Method for dynamically optimizing and adjusting control parameters of intelligent valve positioner

Abstract

The invention belongs to the technical field of intelligent regulating valves, and particularly relates to a method for dynamically optimizing and regulating control parameters of an intelligent valve positioner, which comprises the following steps of: acquiring equipment operation data through a multi-source sensor group and an industrial internet of things platform, and sequentially performing combined filtering and denoising, anomaly detection and linear normalization preprocessing on the acquired equipment operation data; performing time-frequency domain conjoint analysis based on short-time Fourier transform on the preprocessed data, and extracting a flow change rate, a pressure fluctuation range and a control signal frequency as characteristic parameters; adopting a support vector machine model to classify the equipment working conditions into three types, namely a steady state, a dynamic change and an abnormal state; according to the method for dynamically optimizing and adjusting the control parameters of the intelligent valve positioner, the flow change rate and the pressure fluctuation are extracted based on short-time Fourier transform, working conditions are rapidly classified in combination with a support vector machine, a basis is provided for control strategy switching, the problem that the real-time performance is poor when the working conditions are suddenly changed in a traditional fixed parameter mode is solved, and then the control precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent regulating valves, and in particular relates to a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner. Background Art

[0002] The intelligent automatic regulating valve is a pressure control device for natural gas parallel ports. It stabilizes downstream pipeline pressure by adjusting the output flow. This device consists of a quarter-turn electric actuator, a regulating element, a sealing element, a power supply system, and a control system. It automatically changes the size of the throttle opening within the valve, adjusting the flow area between the throttle and the seal, in response to the system's control signal. This allows for the regulation and control of process parameters such as medium flow and pressure. It also offers advantages such as a large rated flow coefficient, low axial balancing force on the regulating element, stable operation, a low output curve slope, and a large operating pressure differential. The valve positioner is a core accessory that matches the regulating valve opening to the control command, and together with the regulating valve, it forms a closed-loop control system.

[0003] Chinese invention patent publication number CN115163910A describes a method for dynamically optimizing and adjusting the control parameters of an intelligent valve positioner, applicable to the field of intelligent control valves. This invention dynamically optimizes control parameters for pneumatic control valves equipped with piezoelectric intelligent valve positioners, achieving optimal control performance. Key technical features of this solution include: improving the control algorithm's performance and adaptability to environmental changes, while also avoiding downtime for control valve maintenance.

[0004] However, the above technologies often have the following defects: traditional adjustment methods cannot adapt to complex and changing working conditions in real time, resulting in low control accuracy and high energy consumption.

[0005] To this end, the present invention provides a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner. Summary of the Invention

[0006] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a method for dynamically optimizing and adjusting the control parameters of an intelligent valve positioner according to the present invention comprises:

[0008] The equipment operation data is collected through a multi-source sensor group and the industrial Internet of Things platform, and the collected equipment operation data is sequentially subjected to combined filtering and denoising, anomaly detection, and linear normalization preprocessing;

[0009] A short-time Fourier transform-based time-frequency domain joint analysis is performed on the preprocessed data to extract the flow rate change rate, pressure fluctuation range, and control signal frequency as characteristic parameters. A support vector machine model is used to classify the equipment operating conditions into three categories: steady state, dynamic change, and abnormality.

[0010] A reinforcement learning algorithm is used to generate control parameter adjustment actions. The positioning accuracy, response time, and energy consumption of the control valve are used as the reward function. The recursive least squares method is used for online system identification. Based on real-time input and output data, the parameters of the control valve dynamic model are continuously updated. The optimized control parameters are then output based on the model's predictive control output.

[0011] A fault prediction model is built using an LSTM network. The model inputs historical equipment operation data and real-time characteristic parameters, outputs fault type, predicted occurrence time and location, and generates a preventive maintenance strategy.

[0012] The optimized parameters are written into the valve positioner control system in real time, and the operating data under the new parameters are synchronously collected and fed back to the system update module to form a closed-loop control circuit.

[0013] Furthermore, the multi-source sensor includes pressure, flow, displacement, temperature, vibration and sound sensors;

[0014] The combined filtering denoising adopts a cascade structure of a low-pass filter and a median filter, and the filtering parameters are dynamically adjusted according to the sensor type, wherein the low-pass filter removes high-frequency noise, and the median filter further eliminates random pulse noise.

[0015] Furthermore, the anomaly detection process satisfies:

[0016] For data that follow a normal distribution, data points outside the range of the mean ± 3 times the standard deviation are marked as outliers;

[0017] After the abnormal data are removed, the linear interpolation method is used to supplement it. The interpolation formula is:

[0018]

[0019] in, and is the known data point near the abnormal point; X is the independent variable value of the point to be interpolated; y is the dependent variable value of the point to be interpolated, that is, the result obtained by interpolation calculation. and are the independent and dependent variable values ​​of a known data point near the outlier; and are the independent and dependent variable values ​​of another known data point near the outlier.

[0020] Furthermore, the linear normalization process satisfies:

[0021]

[0022] Map the data to the interval [0,1], where and are the maximum and minimum values ​​of the current data window respectively, and the normalized results are mapped to the interval.

[0023] Furthermore, the reinforcement learning algorithm adopts the PPO algorithm, whose state space includes characteristic parameters and equipment condition classification results, and the action space is the incremental adjustment range of PID control parameters ±20%.

[0024] Furthermore, the LSTM network includes an input layer, multiple LSTM hidden layers and an output layer. The number of input layer nodes is determined according to the number of extracted feature parameters, and the number of hidden layer nodes is determined through experimental optimization. The historical operation data and real-time feature parameters of the equipment are input, and the fault type, predicted occurrence time and location are output, and a preventive maintenance strategy is generated.

[0025] Furthermore, it also includes intelligent fault-tolerant control:

[0026] When a critical sensor fails, it switches to a Kalman filter-based state estimator to generate alternative data;

[0027] In view of valve core wear fault, feedforward compensation algorithm is used to correct the control signal, and the compensation amount The calculation formula is:

[0028]

[0029] Among them, F is the valve core force data; It is the derivative of the valve core force data F with respect to time t, indicating the rate of change of the valve core force with time, and its unit is force per unit time.

[0030] Furthermore, it also includes simulation optimization based on digital twins:

[0031] Establish a digital twin that includes a fluid dynamics model and a mechanical transmission model, receive sensor data in real time, and output virtual performance indicators;

[0032] The control parameter optimization process is first tested on the digital twin, and the actual system only executes the verified parameter combinations.

[0033] Furthermore, the update frequency of the digital twin is 8 to 10 Hz, and when the error between the actual system and the virtual performance index exceeds 5%, the model recalibration is triggered.

[0034] Furthermore, it also includes dynamic resource allocation:

[0035] Dynamically adjust resource weight coefficients in three dimensions: control accuracy, response speed, and fault detection sensitivity based on the remaining battery power and CPU load rate. Satisfy the constraints:

[0036]

[0037] and .

[0038] The beneficial effects of the present invention are as follows:

[0039] 1. Based on the short-time Fourier transform, dynamic features such as flow rate change rate and pressure fluctuation are extracted, and combined with the support vector machine to quickly classify the working conditions, a real-time basis is provided for control strategy switching, avoiding the poor real-time response lag of the traditional fixed parameter mode when the working conditions suddenly change, thereby improving control accuracy;

[0040] 2. The PPO algorithm is used to dynamically adjust PID parameters, with positioning accuracy, response time, and energy consumption comprehensive indicators as the reward function to achieve multi-objective optimization. The recursive least squares method is combined to update the system model online, and the optimal parameters are calculated in advance through model predictive control. Compared with traditional PID control, energy consumption is reduced by 15% to 20%, and the purpose of reducing energy consumption is achieved based on this. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below with reference to the accompanying drawings.

[0042] Figure 1 This is an overall flow chart of a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to the present invention;

[0043] Figure 2 This is a schematic diagram of a multi-sensor data fusion process of a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to the present invention;

[0044] Figure 3 This is a LSTM network training loss curve diagram of a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to the present invention;

[0045] Figure 4 It is a valve core flow field pressure distribution diagram output by a digital twin of a method for dynamic optimization and adjustment of control parameters of an intelligent valve positioner of the present invention;

[0046] Figure 5 The present invention provides a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner, and shows a curve diagram of accuracy versus the number of hidden layer nodes and a curve diagram of false alarm rate versus the number of hidden layer nodes. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] See also Figure 1-5 , Example 1:

[0049] This embodiment provides: a method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner, comprising:

[0050] The equipment operation data is collected through a multi-source sensor group and the industrial Internet of Things platform, and the collected equipment operation data is sequentially subjected to combined filtering and denoising, anomaly detection, and linear normalization preprocessing;

[0051] It should be noted that the industrial Internet of Things platform adopts a layered architecture, with the bottom layer being the data collection layer. The protocol communicates with sensors and transmits the collected data to the middle data processing layer, which uses Kafka for data caching and streaming processing. The upper layer is the application layer, which provides functions such as data display, analysis, and control instruction issuance. The platform's data storage uses the distributed file system HDFS, which can store a large amount of device operation data.

[0052] Multi-source sensors include pressure, flow, displacement, temperature, vibration, and sound sensors;

[0053] Combined filtering denoising adopts a cascade structure of low-pass filter and median filter. The filtering parameters are dynamically adjusted according to the sensor type. The low-pass filter removes high-frequency noise, and the median filter further eliminates random pulse noise.

[0054] The specific parameter adjustment logic of the combined filter includes:

[0055] Dynamic parameter adjustment mechanism

[0056] Low-pass filter cutoff frequency: Dynamically set according to the sensor sampling frequency, satisfying:

[0057]

[0058] The cutoff frequency of the low-pass filter is in Hertz, which indicates the highest frequency component allowed to pass through the filter.

[0059] is the sampling frequency of the sensor, in Hertz, which is the number of times the sensor samples the signal per second;

[0060] is the frequency of the signal, in Hertz, which may refer to the relevant frequency components in the signal collected by the sensor;

[0061] To obtain the maximum frequency value in the signal;

[0062] The function of this formula is to dynamically determine the cutoff frequency of the low-pass filter according to the sensor sampling frequency and signal frequency to meet the needs of signal processing.

[0063] in The main frequency of the current signal is obtained by FFT analysis to obtain the median filter window size: based on the impulse noise density (Noise density is calculated by counting the number of units of time exceeding 3 Calculation of the proportion of data points) Adaptive adjustment:

[0064]

[0065] It should be noted that when performing FFT analysis to obtain the current signal's main frequency, the sampling frequency should be at least twice the highest frequency of the signal. Based on the expected frequency range of the sensor's signal acquisition, the sampling frequency is set to 500Hz. The number of sampling points N is selected as 1024 points, which can ensure a frequency resolution of , perform FFT transformation on the collected N data points to obtain the frequency domain signal, and then find the frequency corresponding to the maximum amplitude in the frequency domain signal, which is the main frequency of the current signal .

[0066] Sensor differentiation through:

[0067]

[0068] From the above table, we can conclude

[0069] Anomaly detection processing satisfies:

[0070] For data that follow a normal distribution, data points outside the range of the mean ± 3 times the standard deviation are marked as outliers;

[0071] After the abnormal data are removed, the linear interpolation method is used to supplement it. The interpolation formula is:

[0072]

[0073] in, and is a known data point near the outlier; x is the independent variable value of the point to be interpolated; y is the dependent variable value of the point to be interpolated, that is, the result obtained by interpolation calculation. and are the independent and dependent variable values ​​of a known data point near the outlier; and are the independent and dependent variable values ​​of another known data point near the outlier.

[0074] It should be noted that when using the linear interpolation method to supplement abnormal data, data points near the abnormal point are selected in chronological order. Specifically, with the abnormal point as the reference, the two normal data points closest to the abnormal point in time are selected forward and backward as known data points. If the time intervals between the adjacent data points and the abnormal point are the same, the data points collected first are given priority.

[0075] Linear normalization satisfies:

[0076]

[0077] Map the data to the interval [0,1], where and are the maximum and minimum values ​​of the current data window respectively, and the normalized results are mapped to the interval.

[0078] A short-time Fourier transform-based time-frequency domain joint analysis is performed on the preprocessed data to extract the flow rate change rate, pressure fluctuation range, and control signal frequency as characteristic parameters. A support vector machine model is used to classify the equipment operating conditions into three categories: steady state, dynamic change, and abnormality.

[0079] A reinforcement learning algorithm is used to generate control parameter adjustment actions. The positioning accuracy, response time, and energy consumption of the control valve are used as the reward function. The recursive least squares method is used for online system identification. Based on real-time input and output data, the parameters of the control valve dynamic model are continuously updated. The optimized control parameters are then output based on the model's predictive control output.

[0080] The reinforcement learning algorithm adopts the PPO algorithm, whose state space includes characteristic parameters and equipment condition classification results, and the action space is the incremental adjustment range of PID control parameters ±20%.

[0081] It is important to note that the state space is composed of extracted characteristic parameters such as flow rate change, pressure fluctuation range, and control signal frequency, as well as the equipment operating condition classification results (steady state, dynamic change, and abnormal). The control parameter incremental adjustment range of ±20% is specifically adjusted by multiplying the current control parameter by (1 + adjustment step size) or (1 - adjustment step size), with the adjustment step size set to 0.05. In each iteration, the control parameter is adjusted based on the reward function (combined indicators of the control valve's positioning accuracy, response time, and energy consumption).

[0082] A fault prediction model is built using an LSTM network. The model inputs historical equipment operation data and real-time characteristic parameters, outputs fault type, predicted occurrence time and location, and generates a preventive maintenance strategy.

[0083] The LSTM network consists of an input layer, multiple LSTM hidden layers, and an output layer. The number of input layer nodes is determined based on the number of extracted feature parameters, and the number of hidden layer nodes is determined through experimental optimization. The network inputs historical equipment operation data and real-time feature parameters, outputs fault type, predicted occurrence time and location, and generates a preventive maintenance strategy.

[0084] It should be noted that the grid search method is used to optimize the number of hidden layer nodes; that is, the candidate range of the number of hidden layer nodes is set to 10-100, with an interval of 10, and the accuracy and false alarm rate are used as evaluation indicators. A 5-fold cross-validation is performed on each candidate number of nodes. During the experiment, the accuracy and false alarm rate corresponding to each number of nodes are recorded, and the curves of accuracy-number of hidden layer nodes and false alarm rate-number of hidden layer nodes are plotted. By observing Figure 5 That is, the curve selects the number of nodes with higher accuracy and lower false alarm rate as the optimal number of hidden layer nodes, which is finally determined to be 50.

[0085] pass Figure 5 It can be concluded that: during the experiment, when the number of hidden layer nodes is 10, the accuracy rate is 78% and the false alarm rate is 12%; when the number of nodes is 20, the accuracy rate increases to 82% and the false alarm rate drops to 10%; as the number of nodes increases to 50, the accuracy rate reaches 90% and the false alarm rate is 8%; when the number of nodes continues to increase to 100, the accuracy rate does not increase significantly, but the false alarm rate increases slightly.

[0086] The training process of the fault prediction model includes:

[0087] Source of training data: Historical equipment operation data is collected, including data under normal working conditions and fault conditions, with a total of 1,200 sets of samples, a sampling frequency of 100 Hz, and a sampling time of 2 hours before the fault.

[0088] Data preprocessing method: clean and normalize the data to remove outliers and noise.

[0089] Model training parameters: the learning rate is set to 0.001 and the number of iterations is 1000.

[0090] Model performance evaluation: Accuracy and false alarm rate are used as evaluation indicators, and evaluation is performed through 5-fold cross validation.

[0091] Also includes intelligent fault-tolerant control:

[0092] When a critical sensor fails, it switches to a Kalman filter-based state estimator to generate alternative data;

[0093] In view of valve core wear fault, feedforward compensation algorithm is used to correct the control signal, and the compensation amount The calculation formula is:

[0094]

[0095] in, The wear coefficient is a coefficient related to the degree of valve core wear, reflecting the influence of valve core wear on control signal compensation. It is a dimensionless value, and its specific value needs to be determined according to the actual wear of the valve core and experimental calibration.

[0096] F is the valve core force data; It is the derivative of the valve core force data F with respect to time t, indicating the rate of change of the valve core force over time. The unit is force per unit time, which reflects the speed of the change of the valve core force. is the compensation amount, which is the value calculated by the feedforward compensation algorithm and used to correct the control signal.

[0097] It should be noted that in the Kalman filter state estimator, the process noise covariance matrix Q is set according to the dynamic characteristics of the system. After many experiments and adjustments, Q is set to a diagonal matrix with diagonal elements of 0.1, 0.08, and 0.05, which correspond to the process noise variances of different state variables; the measurement noise covariance matrix R is determined according to the measurement accuracy of the sensor. For the pressure sensor, the corresponding measurement noise variance is set to 0.02, and for the flow sensor, it is set to 0.03.

[0098] Also includes simulation optimization based on digital twins:

[0099] Establish a digital twin that includes a fluid dynamics model and a mechanical transmission model, receive sensor data in real time, and output virtual performance indicators;

[0100] The control parameter optimization process is first tested on the digital twin, and the actual system only executes the verified parameter combinations.

[0101] The update frequency of the digital twin is 8 to 10 Hz, and when the error between the actual system and the virtual performance index exceeds 5%, the model recalibration is triggered.

[0102] The fluid dynamics equations include:

[0103] Core governing equations

[0104] Valve core force model:

[0105]

[0106] in The flow coefficient is a dimensionless number that reflects the flow characteristics of the fluid when passing through the valve. It is related to factors such as the valve's geometry and opening, and is used to correct the difference between the theoretical flow rate and the actual flow rate.

[0107] The flow area is measured in square meters and refers to the effective cross-sectional area of ​​the fluid passing through the valve. Its size affects the flow rate and flow velocity of the fluid.

[0108] The unit of fluid density is kilograms per cubic meter, which represents the mass of the fluid per unit volume. The density of the fluid will affect its physical quantities such as inertia and pressure;

[0109] The pressure difference is in Pascals, which is the pressure difference between the valve inlet and outlet. The pressure difference is one of the driving forces for fluid flow.

[0110] The valve core shape resistance coefficient is a dimensionless number. It is related to the shape of the valve core and reflects the size of the valve core shape's resistance to the fluid. Different shapes of valve cores will have different resistance coefficients.

[0111] The fluid velocity is measured in meters per second, indicating the flow rate of the fluid at the valve. The magnitude of the velocity will affect the interaction force between the fluid and the valve core.

[0112] It is the force exerted by the fluid on the valve core, measured in Newtons. It comprehensively reflects the total effect of the fluid pressure and the resistance generated by the relative movement of the fluid and the valve core on the valve core.

[0113] Flow field simulation module

[0114] use Turbulence model solution equation:

[0115]

[0116]

[0117] in, It is the fluid density in kilograms per cubic meter, which has the same meaning as the valve core force model and describes the mass distribution characteristics of the fluid.

[0118] It is time, measured in seconds, used to describe the change of physical quantities over time.

[0119] is the fluid velocity vector, measured in meters per second. It contains the velocity components of the fluid in all directions and is used to describe the motion state of the fluid.

[0120] It is the Hamiltonian operator, which is used to represent the spatial gradient in the rectangular coordinate system and to calculate the divergence of the velocity field in the continuity equation;

[0121] The fluid pressure, measured in Pascals, represents the normal force per unit area of ​​the fluid and is an important physical quantity in fluid statics and dynamics.

[0122] It is the viscous stress tensor with the unit of Pascal. It describes the stress generated by the viscosity inside the fluid, reflects the influence of fluid viscosity on the flow, and is related to factors such as the velocity gradient of the fluid.

[0123] Real-time optimization

[0124] Model reduction: Project the high-dimensional flow field into the 10th-order modal space through POD;

[0125] Parameter mapping table: Pre-calculate the parameters at different openings 、 Relationship, runtime lookup table interpolation.

[0126] It should be noted that the fluid dynamics model uses computational fluid dynamics (CFD) software Modeling is carried out, which includes the following processes;

[0127] First, a 3D model is built based on the valve's geometric structure. Then, the mesh is divided and boundary conditions, such as inlet and outlet pressures, are set.

[0128] Select an appropriate turbulence model (e.g. The mechanical transmission model is constructed using the multi-body dynamics software ADAMS. The connection relationship, mass, moment of inertia and other parameters of each component are defined according to the mechanical structure of the valve. The drive and constraint conditions are set to perform kinematic and dynamic analysis.

[0129] By comparing the model with the actual equipment operation data, the model parameters are adjusted to make the output of the model as consistent as possible with the actual data.

[0130] Also includes dynamic resource allocation:

[0131] Dynamically adjust resource weight coefficients in three dimensions: control accuracy, response speed, and fault detection sensitivity based on the remaining battery power and CPU load rate. Satisfy the constraints:

[0132] ,and .

[0133] The specific algorithm is as follows:

[0134] Get the remaining battery power B and CPU load rate L in real time.

[0135] Adjust the weight coefficients according to the following rules:

[0136] When B<20% or L>80%, reduce the control accuracy weight. , increase the response speed weight ;

[0137] When the equipment is in abnormal working condition, the fault detection sensitivity weight is increased .

[0138] It should be noted that the remaining battery power is obtained in real time through the interface provided by the battery management system. The battery management system monitors the battery voltage, current and other parameters in real time, and calculates the remaining power based on the battery's charge and discharge characteristics. The data is transmitted to the intelligent valve positioner control system through serial communication. The CPU load rate is obtained through the system call interface provided by the operating system. In the Linux system, the CPU load rate can be obtained by reading the relevant information in the / proc / stat file and calculating it. The data is also transmitted to the control system through the system's internal communication mechanism.

[0139] The optimized parameters are written into the valve positioner control system in real time, and the operating data under the new parameters are synchronously collected and fed back to the system update module to form a closed-loop control circuit.

[0140] Example 1:

[0141] Comparison between traditional PID control (fixed parameters) and the proposed method (PPO+MPC) on the same test platform

[0142] Test conditions:

[0143] Steady-state working conditions (pressure 0.5MPa, flow rate 10L / min)

[0144] Dynamic mutation working condition (pressure 0.3→0.8MPa step change)

[0145] Abnormal operating conditions (simulated sensor drift ±5%)

[0146] Energy consumption measurement: , that is, total energy consumption = (current 2 × resistance + gas source power consumption) accumulated over time.

[0147] Measured data: (Test period T = 8 hours, solenoid valve rated power 50W, air source pressure 0.7MPa)

[0148]

[0149] From the above table we can conclude that:

[0150] The PPO algorithm reduces overshoot under dynamic conditions (conventional PID overshoot 15% → this invention 5%).

[0151] MPC calculates the optimal trajectory in advance, reducing gas consumption by 23%;

[0152] Example 2

[0153] We artificially created valve core stuck faults (particulate contamination / mechanical deformation) on 50 control valves of different models and collected data (sampling rate 100 Hz) for two hours before the fault, obtaining a total of 1,200 sets of samples.

[0154] Among them, friction warning:

[0155] Under normal circumstances, the force on the valve core is ≤8 Newtons;

[0156] Fault precursor ≥ 15 Newtons (up to 18N when stuck)

[0157] Abnormal vibration:

[0158] The vibration energy in the frequency range of 200-500Hz is 50% higher than normal.

[0159] Test results: (Test set: 300 independent samples, 5-fold cross validation)

[0160]

[0161] The above table shows that the system's prediction accuracy for valve core stuck faults reaches 92%.

[0162] The above-mentioned front, back, left, right, up and down are all based on the Figure 1 As a benchmark, according to the person's observation perspective, the side of the device facing the observer is defined as the front, the left side of the observer is defined as the left, and so on.

[0163] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the scope of protection of the present invention.

[0164] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner, characterized by: include: The equipment operation data is collected through a multi-source sensor group and the industrial Internet of Things platform, and the collected equipment operation data is sequentially subjected to combined filtering and denoising, anomaly detection, and linear normalization preprocessing; Perform time-frequency domain joint analysis based on short-time Fourier transform on the preprocessed data to extract flow rate change rate, pressure fluctuation range and control signal frequency as characteristic parameters; The support vector machine model is used to classify equipment operating conditions into three categories: steady state, dynamic change, and abnormality; A reinforcement learning algorithm is used to generate control parameter adjustment actions. The positioning accuracy, response time, and energy consumption of the control valve are used as the reward function. The recursive least squares method is used for online system identification. Based on real-time input and output data, the parameters of the control valve dynamic model are continuously updated. The optimized control parameters are then output based on the model's predictive control output. A fault prediction model is built using an LSTM network. The model inputs historical equipment operation data and real-time characteristic parameters, outputs fault type, predicted occurrence time and location, and generates a preventive maintenance strategy. The optimized parameters are written into the valve positioner control system in real time, and the operating data under the new parameters are synchronously collected and fed back to the system update module to form a closed-loop control circuit.

2. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: The multi-source sensors include pressure, flow, displacement, temperature, vibration and sound sensors; The combined filtering denoising adopts a cascade structure of a low-pass filter and a median filter, and the filtering parameters are dynamically adjusted according to the sensor type, wherein the low-pass filter removes high-frequency noise, and the median filter further eliminates random pulse noise.

3. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: The anomaly detection process satisfies: For data that follow a normal distribution, data points outside the range of the mean ± 3 times the standard deviation are marked as outliers; After the abnormal data are removed, the linear interpolation method is used to supplement it. The interpolation formula is: , in, and is a known data point near the outlier; x is the independent variable value of the point to be interpolated; y is the dependent variable value of the point to be interpolated, that is, the result obtained by interpolation calculation. and are the independent and dependent variable values ​​of a known data point near the outlier; and are the independent and dependent variable values ​​of another known data point near the outlier.

4. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: The linear normalization process satisfies: Map the data to the interval [0,1], where and are the maximum and minimum values ​​of the current data window respectively, and the normalized results are mapped to the interval.

5. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: The reinforcement learning algorithm adopts the PPO algorithm, whose state space includes characteristic parameters and equipment condition classification results, and the action space is the incremental adjustment range of PID control parameters ±20%.

6. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: The LSTM network includes an input layer, multiple LSTM hidden layers, and an output layer. The number of input layer nodes is determined based on the number of extracted feature parameters, and the number of hidden layer nodes is determined through experimental optimization. The network inputs historical equipment operation data and real-time feature parameters, outputs fault type, predicted occurrence time and location, and generates a preventive maintenance strategy.

7. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: Also includes intelligent fault-tolerant control: When a critical sensor fails, it switches to a Kalman filter-based state estimator to generate alternative data; In response to valve core wear failure, the feedforward compensation algorithm is used to correct the control signal and the compensation amount The calculation formula is: , where F is the valve core force data; It is the derivative of the valve core force data F with respect to time t, indicating the rate of change of the valve core force with time, with the unit being force per unit time.

8. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: Also includes digital twin-based simulation optimization: Establish a digital twin that includes a fluid dynamics model and a mechanical transmission model, receive sensor data in real time, and output virtual performance indicators; The control parameter optimization process is first tested on the digital twin, and the actual system only executes the verified parameter combinations.

9. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 8, characterized in that: The update frequency of the digital twin is 8-10 Hz, and when the error between the actual system and the virtual performance index exceeds 5%, the model recalibration is triggered.

10. The method for dynamically optimizing and adjusting control parameters of an intelligent valve positioner according to claim 1, characterized in that: Also includes dynamic resource allocation: Dynamically adjust resource weight coefficients in three dimensions: control accuracy, response speed, and fault detection sensitivity based on the remaining battery power and CPU load rate. Satisfy the constraints: ,and .

Images