Vacuum radiation water vapor phase drying method

Abstract

The invention provides a vacuum radiation water gas-phase drying method, and belongs to the technical field of vacuum radiation water gas-phase drying. According to the vacuum radiation water gas-phase drying method, before water gas is adopted for drying, a tank body is preheated through radiation in advance, and the temperature difference between the tank body and a to-be-detected component is reduced; and according to conversion parameters among the water vapor, the pressure, the temperature and the radiation power in the tank body in the drying process of the water vapor, the corresponding radiation power and the air intake are adjusted. Firstly, by collecting the vacuum degree and the initial temperature of a tank body, the initial power gain value of a radiant panel is calculated for preheating; after the preheating temperature inflection point is reached, the system establishes an air inlet valve opening degree adjusting function by monitoring the temperature difference value of an iron core and a winding, and the water vapor concentration is calculated in combination with the pressure change rate. A positive active adjustment value is determined through the corresponding relation between the environment temperature and the water vapor concentration, meanwhile, a negative passive adjustment value is calculated based on the surface temperature and the water content, and a two-way control mechanism is formed. The system achieves automatic control of the process by comparing the negative passive adjustment value with a preset drying threshold value, and finally the purpose of uniform drying is achieved. The technical problems that in an existing vacuum radiation drying technology, long drying time is needed for achieving thorough drying, and rusting is caused are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vacuum radiation water vapor phase drying, and in particular relates to a vacuum radiation water vapor phase drying method. Background Art

[0002] Large electrical equipment, such as transformers, often requires insulation treatment during operation and maintenance. Drying is a critical step in ensuring the insulation performance of these equipment. Traditional drying methods primarily include hot air drying, vacuum drying, and a combination of the two. Hot air drying heats the air and utilizes the temperature difference between the hot air and the surface of the equipment to transfer heat, evaporating moisture and achieving drying. Vacuum drying involves applying a vacuum within a sealed container to lower the boiling point of the water and promote evaporation. These traditional processes are widely used in industry, playing a particularly important role in the maintenance and production of large transformers.

[0003] However, these traditional drying technologies have many limitations. During the hot air drying process, the heat transfer efficiency is low and the heat distribution is uneven, which can easily cause local overheating or underheating. Although vacuum drying can reduce the evaporation temperature of water, it is difficult to provide sufficient heat by relying solely on a vacuum environment, resulting in a long drying cycle. Although the combined process has improved the above problems to a certain extent, it still faces technical difficulties such as uneven temperature field distribution, low energy utilization efficiency, and poor process control accuracy. Especially when dealing with electrical equipment with complex structures, the temperature difference effect between components of different materials will lead to significant differences in the drying effect, and may even cause material damage or rust.

[0004] At present, the industry has tried to solve the above problems by improving the heating method, optimizing the vacuum system, increasing the temperature monitoring points, etc. However, these improvement measures can often only locally optimize a certain link and cannot fundamentally solve the problem of uneven temperature field distribution. Especially when dealing with large and complex equipment, traditional technology has difficulty in achieving precise control of the temperature difference between components of different materials, resulting in a large deviation in the temperature field distribution during the drying process, affecting the uniformity and stability of the drying effect. In other words, there is a technical problem in the prior art that the temperature field distribution cannot be accurately controlled during the vacuum radiation drying process, resulting in uneven drying effect. For example, a long drying time is required to complete the thorough drying of all positions. If the time is shortened, rust may occur. This technical problem has become a key bottleneck restricting the further development of the drying process. Summary of the Invention

[0005] In view of this, the present invention provides a vacuum radiation water vapor drying method, which can solve the technical problem in the existing vacuum radiation drying technology that a long drying time is required to achieve thorough drying.

[0006] The present invention is implemented as follows: The present invention provides a vacuum radiation water vapor phase drying method comprising the following steps: before starting drying with water vapor, preheating a tank body by radiation to reduce the temperature difference between the tank body and the component to be tested; collecting a vacuum value and an initial temperature value inside the tank body; calculating an initial power gain value of a radiation plate based on the initial temperature value; monitoring a temperature value of the component to be tested inside the tank body; collecting a core temperature value and a winding temperature value inside the tank body and calculating a temperature difference; establishing an intake valve opening adjustment function based on the temperature difference; controlling the intake valve opening and adjusting the output power of the radiation plate based on the intake valve opening adjustment value; According to the conversion parameters between the water vapor, pressure, temperature and radiation power in the tank body during the water vapor drying process, the corresponding radiation power and air intake are adjusted: the pressure value in the tank body is collected and the pressure change rate is calculated; a water vapor concentration calculation model is established and the water vapor concentration value is calculated; the ambient temperature value in the tank body is collected; a positive active adjustment value is calculated based on the ambient temperature value and the water vapor concentration value; the output power of the radiation plate is adjusted based on the positive active adjustment value; the surface temperature value and humidity value of the component to be tested are collected; a negative passive adjustment value is calculated; it is determined whether the negative passive adjustment value reaches the preset drying threshold value. When the negative passive adjustment value does not reach the preset drying threshold value, the steps of calculating the temperature difference to determining the negative passive adjustment value are repeated. When the negative passive adjustment value reaches the preset drying threshold value, the output power of the radiation plate is reduced and the air intake valve is closed.

[0007] Among them, the calculation step of the initial power gain value adopts a nonlinear mapping algorithm, which is specifically implemented through a three-layer BP neural network structure. The input layer of the three-layer BP neural network structure is the initial temperature value inside the tank body, the output layer is the output power of the radiation panel, the number of hidden layer nodes is 7, an S-type transfer function is used between the input layer and the hidden layer, and a linear transfer function is used between the hidden layer and the output layer. The training sample is 500 sets of historical operation data.

[0008] Among them, the step of establishing the intake valve opening adjustment function adopts a fuzzy adaptive control algorithm, takes the temperature difference as the input variable, and the intake valve opening adjustment value as the output variable. The domain of the input variable is 0°C to 100°C and is divided into 5 fuzzy subsets. The membership function adopts a combination of trapezoidal and triangular shapes. The domain of the output variable is 0 to 1, and the Mamdani reasoning method is adopted.

[0009] The water vapor concentration calculation model is established using the Kalman filter algorithm, with water vapor concentration as the state variable and pressure change rate as the observation variable. Both the system noise and the observation noise are assumed to be Gaussian white noise. The sampling period of the state estimation is 1 second, the system noise variance is 0.01, and the observation noise variance is 0.1.

[0010] Among them, the calculation step of the positive active adjustment value adopts a fuzzy neural network algorithm. The fuzzy neural network adopts a four-layer network structure, including an input layer, a fuzzy layer, a hidden layer and an output layer. The input layer contains two neurons: the ambient temperature value and the water vapor concentration value. The fuzzy layer fuzzifies the input variables and adopts a Gaussian function as the membership function.

[0011] Among them, the calculation step of the negative passive adjustment value adopts the support vector regression algorithm, normalizes the temperature value and the humidity value, selects the radial basis kernel function as the kernel function, sets the penalty factor to 100, sets the kernel function parameter to 0.1, and selects the optimal parameter through the cross-validation method.

[0012] Among them, the preset drying threshold is determined by the fuzzy C-means clustering algorithm, and cluster analysis is performed on the historical operation data. The number of clusters is set to 3, the Euclidean distance is used as the distance metric, the membership function adopts an exponential function, the number of iterations is 100, and the termination threshold is 0.001.

[0013] The step of collecting the temperature values ​​of the core and the winding inside the tank is achieved by temperature sensors installed on the surface of the core and the winding, and the temperature sensors are all PT100 platinum thermal resistors.

[0014] The preheating temperature inflection point value is set to 122° C. based on the experience of historical drying operation records, the positive active adjustment value ranges from 0.6 to 1.4, and the negative passive adjustment value ranges from 0.4 to 1.2.

[0015] Among them, when the negative passive adjustment value reaches the preset dryness threshold, the gradient descent method is used to reduce the output power of the radiation panel, and the power reduction rate is 2% per second. At the same time, the proportional integral control algorithm is used to control the closing speed of the intake valve, with a proportional coefficient of 0.6 and an integration time of 30 seconds.

[0016] Compared to existing technologies, the present invention provides a vacuum radiation water vapor drying method. By establishing a multi-parameter coordinated control system for radiation panel power, intake valve opening, and water vapor concentration, this method achieves precise regulation of the temperature field distribution during the drying process. This method innovatively incorporates the material's temperature differential characteristics into the control strategy. By real-time monitoring of the temperature difference between the core and winding, it establishes an adaptive intake valve opening adjustment mechanism based on the temperature difference, achieving a uniform temperature field distribution.

[0017] The present invention adopts a bidirectional control strategy that combines positive active adjustment with negative passive adjustment. The positive active adjustment value is calculated through the coupling function of ambient temperature and water vapor concentration, and the negative passive adjustment value is constructed based on the surface temperature and moisture content, forming a complete closed-loop control system. This innovative control method can not only accurately adjust the temperature distribution of each part, but also dynamically optimize the control parameters according to the actual drying effect, significantly improving the uniformity and stability of the drying process. In addition, the water vapor concentration calculation model introduced in the present invention realizes real-time monitoring of the drying environment through the pressure change rate, providing reliable data support for temperature field regulation.

[0018] The present invention successfully solves the technical problem of the existing vacuum radiation drying technology that a long drying time is required to achieve thorough drying. It also correspondingly avoids the problem of easy rusting in the existing technology if the drying time is shortened. The fundamental reason is that a complete set of multi-parameter collaborative control systems has been established. By incorporating key parameters such as temperature difference characteristics, water vapor concentration, and ambient temperature into the control system, all-round regulation of the drying process is achieved. Especially when dealing with large and complex equipment, this method can accurately identify and adjust the temperature difference between components of different materials to ensure the uniformity of the drying effect. At the same time, the application of the two-way control strategy enables the system to have adaptive adjustment capabilities, and can dynamically optimize the control parameters according to actual conditions, further improving the drying effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention.

[0020] Figure 2 This is a curve diagram of the error change of the BP neural network during the training process in Example 2.

[0021] Figure 3 This is a graph showing how the temperature of the component to be tested changes with time in Example 2.

[0022] Figure 4 This is a graph showing the temperature difference between the core and the winding over time in Example 2.

[0023] Figure 5 This is a graph showing an obvious linear correlation between the tank pressure and the water vapor concentration in Example 2. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] like Figure 1FIG. 1 is a flow chart of a vacuum radiation water vapor drying method provided by the present invention, and the method comprises the following steps: Before starting the drying process with water vapor, the tank is preheated by radiation to reduce the temperature difference between the tank and the part to be tested; including: S01, collecting the vacuum value and the initial temperature value inside the tank; S02. Calculating an initial power gain value of the radiation plate according to an initial temperature value inside the tank, and adjusting the output power of the radiation plate according to the initial power gain value; S03, monitoring the temperature value of the component to be tested in the tank until the temperature value of the component to be tested reaches the preheating temperature inflection point value; S04, collecting the temperature of the iron core in the tank body and the temperature of the winding in the tank body, and calculating the temperature difference between the temperature of the iron core in the tank body and the temperature of the winding in the tank body; S05. Establishing an intake valve opening adjustment function according to the temperature difference, and calculating an intake valve opening adjustment value according to the intake valve opening adjustment function; S06. Controlling the opening of the air intake valve according to the air intake valve opening adjustment value, and adjusting the output power of the radiation panel; According to the conversion parameters between the water vapor, pressure, temperature and radiation power in the tank during the drying process, the corresponding radiation power and air intake are adjusted, including: S07, collecting the pressure value inside the tank in real time, and calculating the pressure change rate according to the pressure value inside the tank; S08. Establishing a water vapor concentration calculation model according to the pressure change rate, and calculating a water vapor concentration value using the water vapor concentration calculation model; S09, collecting the ambient temperature value inside the tank; S10, calculating a positive active adjustment value according to a correspondence between the ambient temperature value in the tank and the water vapor concentration value; S11, adjusting the output power of the radiation panel according to the forward active adjustment value; S12, collecting the surface temperature value and the surface humidity value of the component to be measured; S13, calculating a negative passive adjustment value according to the surface temperature value and the surface humidity value of the component to be measured; S14, determining whether the negative passive adjustment value reaches a preset dryness threshold; S15. When the negative passive adjustment value does not reach the preset dryness threshold, repeat steps S04 to S14. When the negative passive adjustment value reaches the preset dryness threshold, reduce the output power of the radiation panel and close the air intake valve.

[0026] The initial power gain value is determined by a nonlinear mapping relationship between the initial temperature value in the tank and the output power of the radiation panel; The intake valve opening adjustment function is determined by an adaptive control relationship between the temperature difference and the intake valve opening adjustment value; The positive active adjustment value is determined by the coupling relationship between the ambient temperature value in the tank and the water vapor concentration value; The negative passive adjustment value is determined by a composite relationship between the surface temperature value of the component to be measured and the surface humidity value of the component to be measured; The water vapor concentration calculation model establishes a dynamic calculation relationship through the pressure change rate.

[0027] The specific implementation of the above steps is described in detail below. The specific implementation of step S01 is to collect the vacuum value and the initial temperature value of the tank body by using a vacuum sensor and a temperature sensor installed in the tank body. The vacuum sensor adopts a capacitive vacuum gauge with a measurement range of 0.1Pa to 100kPa and a measurement accuracy of ±0.5%. The temperature sensor adopts a PT100 platinum resistance thermometer. The collected data is transmitted to the controller through the data acquisition module to complete the real-time data acquisition and storage.

[0028] The specific implementation of step S02 is to calculate the initial power gain value of the radiation plate based on the initial temperature value in the tank body using a nonlinear mapping algorithm. First, a nonlinear mapping relationship between temperature and power is established, and a radial basis function neural network is used for training. The input layer is the initial temperature value in the tank body, and the output layer is the output power of the radiation plate. The hidden layer uses a Gaussian function as the activation function. The training sample uses historical operation data. After the training is completed, a nonlinear mapping model is obtained. The corresponding initial power gain value is calculated according to the current initial temperature value in the tank body, and the calculation result is output to the power controller to control the output power of the radiation plate. The value range of the initial power gain value is 0.8 to 1.2. Among them, the nonlinear mapping algorithm used to calculate the initial power gain value of the radiation panel is based on the BP neural network with a three-layer structure. The input layer is the initial temperature value inside the tank, the output layer is the output power of the radiation panel, and the number of hidden layer nodes is determined to be 7 by an empirical formula. The S-type transfer function is used between the input layer and the hidden layer, and the linear transfer function is used between the hidden layer and the output layer. The number of neurons in the input layer is the same as the dimension of the input variable. The hidden layer uses the Gaussian function as the basis function. The number of hidden layer nodes is determined by the orthogonal least squares method. The output layer is a linear weight combination. The network parameters include the basis function center, expansion constant and network weight. The basis function center is determined by the K-means clustering algorithm, the expansion constant is determined by the P nearest neighbor method, and the network weight is calculated by the least squares method. The training process adopts an adaptive adjustment mechanism to dynamically adjust the learning parameters according to the network output error. The training sample is 500 sets of historical operation data. The Levenberg-Marquardt algorithm is used for training, the learning rate is adaptively adjusted, and the training target error is 0.0001.

[0029] The specific implementation method of step S03 is to monitor the temperature value of the component to be tested in real time through a temperature sensor installed on the surface of the component to be tested. The temperature sensor adopts a K-type thermocouple with a measurement range of 0°C to 1200°C and a measurement accuracy of ±1°C. The collected temperature value is compared with the preset preheating temperature inflection point value. When the temperature value of the component to be tested reaches the preheating temperature inflection point value, the next operation is triggered, wherein the preheating temperature inflection point value is set to 122°C. When the temperature exceeds this value, the surface of the component to be tested is prone to rust.

[0030] The specific implementation of step S04 is to collect the temperature value of the iron core and the winding inside the tank body respectively through temperature sensors installed on the surface of the iron core and the winding. The temperature sensors all use PT100 platinum thermal resistors. The difference between the two collected temperature values ​​is calculated to obtain the temperature difference. The temperature difference is calculated using a real-time data processing algorithm. The temperature data is smoothed by a sliding window method. The window length is 10 sampling points and the sampling interval is 1 second to reduce the influence of measurement noise.

[0031] The specific implementation method of step S05 is to establish an intake valve opening adjustment function based on the temperature difference using a fuzzy adaptive control algorithm. Specifically, the temperature difference is used as the input variable and the intake valve opening adjustment value is used as the output variable. The domain of the input variable is 0°C to 100°C, which is divided into 5 fuzzy subsets, represented by NB, NS, ZO, PS, and PB respectively. The membership function adopts a combination of trapezoidal and triangular shapes. The domain of the output variable is 0 to 1, which is also divided into 5 fuzzy subsets. The fuzzy rules adopt 25 rules summarized by expert experience, and the Mamdani reasoning method and the center of gravity method are used for defuzzification. At the same time, an adaptive mechanism is introduced to adjust the parameters of the membership function online according to the system response, with an adjustment step size of 0.05 to obtain the intake valve opening adjustment value, wherein the value range of the intake valve opening adjustment value is 0.2 to 0.8.

[0032] The specific implementation method of step S06 is to output the calculated intake valve opening adjustment value to the actuator to control the opening of the intake valve. At the same time, according to the change of the intake valve opening, the proportional integral differential control algorithm is used to adjust the output power of the radiation panel, where the proportional coefficient is 0.8, the integral time is 60 seconds, and the differential time is 10 seconds. The stable operation of the system is achieved through closed-loop control.

[0033] The specific implementation method of step S07 is to collect the pressure value inside the tank in real time through a pressure sensor installed in the tank. The pressure sensor adopts a capacitive pressure transmitter with a measurement range of 0Pa to 100kPa and a measurement accuracy of ±0.1%. According to the collected pressure value, the pressure change rate is calculated by the least squares method. The calculation window length is 60 seconds and the sampling interval is 1 second. The pressure change trend over time is obtained by linear fitting, and the pressure change rate is calculated.

[0034] The specific implementation method of step S08 is to establish a water vapor concentration calculation model based on the pressure change rate using the Kalman filter algorithm. First, the state equation and observation equation are established, the state variable is the water vapor concentration, the observation variable is the pressure change rate, and the system noise and observation noise are both assumed to be Gaussian white noise. The state is estimated by recursion to obtain the water vapor concentration value, where the sampling period of the state estimation is 1 second, the system noise variance is 0.01, the observation noise variance is 0.1, and the initial state estimation error variance is 1.

[0035] The specific implementation method of step S09 is to collect the ambient temperature value inside the tank by means of a temperature and humidity sensor installed in the tank. The temperature and humidity sensor adopts a digital thermometer and hygrometer with a temperature measurement range of 50°C to 150°C and a measurement accuracy of ±0.5°C. The humidity measurement range is 0% to 100% RH and a measurement accuracy of ±2% RH. The sampling period is 1 second, and the collected data is transmitted to the controller through the data acquisition module.

[0036] The specific implementation method of step S10 is based on the ambient temperature value and the water vapor concentration value in the tank body, and the grey correlation analysis method is used to establish the coupling relationship between the two. First, the ambient temperature value and the water vapor concentration value are dimensionlessly processed, and the correlation coefficient is calculated. The correlation threshold is set to 0.85. When the correlation is greater than the threshold, the fuzzy neural network algorithm is used to calculate the positive active adjustment value. The input layer is the ambient temperature value and the water vapor concentration value, the output layer is the positive active adjustment value, the hidden layer uses the hyperbolic tangent function as the activation function, and the training sample uses historical data. The value range of the positive active adjustment value is 0.6 to 1.4.

[0037] Specifically, in the step of calculating the positive active adjustment value using the fuzzy neural network algorithm, a four-layer network structure is adopted: input layer, fuzzy layer, hidden layer and output layer. The input layer contains 2 neurons, corresponding to the ambient temperature value and the water vapor concentration value respectively. The fuzzy layer fuzzifies the input variables and uses the Gaussian function as the membership function. Three membership functions are set for each input variable, representing low, medium and high levels respectively. The hidden layer uses the hyperbolic tangent function as the activation function, the number of neurons is 8, the output layer is the positive active adjustment value, and the BP algorithm is used for network training. The learning rate is set to 0.01, the momentum factor is set to 0.9, and the training is stopped when the training error is less than 0.001. After the training is completed, the calculation model of the positive active adjustment value is obtained.

[0038] The specific implementation method of step S11 is to output the calculated positive active adjustment value to the power controller, and use the adaptive PID control algorithm to adjust the output power of the radiation panel, wherein the proportional coefficient, integral time and differential time are all adjusted online through fuzzy rules, the proportional coefficient has a value range of 0.5 to 1.5, the integral time has a value range of 30 seconds to 90 seconds, and the differential time has a value range of 5 seconds to 15 seconds, thereby improving the control performance of the system through adaptive adjustment.

[0039] The specific implementation method of step S12 is to collect the surface temperature value and surface humidity value of the component to be tested respectively by a temperature sensor and a humidity sensor installed on the surface of the component to be tested. The temperature sensor adopts a K-type thermocouple with a measuring range of 0°C to 1200°C and a measuring accuracy of ±1°C. The humidity sensor adopts a capacitive hygrometer with a measuring range of 0% to 100% RH and a measuring accuracy of ±1% RH. The sampling period is 1 second. Real-time data collection and storage are realized through the data acquisition module.

[0040] The specific implementation method of step S13 is based on the surface temperature value and surface humidity value of the component to be measured, and a support vector regression algorithm is used to establish a composite relationship between the two. First, the temperature value and humidity value are normalized, and the radial basis kernel function is selected as the kernel function. The penalty factor is set to 100, and the kernel function parameter is set to 0.1. The optimal parameters are selected through the cross-validation method, a regression model is established, and the negative passive adjustment value is calculated, where the value range of the negative passive adjustment value is 0.4 to 1.2.

[0041] The specific implementation of step S14 is to compare the calculated negative passive adjustment value with a preset drying threshold. The preset drying threshold is determined using a fuzzy clustering algorithm, specifically based on the fuzzy C-means clustering algorithm. First, historical operating data, including the negative passive adjustment value and the corresponding drying effect evaluation index, is collected. The data is normalized, the number of clusters is set to 3, the cluster centers are randomly initialized, the distances from the sample points to the cluster centers are calculated, and the Euclidean distance is used as the distance metric. The membership matrix is ​​calculated based on the distances, using an exponential membership function. The cluster centers are updated, and the calculation is repeated until the change in the membership matrix is ​​less than 0.001 or the maximum number of iterations, 100, is reached. Finally, the optimal cluster center is selected as the preset drying threshold. The preset drying threshold is set to 0.85.

[0042] The specific implementation of step S15 is that when the negative passive adjustment value does not reach the preset dryness threshold, steps S04 to S14 are repeated, achieving continuous system operation through closed-loop control. When the negative passive adjustment value reaches the preset dryness threshold, the output power of the radiation panel is reduced using a gradient descent method at a power reduction rate of 2% per second until it reaches the minimum power. Simultaneously, a proportional-integral control algorithm is used to control the closing speed of the intake valve, with a proportional coefficient of 0.6 and an integration time of 30 seconds, to achieve a smooth shutdown of the system. The specific implementation of using the proportional-integral control algorithm to control the closing speed of the intake valve is based on an improvement of the traditional PID control algorithm, retaining only the proportional and integral components and removing the differential component to prevent system oscillation. The controller input is the deviation between the valve opening and the target opening, and the output is a control signal for the valve actuator. The proportional coefficient is adjusted using the step response method and is set to 0.6. The integration time is adjusted using the quarter-decay method and is set to 30 seconds. The sampling period is 1 second. The integral limit is ±20% of the output range, and a trapezoidal integration algorithm is used to reduce calculation errors.

[0043] The mathematical model or calculation process involved in the present invention is described in detail as follows: 1. Mathematical expression for calculating the initial power gain value: The calculation model of the initial power gain value is specifically expressed as follows: ; Where, Output power for the radiation panel; is the output layer weight; is the weight from the input layer to the hidden layer; is the input temperature feature; is the hidden layer bias; is the output layer bias; is an S-type transfer function, expressed as ; is the input feature dimension.

[0044] 2. Mathematical expression of the intake valve opening adjustment function: The intake valve opening adjustment function is specifically expressed as follows: ; ; Where, is the valve opening adjustment value; is the temperature difference; For the The membership function of fuzzy subsets; is the central value of the output fuzzy subset; is the parameter of the membership function.

[0045] 3. Mathematical expression of water vapor concentration calculation model: The water vapor concentration state estimation equation is specifically expressed as follows: ; ; Where, is the water vapor concentration state vector; is the pressure observation vector; is the state transfer matrix; is the control matrix; is the observation matrix; are system noise and observation noise respectively, satisfying Gaussian distribution and .

[0046] 4. The calculation model of the positive active adjustment value is specifically expressed as follows: ; ; Where, It is the positive active adjustment value; is the network weight; is the Gaussian basis function; is the fuzzy membership function; are input variables, including ambient temperature and water vapor concentration; is the basis function center; is the basis function width.

[0047] 5. The calculation model for negative passive adjustment value is specifically expressed as follows: ; ; Where, Negative passive adjustment value; is the Lagrange multiplier; is the kernel function; is the input feature vector, including surface temperature and moisture content; is the kernel function parameter; is the bias term.

[0048] 6. The pressure change rate calculation model is specifically expressed as follows: ; ; ; Where, is the pressure change rate; is the time sampling point; is the pressure sampling value; is the number of sampling points; is the time average; is the average pressure.

[0049] 7. The fuzzy clustering model with preset drying threshold is specifically expressed as follows: ; ; ; Where, is the objective function; For samples Cluster centers The degree of membership; is the fuzzy index, the value is 2; is the number of clusters, which is 3; is the sample size; is the input sample; is the cluster center.

[0050] The steps for deriving or establishing each formula or equation are described in detail below.

[0051] 1. Derivation of the initial power gain value calculation model: First, establish the mapping matrix from initial temperature to power gain: ; Where, Indicates the The first sample temperature characteristics.

[0052] The weight matrix is ​​expressed as: ; Where, Represents the connection weight from the input layer to the hidden layer.

[0053] The optimization objective function is: ; Where, For the The predicted output power of samples, is the actual output power.

[0054] Update the weights through the back-propagation algorithm: ; ; Where, is the learning rate, and its value is 0.01.

[0055] 2. Derivation of intake valve opening adjustment function: First, construct the fuzzy rule matrix: ; Where, Indicates the The input fuzzy subset and The correlation degree of the output fuzzy subsets.

[0056] The adaptive adjustment process is expressed as: ; ; ; Where, is the adjustment amount of the membership function parameter, calculated by the gradient descent method: ; ; ; in, To adjust the step size, take the value as 0.05. is the performance indicator function.

[0057] 3. Derivation of water vapor concentration calculation model: Kalman filter prediction and update process: Prediction process: ; ; Update process: ; ; ; Where, is the Kalman gain matrix, is the state estimation error covariance matrix, Is the identity matrix. The initial state is set as: , .

[0058] 4. Derivation of the positive active adjustment value calculation model: The conversion matrix from the input layer to the fuzzy layer of the fuzzy neural network: ; Where, Indicates the The input variables are The membership degree of a fuzzy set.

[0059] The basis function center vector is expressed as: ; Where, For the The center point of the Gaussian basis function.

[0060] Network optimization objective function: ; Where, is the actual output value, is the predicted output value, is the regularization parameter, and its value is 0.001.

[0061] Parameter update equation: ; ; ; Where, is the learning rate, with an initial value of 0.1 and is adjusted adaptively.

[0062] 5. Derivation of the negative passive adjustment value calculation model: The optimization problem of support vector regression is expressed as: ; Constraints: ; ; ; The dual problem is transformed into: ; Constraints: ; ; Where, is the allowable error, the value is 0.01, is the penalty factor, determined by cross-validation.

[0063] 6. Derivation of the pressure change rate calculation model: Construct the pressure sampling data matrix: ; Where, For the The pressure value of each sampling point, is the corresponding time value.

[0064] Least squares optimization objective: ; Where, is the pressure change rate, is the intercept term.

[0065] 7. Derivation of fuzzy clustering model with preset drying threshold: The sample matrix is ​​expressed as: ; Where, Indicates the The first sample Dimensional features.

[0066] The membership matrix is ​​expressed as: ; Where, Indicates the Sample pair The membership degree of each cluster center.

[0067] The iterative optimization process includes: Step 1: Initialize the membership matrix ; Step 2: Calculate the cluster center; Step 3: Update the membership matrix; Step 4: Calculate the objective function value; Step 5: Determine whether the termination condition is met. If not, return to step 2.

[0068] Specifically, the core technical principles of this invention are based on heat transfer, phase change mass transfer, and multi-parameter coordinated control. During the preheating phase, the system achieves uniform heating of the equipment by precisely controlling the power of the radiant panels. Compared to traditional hot air drying, radiant heating offers the advantages of high heat transfer efficiency and uniform temperature distribution. By calculating the initial power gain value, the system can establish an optimal power output curve based on the initial temperature, ensuring a uniform temperature rise across all parts during the preheating process.

[0069] During the drying phase, the present invention innovatively utilizes the material temperature differential effect. Due to differences in thermophysical properties, components made of different materials will experience temperature differences during the heating process. While traditional technologies view this temperature difference as a disadvantage, the present invention transforms it into a control parameter, establishing an intake valve opening adjustment function by real-time monitoring of the temperature difference between the core and windings. This temperature-difference-based control strategy automatically adjusts the intake volume based on material properties, achieving precise control of the local temperature field. Furthermore, a water vapor concentration calculation model established through the pressure change rate provides an additional control dimension for temperature field regulation.

[0070] The bidirectional control strategy employed in this invention is key to achieving precise control. A positive active adjustment value, calculated from a coupled function of ambient temperature and water vapor concentration, is used to actively adjust system parameters. A negative passive adjustment value, based on real-time monitoring of surface temperature and moisture content, reflects the actual drying effect. This bidirectional adjustment mechanism ensures that the control system can simultaneously account for both process parameters and drying effect, forming a complete closed-loop control system. By continuously comparing the negative passive adjustment value with a preset drying threshold, the system achieves automatic control and optimal regulation of the drying process.

[0071] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0072] The specific implementation method of step S01 is to realize real-time acquisition of vacuum degree and temperature through a sensor array arranged in the tank body, wherein the vacuum degree sensor adopts a capacitive vacuum gauge with a measurement range of 0.1Pa to 100kPa and a measurement accuracy of ±0.5%. The temperature sensor adopts a PT100 platinum thermal resistor, and the sampling period is set to 1 second. The data acquisition module adopts 16-bit AD conversion accuracy, and the collected data is transmitted to the central controller in real time for storage and processing through industrial Ethernet. The data storage adopts a circular buffer mechanism, and the buffer size is 1024 data points.

[0073] The specific implementation of step S02 is to establish a nonlinear mapping relationship between the initial temperature value and the radiation panel power based on the BP neural network. The calculation model of the initial power gain value is specifically expressed as: , where is the output power of the radiation panel, is the output layer weight, is the weight from input layer to hidden layer, is the input temperature characteristic, is the hidden layer bias, is the output layer bias, is an S-type transfer function, expressed as , is the input feature dimension; first construct the temperature feature matrix: , where Indicates the The first sample temperature features; then construct the weight matrix: , where Represents the connection weight from the input layer to the hidden layer; the optimization objective function is: , where For the The predicted output power of samples, is the actual output power; the back propagation algorithm is used to update the weights: , , where is the learning rate, which is set to 0.01. The number of training samples is 500. When the training error is less than 0.001, the training is stopped to obtain the calculation model of the initial power gain value.

[0074] The specific implementation method of step S03 is to monitor the temperature value of the component to be tested in real time through a K-type thermocouple arranged on the surface of the component to be tested. The measurement range is 0°C to 1200°C, the measurement accuracy is ±1°C, the sampling period is 1 second, and the temperature data is smoothed by a sliding average algorithm. The sliding window length is 10 sampling points. The average temperature value is calculated in real time and compared with the preset preheating temperature inflection point value of 122°C. When the average temperature value reaches the preheating temperature inflection point value, the next operation is triggered.

[0075] The specific implementation method of step S04 is to collect temperature data through PT100 platinum thermal resistors arranged on the surface of the iron core and the winding, with a sampling period of 1 second, and calculate the temperature difference using a real-time data processing algorithm. First, the collected temperature data is digitally filtered using a 5th-order Butterworth low-pass filter with a cutoff frequency of 0.5 Hz. Then, the temperature data is smoothed using a sliding window method with a window length of 10 sampling points, and the difference between the core temperature value and the winding temperature value is calculated.

[0076] The specific implementation of step S05 is to establish an intake valve opening adjustment function based on a fuzzy adaptive control algorithm. The specific calculation model is: , , where is the valve opening adjustment value, is the temperature difference, For the The membership function of fuzzy subsets is is the central value of the output fuzzy subset, is the parameter of the membership function; first construct the fuzzy rule matrix: , where Indicates the The input fuzzy subset and The correlation degree of the output fuzzy subsets is calculated by using an adaptive adjustment process: , , , where is the adjustment amount of the membership function parameter, calculated by the gradient descent method: , , ,in, To adjust the step size, take the value as 0.05. is the performance indicator function.

[0077] The specific implementation of step S06 is to control the intake valve opening based on the intake valve opening adjustment value, and at the same time use the proportional integral differential control algorithm to adjust the output power of the radiation panel, wherein the proportional coefficient is 0.8, the integral time is 60 seconds, and the differential time is 10 seconds. The controller uses an incremental PID control algorithm, and the control output increment calculation formula is: , where is the control quantity increment, is the current deviation, is the last deviation, is the previous deviation, is the proportionality coefficient, is the integration coefficient, is the differential coefficient; stable operation of the system is achieved through closed-loop control.

[0078] The specific implementation of step S07 is to collect pressure values ​​in real time through a capacitive pressure transmitter arranged in the tank, with a measurement range of 0Pa to 100kPa, a measurement accuracy of ±0.1%, and a sampling period of 1 second. A pressure sampling data matrix is ​​constructed based on the collected pressure data: , where For the The pressure value of each sampling point, is the corresponding time value; the pressure change rate is calculated using the least squares method, and the calculation model is: , , , where is the pressure change rate, is the number of sampling points, is the time average, is the average pressure.

[0079] The specific implementation of step S08 is to establish a water vapor concentration calculation model based on the Kalman filter algorithm, and the state equation and observation equation are: , , where is the water vapor concentration state vector, is the pressure observation vector, is the state transition matrix, is the control matrix, is the observation matrix, are system noise and observation noise respectively; the prediction and update process of Kalman filter includes: prediction process: , ;Update process: , , , where is the Kalman gain matrix, is the state estimation error covariance matrix, is the identity matrix.

[0080] The specific implementation method of step S09 is to collect ambient temperature data through a digital thermometer and hygrometer arranged in the tank body. The temperature measurement range is 50°C to 150°C, the measurement accuracy is ±0.5°C, the humidity measurement range is 0% to 100%RH, the measurement accuracy is ±2%RH, the sampling period is 1 second, and a digital low-pass filtering algorithm is used to filter the collected temperature data. The filter cutoff frequency is 1Hz, the filter order is 3rd order, and a stable ambient temperature value is obtained.

[0081] The specific implementation of step S10 is to establish a coupling relationship between the ambient temperature value and the water vapor concentration value based on the fuzzy neural network algorithm to calculate the positive active adjustment value. The calculation model is: , , where is the positive active adjustment value, is the network weight, is the Gaussian basis function, is the fuzzy membership function, is the input variable, is the basis function center, is the basis function width; construct the conversion matrix from the input layer to the fuzzy layer: , where Indicates the The input variables are The membership degree of a fuzzy set; the basis function center vector is expressed as: , where For the The center point of the Gaussian basis function.

[0082] The specific implementation of step S11 is to use an adaptive PID control algorithm to adjust the output power of the radiation panel based on the positive active adjustment value, and the controller parameters are adjusted in real time through fuzzy rules, specifically: , , , where , , are the proportional, integral and differential coefficients at the current moment, , , is the parameter adjustment amount; the controller output is: , where To control the output, To control deviation.

[0083] The specific implementation method of step S12 is to collect surface temperature and surface humidity values ​​respectively by arranging a K-type thermocouple and a capacitive hygrometer on the surface of the component to be tested. The measurement range of the temperature sensor is 0°C to 1200°C, and the measurement accuracy is ±1°C. The measurement range of the humidity sensor is 0% to 100% RH, and the measurement accuracy is ±1% RH. The sampling period is 1 second. The collected data is processed by a Butterworth digital filter. The filter order is 4th order and the cutoff frequency is 2Hz. The filtered data is transmitted to the controller through the data acquisition module for storage and processing.

[0084] The specific implementation of step S13 is to establish a composite relationship between the surface temperature value and the surface humidity value based on the support vector regression algorithm, and calculate the negative passive adjustment value. The optimization problem is expressed as: , constraints: , , ; The dual problem is transformed into: , constraints: , ; Solve to get the negative passive adjustment value calculation formula: , , where is the allowable error, the value is 0.01, is the penalty factor, is the kernel function parameter.

[0085] The specific implementation of step S14 is to use the fuzzy C-means clustering algorithm to determine the preset dryness threshold, and the sample matrix is ​​expressed as: , where Indicates the The first sample dimensional features; the membership matrix is ​​expressed as: , where Indicates the Sample pair The membership degree of cluster centers; the objective function is: , update formula: , , where is the fuzzy index, and its value is 2.

[0086] The specific implementation of step S15 is that when the negative passive adjustment value does not reach the preset dryness threshold, steps S04 to S14 are repeated. When the negative passive adjustment value reaches the preset dryness threshold, the output power of the radiation panel is reduced by a gradient descent method, and the power reduction rate is 2% per second. The power adjustment formula is: , where is the current power value, is the power value at the previous moment; at the same time, the proportional integral control algorithm is used to control the closing speed of the intake valve, and the controller output is: , where is the proportional coefficient, which is 0.6. is the integration coefficient, the integration time is 30 seconds, is the valve position deviation.

[0087] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: In a certain power equipment maintenance center, a 110kV transformer is dried after overhaul. The rated capacity of the transformer is 50MVA, the core weight is 15 tons, and the winding weight is 8 tons. It needs to be treated using a vacuum radiation water vapor phase drying method. The specific implementation process is described as follows.

[0088] First, collect the vacuum degree and initial temperature data inside the tank. The initial sampling data of vacuum degree and temperature are shown in Table 1: Table 1 Vacuum degree and temperature sampling data

[0089] Based on the collected initial temperature data, the BP neural network is used to calculate the initial power gain value of the radiation panel. The network training data is shown in Table 2: Table 2 Neural network training data table

[0090] like Figure 2 The figure shows the error curve of the BP neural network during training, showing a good convergence trend for both the training and validation errors. The power gain value obtained after neural network training was 1.15, and the initial output power of the radiation panel was set to 45kW based on this value. During the drying process, temperature changes were monitored by temperature sensors placed on the surface of the test component. The temperature change trend is shown in Table 3: Table 3 Surface temperature change data of the tested components

[0091] like Figure 3 As shown in the figure, the temperature of the component under test shows a nonlinear upward trend over time and tends to be flat when approaching the preheating temperature inflection point. When the surface temperature of the component under test reaches 120.6℃, close to the preheating temperature inflection point of 122℃, the temperature data of the core and windings are collected, and the temperature sampling data are shown in Table 4: Table 4 Core and winding temperature sampling data

[0092] like Figure 4 As shown in Figure 5, the temperature difference between the core and the winding changes with time, reflecting the heat conduction process between the two. The intake valve opening adjustment function is established based on the temperature difference, and the fuzzy rule base is shown in Table 5: Table 5 Fuzzy rule data table

[0093] The opening of the intake valve is adjusted according to the calculated intake valve opening adjustment value, and the pressure data in the tank is collected in real time. The pressure change data is shown in Table 6: Table 6 Tank pressure change data

[0094] The Kalman filter algorithm is used to establish the water vapor concentration calculation model, and the initial parameters are set as follows: system noise variance , observation noise variance , the initial state estimate , initial estimate error variance .like Figure 5 As shown in the figure, there is an obvious linear correlation between the tank pressure and the water vapor concentration, and the actual measured values ​​are basically consistent with the theoretical predicted values.

[0095] The ambient temperature data inside the tank is collected, and the positive active adjustment value is calculated based on the ambient temperature value and the water vapor concentration value. The adjustment results are shown in Table 7: Table 7 Calculation results of positive active adjustment value

[0096] Finally, the surface temperature and moisture content data of the tested component are collected and the negative passive adjustment value is calculated. The calculation results are shown in Table 8: Table 8 Negative passive adjustment value calculation results

[0097] Traditional transformer drying methods mainly use hot air circulation drying or vacuum drying. These methods have problems such as long drying time, high energy consumption, and poor drying uniformity. The vacuum radiation water vapor phase drying method proposed in the present invention achieves precise control of the drying process by introducing algorithms such as fuzzy adaptive control and Kalman filtering. After comparative tests, the drying time was shortened from the traditional 72 hours to 48 hours using the method of the present invention, energy consumption was reduced by about 25%, insulation resistance increased by 2.5 times after drying, and breakdown voltage increased by 1.8 times, significantly improving the drying effect and efficiency. In addition, the method of the present invention also has adaptive adjustment capabilities, and can dynamically adjust drying parameters according to real-time monitoring data, effectively avoiding the problems of overheating and uneven drying, and providing a more reliable technical guarantee for the repair and maintenance of large transformers.

[0098] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 9 below.

[0099] Table 9 Variable Explanation Table

[0100] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A vacuum radiation water vapor phase drying method, characterized in that: The following steps are involved: Before starting to dry with water vapor, preheat the tank by radiation to reduce the temperature difference between the tank and the part to be tested; According to the conversion parameters among the water vapor, pressure, temperature and radiation power in the tank during the drying process of the water vapor, the corresponding radiation power and air intake are adjusted.

2. The vacuum radiation water vapor phase drying method according to claim 1, characterized in that: The method of preheating the tank body by radiation in advance to reduce the temperature difference between the tank body and the component to be tested includes: collecting the vacuum value and the initial temperature value in the tank body; calculating the initial power gain value of the radiation plate according to the initial temperature value; monitoring the temperature value of the component to be tested containing the iron core and the winding in the tank body; collecting the temperature value of the iron core and the winding in the tank body and calculating the temperature difference; establishing an intake valve opening adjustment function according to the temperature difference; controlling the intake valve opening according to the intake valve opening adjustment value and adjusting the output power of the radiation plate of the preheated tank body; The adjustment of the corresponding radiation power and air intake includes: collecting the pressure value inside the tank body and calculating the pressure change rate; establishing a water vapor concentration calculation model and calculating the water vapor concentration value; collecting the ambient temperature value inside the tank body; calculating the positive active adjustment value according to the ambient temperature value and the water vapor concentration value; adjusting the output power of the radiation plate according to the positive active adjustment value; collecting the surface temperature value and humidity value of the component to be tested; calculating the negative passive adjustment value; judging whether the negative passive adjustment value reaches a preset dryness threshold, when the negative passive adjustment value does not reach the preset dryness threshold, repeating the steps of calculating the temperature difference to judging the negative passive adjustment value, when the negative passive adjustment value reaches the preset dryness threshold, reducing the output power of the radiation plate and closing the air intake valve.

3. The vacuum radiation water vapor phase drying method according to claim 2, characterized in that: The calculation step of the initial power gain value adopts a nonlinear mapping algorithm, which is specifically implemented through a three-layer BP neural network structure. The input layer of the three-layer BP neural network structure is the initial temperature value in the tank body, the output layer is the output power of the radiation panel, the number of hidden layer nodes is 7, an S-type transfer function is used between the input layer and the hidden layer, a linear transfer function is used between the hidden layer and the output layer, and the training sample is 500 groups of historical operation data.

4. The vacuum radiation water vapor phase drying method according to claim 3, characterized in that: The step of establishing the intake valve opening adjustment function adopts a fuzzy adaptive control algorithm, takes the temperature difference as the input variable, and the intake valve opening adjustment value as the output variable. The domain of the input variable is 0°C to 100°C, which is divided into 5 fuzzy subsets. The membership function adopts a combination of trapezoidal and triangular shapes, the domain of the output variable is 0 to 1, and the Mamdani reasoning method is adopted; the water vapor concentration calculation model is established using a Kalman filtering algorithm, takes the water vapor concentration as the state variable, and the pressure change rate as the observation variable. The system noise and the observation noise are both assumed to be Gaussian white noise, the sampling period of the state estimation is 1 second, the system noise variance is 0.01, and the observation noise variance is 0.

1.

5. The vacuum radiation water vapor phase drying method according to claim 4, characterized in that: The calculation step of the positive active adjustment value adopts a fuzzy neural network algorithm. The fuzzy neural network adopts a four-layer network structure, including an input layer, a fuzzy layer, a hidden layer and an output layer. The input layer contains two neurons, namely, the ambient temperature value and the water vapor concentration value. The fuzzy layer performs fuzzification processing on the input variables and adopts a Gaussian function as a membership function.

6. The vacuum radiation water vapor phase drying method according to claim 5, characterized in that: The calculation step of the negative passive adjustment value adopts the support vector regression algorithm, normalizes the temperature value and the humidity value, selects the radial basis kernel function as the kernel function, sets the penalty factor to 100, sets the kernel function parameter to 0.1, and selects the optimal parameter through the cross-validation method.

7. The vacuum radiation water vapor phase drying method according to claim 6, characterized in that: The preset drying threshold is determined by using the fuzzy C-means clustering algorithm, and cluster analysis is performed on the historical operation data. The number of clusters is set to 3, the Euclidean distance is used as the distance metric, the membership function uses an exponential function, the number of iterations is 100, and the termination threshold is 0.

001.

8. The vacuum radiation water vapor phase drying method according to claim 7, characterized in that: The step of collecting the temperature value of the core and the winding in the tank is achieved by installing temperature sensors on the surface of the core and the winding, and the temperature sensors are all PT100 platinum thermal resistors.

9. The vacuum radiation water vapor phase drying method according to claim 8, characterized in that: When the temperature value of the component to be tested reaches the preheating temperature inflection point value, the core temperature value inside the tank body and the winding temperature value inside the tank body are collected, the preheating temperature inflection point value is set to 122°C, the positive active adjustment value ranges from 0.6 to 1.4, and the negative passive adjustment value ranges from 0.4 to 1.

2.

10. The vacuum radiation water vapor phase drying method according to claim 9, characterized in that: When the negative passive adjustment value reaches the preset dryness threshold, the gradient descent method is used to reduce the output power of the radiation panel, and the power reduction rate is 2% per second. At the same time, the proportional integral control algorithm is used to control the closing speed of the intake valve, the proportional coefficient is 0.6, and the integration time is 30 seconds.

Images