltage of the piezoelectric ceramic pressure relief valve to increase the opening of the piezoelectric ceramic pressure relief valve. When the pressure deviation is negative, the valve opening of the piezoelectric ceramic proportional valve is adjusted by the final drive voltage of the piezoelectric ceramic proportional valve to increase the opening of the piezoelectric ceramic proportional valve. When the pressure deviation is zero, the valve opening of the electro-ceramic proportional valve and the piezoelectric ceramic pressure relief valve remains unchanged, and the update of the drive voltage signal is stopped.

10. A jet diffusion type oil-gas separation method for a transformer oil online monitoring jet diffusion type oil-gas separation device, characterized in that, Includes the following steps: Start the oil-gas separation process, and sequentially execute the temperature recognition, initial self-test, gas path cleaning and air tightness test processes. After the test is passed, execute the oil circulation and quantitative process. After the oil sample is quantified, the stepper motor is first controlled to transfer the transformer oil in the degassing chamber to the injection cylinder. Then, the stepper motor is controlled to pull the piston to the right limit, creating a vacuum inside the degassing chamber. Next, the driving voltage of the piezoelectric ceramic valve is determined by superimposing the hysteresis feedforward compensation voltage after the dual-valve output consistency calibration compensation and the reinforcement learning fine-tuning voltage. The injection cylinder is then controlled by driving the proportional valve or pressure relief valve to quickly inject the transformer oil in the chamber into the degassing chamber. During the injection process, the dissolved gas in the oil is transferred from the liquid phase to the gas phase and accumulates at the top of the degassing chamber and in the buffer tube. After the injection degassing is completed, the piston position is adjusted by controlling the stepper motor to push the piston rod, transferring the sample gas accumulated in the degassing chamber to the buffer tube and the quantitative tube in the gas injection module. After degassing is completed, the gas injection module is controlled to the injection state, and the degassed gas is sent into the chromatographic column for gas separation. The concentration of each component in the gas sample is obtained by chromatographic analysis.

Images