Petrochemical pipe gallery intelligent monitoring system based on Internet of Things

Abstract

The invention discloses a petrochemical pipe gallery intelligent monitoring system based on the Internet of Things, and particularly relates to the technical field of petrochemical pipe gallery control. A petrochemical pipe gallery deployment sensor array collects data, and the data is subjected to noise suppression and space-time alignment processing to obtain a standardized time sequence data stream; extracting a space-time P matrix of the pressure field and performing singular value decomposition to obtain a dominant mode primary function; training a physical information neural network through the full-order flow field data and the standardized time sequence data flow generated through simulation to obtain a transient pressure wave propagation prediction model; based on a pressure wave prediction result, defining a state vector including a pressure gradient, valve wear and an energy efficiency ratio; a distributed strategy library is trained through a priority experience playback mechanism, and a Pareto optimal control instruction set is generated; through a double-closed-loop control mechanism, the precision adjustment and anti-disturbance capability of the valve action are realized, and the problem of prediction distortion caused by neglecting transient pressure wave propagation characteristics in a traditional method is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical pipe gallery control, and more specifically, to an intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things. Background Art

[0002] In the current operation mode of petrochemical pipe galleries, pipelines are tasked with transporting high-temperature and high-pressure mixed media (such as oil and gas, chemicals, etc.). The petrochemical pipe gallery monitoring system monitors various operating parameters in real time by deploying a variety of sensors such as temperature, pressure, medium viscosity, and pipe wall vibration. Pipelines are usually designed with emergency shut-off valves to respond to emergencies such as equipment failures or external damages. Once an abnormal signal is detected, the shut-off valve will quickly close to isolate the accident area and prevent the accident from spreading.

[0003] However, since the fluid in the pipeline is in a high-temperature and high-pressure state, the physical properties of the medium and the elastic modulus of the pipe wall will change dynamically due to temperature fluctuations, flow rate changes, and long-term aging. When the emergency shut-off valve closes quickly, the fluid in the pipe continues to move due to inertia, resulting in the instantaneous conversion of fluid kinetic energy into pressure energy, thus forming a violent transient pressure wave (i.e., Joukowsky shock). When this shock wave propagates in the pipeline, its speed and waveform are affected by the medium viscosity and the elastic modulus of the pipe wall, and thus may form an abnormally high pressure locally, threatening the structural integrity of the pipeline and potentially causing more serious safety accidents such as secondary leaks.

[0004] Currently, for the problem of transient hydraulic shock after the action of the emergency shut-off valve, traditional technologies mainly use a fixed pressure threshold to trigger mitigation measures, that is, when the pressure in the pipe exceeds a preset value, a mitigation device is started to reduce the shock amplitude, but this method has the following deficiencies:

[0005] Insufficient response sensitivity: The fixed threshold cannot adapt to the dynamic changes of the medium viscosity and the elastic modulus of the pipe wall under working conditions. It may be too sensitive in some cases, frequently triggering mitigation measures, while being insufficient in response during real danger, increasing the risks of misjudgment and missed judgment.

[0006] Ignoring the coupling effect: Traditional methods do not fully consider the complex coupling effect between the fluid and the structure, and it is difficult to accurately capture the propagation characteristics of the transient pressure wave, thus limiting the accuracy and real-time performance of the mitigation strategy.

[0007] Lack of adaptive regulation: The single fixed threshold strategy cannot balance safety and economy under variable working conditions, resulting in difficulty in timely adjusting local abnormal pressures and further exacerbating the risk of pipeline damage. Summary of the Invention

[0008] To overcome the above-mentioned deficiencies of the prior art, the present invention provides an intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things. By deploying a sensor array in the petrochemical pipe gallery, data is collected and processed through noise suppression and spatio-temporal alignment to obtain a standardized time-series data stream; the P matrix of the pressure field spatio-temporal is extracted and singular value decomposition is performed to obtain the dominant modal basis function; the LSTM-PINN physics-informed neural network is trained with the full-order flow field data generated by simulation and the standardized time-series data stream to obtain a transient pressure wave propagation prediction model; based on the pressure wave prediction results, a state vector including pressure gradient, valve wear, and energy efficiency ratio is defined; the NSGA-II algorithm is used to optimize the weights of the multi-objective reward function, and the distributed policy library is trained through the prioritized experience replay mechanism to generate a Pareto optimal control instruction set for the petrochemical pipe gallery valve control, so as to solve the problems proposed in the above-mentioned background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions: An intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things, comprising:

[0010] A petrochemical pipe gallery data acquisition module, by deploying a sensor array in the petrochemical pipe gallery, at least including a pressure sensor, an ultrasonic flowmeter, and a temperature sensor, performs dynamic noise suppression and spatio-temporal alignment processing on multi-modal sensor data to obtain a standardized time-series data stream;

[0011] In the implementation of the petrochemical pipe gallery data acquisition module, a working condition adaptive Kalman filter is adopted to dynamically adjust the process noise covariance matrix according to the Reynolds number Re; distributed sensor node clock synchronization is achieved based on the IEEE1588PTP protocol, and the synchronization accuracy is ≤1 μs to eliminate the phase difference; the 3σ criterion is used to filter out outliers to ensure data validity;

[0012] A pressure wave propagation prediction module, used to build a transient pressure wave propagation prediction model, extract the P matrix of the pressure field spatio-temporal from the standardized time-series data stream, perform singular value decomposition on the P matrix by proper orthogonal decomposition to obtain the dominant modal basis function; the LSTM-PINN physics-informed neural network is trained with the full-order flow field data generated by simulation and the standardized time-series data stream to obtain a transient pressure wave propagation prediction model;

[0013] A valve control strategy generation module, based on the pressure wave prediction results of the transient pressure wave propagation prediction model, defines a state vector including pressure gradient, valve wear, and energy efficiency ratio; the NSGA-II algorithm is used to optimize the weights of the multi-objective reward function, and the distributed Q-learning policy library is trained through the prioritized experience replay mechanism to generate a Pareto optimal control instruction set;

[0014] In the implementation of the valve control strategy generation module, the collaborative effect of NSGA-II and Q-learning is reflected in: adopting a hierarchical optimization framework, NSGA-II optimizes the reward function weights (Pareto front search), and Q-learning generates a policy library based on the optimized weights; the policy generation frequency matches the valve control delay, and a sliding window is used to compress the scale of the policy library, combined with FPGA pipeline acceleration;

[0015] The valve control and adjustment module deploys the Pareto optimal control instruction set to the FPGA hardware, and solidifies the dominant mode basis function and the action value table query logic; based on the standardized time-series data stream, through a dual-closed-loop control mechanism including inner-loop PID adjustment and outer-loop pressure feedback interpolation, it realizes the precision adjustment and anti-disturbance ability of the valve action, and outputs the valve action signal.

[0016] Preferably, the method for obtaining the pressure gradient: based on the predicted pressure field output by the transient pressure wave propagation prediction model, the prediction result includes the pressure distribution at the pipeline spatial coordinates at the predicted time step; the three-dimensional central difference method is used to calculate the pressure gradient field, and through the pressure sensor array deployed at the key nodes of the pipe gallery (every 5 meters along the pipeline axis), the measured pressure gradient is collected in real time, and the predicted gradient is corrected for residuals based on the prediction error;

[0017] Preferably, the method for obtaining the valve wear quantification method is: the valve opening change is recorded in real time through a rotary encoder; the axial force of the valve stem is measured based on a torque sensor; the cumulative wear amount is calculated using the modified Archard wear formula, and the cumulative wear amount is normalized to a dimensionless wear parameter; the calculation formula for the cumulative wear amount L(t) is as follows:

[0018]

[0019] where Ks is the valve material wear coefficient, H is the material hardness, v(τ) is the valve stem movement speed, f(T(τ)) is the temperature compensation function (T is the real-time temperature), which is modeled based on the material thermal expansion coefficient and the temperature-hardness relationship; L(t) is normalized to a dimensionless wear parameter where L max is the valve design life threshold; F(τ) represents the axial force of the valve stem (unit: N), which is measured in real time by a torque sensor installed on the valve stem transmission mechanism.

[0020] Preferably, the acquisition method of the P matrix in the space-time of the pressure field includes:

[0021] Arrange the preprocessed standardized time-series data stream in chronological order to form a matrix structure; the rows represent the positions of each sensor or grid point; the columns represent the pressure values at each sampling moment, including the pressure distribution of the entire pipeline.

[0022] The P matrix is decomposed into three matrices through proper orthogonal decomposition, which are respectively:

[0023] The left singular vectors, with each column representing a spatial fluctuation mode, i.e., the modal basis functions;

[0024] The diagonal matrix, with the diagonal elements being singular values, indicating the importance of each modal basis function;

[0025] The right singular vectors, with each row representing the time evolution law of the corresponding modal basis function.

[0026] Preferably, during the generation process of the P matrix in the pressure field space-time, Kriging interpolation dynamic compensation is performed on the full-order flow field data and real-time sensing data to unify them to the same spatial resolution; the physical constraint ignores the gravity term; a transfer learning mechanism is introduced in the LSTM-PINN, and the full-order flow field data and real-time data distributions are adjusted through the domain adaptation loss function.

[0027] Preferably, the acquisition method of the transient pressure wave propagation prediction model includes the following steps:

[0028] Step S11, CFD full-order flow field simulation and data matrix construction: Use CFD numerical simulation to generate full-order flow field data covering typical working conditions, with the inputs including the geometric parameters of the pipe gallery and the physical properties of the fluid; simulate the typical working conditions by solving the Navier-Stokes equations to obtain the simulation data of the typical working conditions; construct the pressure field snapshot matrix by extracting the spatial pressure distribution of the full-order flow field data;

[0029] Step S12, POD modal basis function extraction and orthogonalization: Perform proper orthogonal decomposition on the pressure field snapshot matrix, decompose it into several orthogonal modal basis functions, obtain the dominant modal basis functions, and verify the modal orthogonality;

[0030] Step S13, construct an LSTM time series encoder. By taking the pressure, flow rate, and temperature data collected in real time as inputs and combining them with the dominant modal basis functions, construct an LSTM-PINN physics-informed neural network, where the LSTM time series encoder is used to process time series data, and the physical constraint layer ensures that the output conforms to the Navier-Stokes equations; train based on the optimizer and training data to obtain the transient pressure wave propagation prediction model;

[0031] Step S14, online dynamic update and accuracy verification: At the set fixed frequency, perform incremental learning on the LSTM-PINN network by collecting new sensing data, and adjust the network parameters to adapt to the changes in real-time working conditions; verify and optimize the transient pressure wave propagation prediction model in real time by calculating the global mean absolute error and the pressure wave peak position error of the pressure wave.

[0032] Preferably, the physical constraint term \(L\) of the domain adaptation loss function of the LSTM-PINN hybrid network phy is as follows:

[0033]

[0034] where \(\lambda\) is the physical constraint weight coefficient, determined by cross-validation; \(P\) pred represents the predicted pressure field, \(\mathbf{u}\) is the velocity vector field, \(\nu\) is the dynamic viscosity coefficient, \(\nabla^2\) is the Laplace operator, \(\alpha_i(t)\) i (t) represents the time-dependent modal weight coefficient, characterizing the contribution degree of the \(i\)-th order modal basis function to the pressure field at time \(t\), \(k\) represents the number of dominant modal basis functions, and \(i\) represents the index; \(\varphi_i\) i (t) represents the POD modal basis function, which is the dominant spatial distribution pattern of pressure waves extracted from the full-order flow field data.

[0035] Preferably, the domain adaptation loss function is composed as follows:

[0036] \(L\) <00,00006>\(=\left\|\mathbf{P}\right.\) pred \(-\mathbf{P}^*\left\|\right.+L_{\text{phys}}\) preal where the data fitting term \(\left\|\mathbf{P}\right.\) phy

[0037] \(-\mathbf{P}^*\left\|\right.\) forces the prediction result to approximate the measured data, and the physical constraint term \(L_{\text{phys}}\) pred is used to ensure that the prediction conforms to the conservation law of fluid mechanics. preal phy

[0038] Preferably, the multi-objective reward function is a weighted sum of pressure fluctuation suppression, valve life optimization, and execution efficiency improvement. Among them, pressure fluctuation suppression is negatively correlated with the real-time pressure deviation, suppressing the risk of pressure overshoot; valve life optimization is negatively correlated with the cumulative mechanical wear of the valve, reducing equipment loss caused by frequent actions; execution efficiency improvement is positively correlated with the valve action torque efficiency, ensuring the balance between response speed and energy consumption. The weight coefficients of each sub-item are determined by a multi-objective optimization algorithm to generate a Pareto optimal control strategy set.

[0039]

[0040] Preferably, a pipe wall elasticity correction term is added to the physical constraint term \(L_{\text{phys}}\) phy to obtain the updated physical constraint term \(XL_{\text{phys}}\) phy :

[0041] where \(E(t)\) represents the real-time inverse pipe wall elastic modulus, obtained by deploying strain sensors at key positions on the pipe wall; \(\epsilon(t)\) represents the pipe wall strain tensor, It is a tensor product, representing the coupling effect of the pressure gradient and the strain field. It represents the feedback effect of the elastic deformation of the pipe wall on the pressure wave, and ρ represents the fluid density in the petrochemical pipe gallery.

[0042] Preferably, the double closed-loop anti-control mechanism includes:

[0043] The inner loop control layer is used for high-frequency regulation: adopting integral separation PID, dynamically amplifying the control signal according to the deviation between the target and the actual opening of the valve; if the deviation between the target and the actual opening is greater than the preset value, the proportional term quickly outputs a strong correction signal; freezing the integral term, only retaining the proportional and derivative actions to prevent the valve from getting stuck at the limit position.

[0044] The outer loop control layer is used for global planning: based on real-time pressure feedback, generating a smooth valve opening curve to avoid mechanical shocks caused by stepwise actions and restricting the pressure gradient not to exceed the preset threshold.

[0045] Preferably, during the inner loop PID regulation process, the inner loop PID proportional coefficient K p (t) is positively correlated with the medium viscosity. When the viscosity increases, the inner loop PID proportional coefficient is increased, which is the core parameter for controlling the response speed. When the viscosity increases, it is increased to compensate for the delay and accelerate the valve response speed to compensate for the delay caused by viscous resistance; when the viscosity decreases, the inner loop PID proportional coefficient is decreased to avoid overshoot and oscillation.

[0046] The technical effects and advantages of the present invention:

[0047] (1) The intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things provided by the present invention realizes dynamic noise suppression and spatio-temporal alignment of multi-modal sensor data through dynamic Kalman filtering and clock synchronization technology, generates a highly consistent standardized time-series data stream, enables the standardized time-series data stream to dynamically adapt to the physical property changes of the fluid and the pipeline, and solves the problem of misjudgment / missing judgment caused by the inability of the traditional fixed pressure threshold method to adapt to the dynamic changes of the medium viscosity and the pipe wall elastic modulus; through the transient pressure wave propagation prediction model and Kriging interpolation dynamic compensation technology, it realizes high-precision simulation of the fluid-structure coupling effect, solves the problem of prediction distortion caused by the traditional method ignoring the transient pressure wave propagation characteristics, and provides a physically interpretable decision basis for valve control.

[0048] (2) The intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things provided by the present invention realizes dynamic balanced regulation of safety and economy through NSGA-II multi-objective optimization (balancing the weights of pressure fluctuation suppression, valve life, and energy efficiency ratio) and double closed-loop control mechanism, combined with FPGA hardware acceleration (end-to-end delay ≤ 5 ms), and solves the problem that a single threshold strategy is difficult to adapt to complex working conditions. Description of the Drawings

[0049] Figure 1 This is the structural block diagram of the intelligent monitoring system for petrochemical pipe corridors of the present invention.

[0050] Figure 2 This is the flow chart for building the transient pressure wave propagation prediction model of the present invention.

[0051] Figure 3 This is the structural block diagram of the intelligent monitoring system for petrochemical pipe corridors based on the domain adaptation loss function of the present invention. Detailed implementation manners

[0052] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0053] Meanwhile, it should be understood that, for the sake of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship.

[0054] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present application and its application or use.

[0055] Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods, and devices should be regarded as part of the description.

[0056] Example 1. Refer to Figure 1 the structural block diagram of the intelligent monitoring system for petrochemical pipe corridors, the present invention provides a kind of intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things as shown in Figure 1 and includes:

[0057] The petrochemical pipe corridor data acquisition module is used to achieve the full-dimensional capture of transient pressure wave characteristics and ensure data consistency. By deploying a sensor array (pressure sensors, ultrasonic flow meters, and temperature sensors) in the petrochemical pipe corridor, dynamic noise suppression and spatio-temporal alignment processing are performed on multi-modal sensor data to obtain a standardized time-series data stream;

[0058] A pressure wave propagation prediction module is used to build a transient pressure wave propagation prediction model, extract the P matrix of the spatio-temporal pressure field from the standardized time-series data stream, perform singular value decomposition on the P matrix using proper orthogonal decomposition (POD) to obtain the dominant mode basis functions; train a long short-term memory - physics informed neural network (LSTM - PINN) with the full - order flow field data generated by simulation and the standardized time - series data stream to obtain the transient pressure wave propagation prediction model; LSTM is used to process time - series data, and PINN is used to introduce physical constraints to ensure that the model conforms to the laws of fluid mechanics.

[0059] A valve control strategy generation module, based on the pressure wave prediction results of the transient pressure wave propagation prediction model, defines a state vector including pressure gradient, valve wear, and energy efficiency ratio; uses the NSGA - II algorithm to optimize the weights of the multi - objective reward function, and trains a distributed Q - learning policy library through a prioritized experience replay mechanism to generate a Pareto - optimal control instruction set.

[0060] Furthermore, the method for obtaining the pressure gradient: Based on the predicted pressure field output by the transient pressure wave propagation prediction model, the prediction results include the pressure distribution at the pipeline spatial coordinates at the predicted time step; use the three - dimensional central difference method to calculate the pressure gradient field, and through a pressure sensor array deployed at key nodes of the pipe gallery (every 5 meters along the pipeline axis), collect the measured pressure gradient in real time, and perform residual correction on the predicted gradient based on the prediction error.

[0061] Furthermore, the method for quantifying valve wear is as follows: The change in valve opening is recorded in real time through a rotary encoder; the axial force of the valve stem is measured based on a torque sensor; the cumulative wear amount is calculated using a modified Archard wear formula, and the cumulative wear amount is normalized to a dimensionless wear parameter; the calculation formula for the cumulative wear amount L(t) is as follows:

[0062]

[0063] where Ks is the wear coefficient of the valve material, H is the material hardness, v(τ) is the movement speed of the valve stem, f(T(τ)) is the temperature compensation function (T is the real - time temperature), which is modeled based on the material thermal expansion coefficient and the temperature - hardness relationship; normalize L(t) to a dimensionless wear parameter where L max is the valve design life threshold; F(τ) represents the axial force of the valve stem (unit: N), which is measured in real time by a torque sensor installed on the valve stem transmission mechanism.

[0064] The valve control and regulation module deploys the Pareto optimal control instruction set to the FPGA hardware, and solidifies the dominant modal basis function and the action value table query logic (after solidification, it can be directly called through the hardware logic, saving the CPU calculation time. The action value table query logic maps the action value table Q-table obtained by policy training to the FPGA lookup table to achieve nanosecond-level control instruction matching), improving the control efficiency; based on the standardized time-series data stream in Step 1, through a dual closed-loop control mechanism (inner-loop PID regulation and outer-loop pressure feedback interpolation), it realizes the precision regulation and anti-disturbance ability of the valve action. On the premise of ensuring the safety constraint of pressure fluctuation, it optimizes the valve response speed and equipment life, and outputs the valve action signal, with the end-to-end delay ≤ 5ms.

[0065] In the embodiment of the present invention, it needs to be further explained that the preprocessed standardized time-series data stream is arranged in chronological order to form a matrix structure: the rows represent the positions of each sensor or grid point (such as 1,000 pressure sensors deployed every 5 meters along the pipeline), and the columns represent the pressure values at each sampling moment (such as collected at 1ms intervals within 2 seconds, a total of 2,000 time points), including the pressure distribution of the entire pipeline;

[0066] The P matrix is decomposed into three matrices through proper orthogonal decomposition, which are respectively:

[0067] The left singular vector, each column represents a spatial fluctuation mode, that is, the modal basis function;

[0068] The diagonal matrix, and the diagonal elements are singular values, indicating the importance of each modal basis function;

[0069] The right singular vector, each row represents the time evolution law of the corresponding modal basis function.

[0070] In the embodiment of the present invention, it needs to be further explained that during the generation process of the spatio-temporal P matrix of the pressure field, Kriging interpolation dynamic compensation is performed on the full-order flow field data and real-time sensing data (sensors at 5m intervals) to unify them to the same spatial resolution; the physical constraint ignores the gravity term; a transfer learning mechanism is introduced in LSTM-PINN, and the full-order flow field data and real-time data distributions are adjusted through the domain adaptation loss function. Using the domain adaptation loss function, the distribution differences between the CFD simulation data (including the assumption of elastic pipe walls) and the real-time sensing data (including strain measurement) are aligned to improve the generalization ability of the prediction model.

[0071] In the embodiment of the present invention, it needs to be further explained that refer to Figure 2 the flow chart for building the transient pressure wave propagation prediction model, and the acquisition method of the transient pressure wave propagation prediction model includes the following steps:

[0072] Step S11, CFD full-order flow field simulation and data matrix construction: Generate full-order flow field data covering typical working conditions through CFD numerical simulation. The inputs include the geometric parameters of the pipe gallery (such as pipe diameter, pipe length) and the physical properties of the fluid (such as viscosity, density, etc.); construct a pressure field snapshot matrix by extracting the spatial pressure distribution of each group of full-order flow field data; the pressure field snapshot matrix is a subset of pressure parameters extracted from the full-order flow field data, which is structured for dimensionality reduction analysis;

[0073] Further, in the embodiment of the present invention, no less than 1,000 groups of full-order flow field data are generated, and the number of grids ≥ 1e 6 , the time step Δt = 1ms, and it is obtained by solving the Navier-Stokes equation (the basic equation describing fluid flow) to simulate typical working conditions (laminar flow, turbulent flow, emergency valve closure);

[0074] Step S12, extraction and orthogonalization of POD modal basis functions: Perform proper orthogonal decomposition (using singular value decomposition in the embodiment of the present invention) on the pressure field snapshot matrix, decompose it into several orthogonal modal basis functions, retain the first k singular values with an energy proportion of 99%, obtain the dominant modal basis functions, and verify the modal orthogonality to ensure that the residual energy proportion does not exceed 1%;

[0075] Explanation: This step reduces the dimensionality of the flow field data, extracts the most representative modal characteristics, reduces the computational complexity, and at the same time maintains the accuracy of fluid mechanics; the modal basis function is understood as the "core pressure fluctuation mode" refined from a large amount of fluid simulation data, similar to restoring complex water pressure changes by combining a few key waveforms; the modal basis function can efficiently characterize the core law of pressure wave propagation in the pipeline;

[0076] Step S13, construct an LSTM (long short-term memory) time series encoder. By taking the pressure, flow rate, and temperature data collected in real time as inputs and combining them with the dominant modal basis functions, construct an LSTM-PINN physics-informed neural network, where the LSTM time series encoder is used to process time series data, and the physical constraint layer ensures that the output conforms to the Navier-Stokes equation; train based on the optimizer and training data to obtain a transient pressure wave propagation prediction model;

[0077] Explanation: This step combines the laws of fluid dynamics with real-time data through a deep learning model to achieve accurate prediction of transient pressure waves and output a transient pressure wave propagation prediction model; among them, the training data acquisition method is:

[0078] Data acquisition: Based on the 1000 groups of full-order flow field data generated in step S11, covering classical working conditions (such as pipe diameters from DN50 to DN1200, viscosity conditions from 0.1 to 500 cP); standardized time-series data streams (pressure, flow rate, temperature, sampling rate 100 kHz) from step one.

[0079] Data division: Divide the acquired data into an offline training set and an online fine-tuning set. The offline training set is used for initial model training; the online fine-tuning set is used for dynamically updating model parameters.

[0080] Data augmentation: Simulate sensor errors by adding Gaussian noise (SNR = 40 dB) to improve the robustness of the model.

[0081] Among them, the full-order flow field data refers to the complete spatial resolution fluid field data obtained through CFD numerical simulation, including:

[0082] Spatial dimension: Transient pressure, flow velocity, and temperature distributions at three-dimensional grid points throughout the pipeline (number of grids ≥ 1×10 6 ).

[0083] Time dimension: The simulation duration covers the complete pressure wave propagation period (such as 0 - 2 s during the emergency valve closure process), and the time step Δt = 1 ms.

[0084] Physical field completeness: Strictly satisfy the Navier - Stokes equation, including multi - flow state characteristics such as laminar flow, turbulent flow, and transient flow.

[0085] Step S14, online dynamic update and accuracy verification: At a set fixed frequency (such as within every 5 minutes), perform incremental learning on the LSTM - PINN network by collecting new sensing data, and adjust the network parameters to adapt to real - time working condition changes; verify and optimize the transient pressure wave propagation prediction model in real - time by calculating the global mean absolute error (MAE) of the pressure wave and the pressure wave peak position error (Δx), ensuring that the error remains within 2%, and the pressure wave peak position error ≤ 0.5 m.

[0086] In the embodiments of the present invention, it needs to be further explained that the physical constraint term L of the domain adaptation loss function of the LSTM - PINN hybrid network phy is:

[0087]

[0088] Among them, λ is the physical constraint weight coefficient, determined by cross - validation (too high λ, such as λ > 0.5 will cause the model to be rigid; too low λ, such as λ < 0.05, then the physical laws cannot be effectively constrained. In the embodiments of the present invention, λ = 0.08); P pred predicted pressure field, u is the flow velocity vector field, ν is the dynamic viscosity coefficient, is the Laplace operator; α i (t) represents the time-dependent modal weight coefficient, which characterizes the contribution degree of the i-th order modal basis function to the pressure field at time t. k represents the number of dominant modal basis functions, and i represents the index; φ i (t) represents the POD modal basis function, which is the dominant spatial distribution pattern of the pressure wave extracted from the full-order flow field data.

[0089] In the embodiments of the present invention, it needs to be further explained that the domain adaptation loss function is constituted as follows:

[0090] L total = ||P pred - P preal || + L phy

[0091] Among them, the data fitting term ||P pred - P preal || means forcing the prediction result to approximate the measured / simulated data. The physical constraint term L phy is used to ensure that the prediction conforms to the hydrodynamic conservation law, avoid the output of the "black box" model from violating physical common sense, ensure that the prediction result conforms to the actual hydrodynamic law, and thus improve the reliability of the prediction.

[0092] In the embodiments of the present invention, it needs to be further explained that the present invention does not make specific restrictions on the quantization form of the multi-objective reward function, as long as it reflects the constraints of pressure fluctuation suppression, valve life optimization, and execution efficiency improvement. Among them, pressure fluctuation suppression is negatively correlated with the real-time pressure deviation, suppressing the risk of pressure overshoot; valve life optimization is negatively correlated with the cumulative mechanical wear of the valve, reducing the equipment loss caused by frequent actions; execution efficiency improvement is positively correlated with the valve action torque efficiency, ensuring the balance between response speed and energy consumption. The weight coefficients of each sub-item are determined through a multi-objective optimization algorithm (such as NSGA-II) to generate a Pareto optimal control strategy set;

[0093] Furthermore, for the convenience of understanding the multi-objective reward function R, the following domain adaptation loss function is listed;

[0094]

[0095] Among them, w1, w2, and w3 are the weight coefficients of each item, and w1 + w2 + w3 = 1.0, which are optimized and determined through the NSGA-II algorithm. The typical values are w1 = 0.6, w2 = 0.3, and w3 = 0.1; ΔP is the absolute deviation between the real-time pressure and the target pressure; L is the cumulative wear of the valve, and L max is the valve design life threshold, τ v (t) is the valve action torque, and G rated is the rated torque of the valve.

[0096] In a possible embodiment, the multi-objective reward function weight coefficients w1, w2, and w3 are changed from static optimization to dynamic adjustment, including: when the viscosity increases, increasing the pressure fluctuation suppression weight w1 to give priority to ensuring safety; when the elastic modulus decreases, increasing the valve life optimization weight w2 to reduce frequent actions.

[0097] In the embodiments of the present invention that need to be further explained, the double-closed-loop anti-control mechanism includes:

[0098] The inner loop control layer is used for high-frequency regulation (e.g., 1 kHz): adopting integral separation PID, dynamically amplifying the control signal according to the valve opening deviation (target and actual); for example, if the target opening is 30%, and the actual opening deviation reaches 5% (i.e., <25% or >35%), the proportional term quickly outputs a strong correction signal; when the opening error ≥ 5%, freeze the integral term (to avoid overshoot caused by continuous error accumulation), and only retain the proportional and derivative actions to prevent the valve from getting stuck at the limit position (such as fully open / fully closed); avoid damage to the actuator due to long-term operation at the limit position and extend the valve life;

[0099] Example scenario: When the pipeline pressure suddenly rises and the valve needs to be quickly closed, the inner loop fine-tunes the opening at a 1 ms period. If the instantaneous error is too large (such as the target closing is 90% and the actual opening is 75%), immediately freeze the integral term and quickly approach the target value only through the proportional term.

[0100] The outer loop control layer is used for global planning (100 Hz): based on real-time pressure feedback (such as the pressure peak position, gradient change), generating a smooth valve opening curve to avoid mechanical shock caused by stepwise actions. For example, when adjusting from 30% opening to 50%, generate a continuous and differentiable S-shaped curve instead of an instantaneous jump; for example, generate a smooth trajectory based on cubic spline interpolation and limit the pressure gradient not to exceed a preset threshold.

[0101] In a possible embodiment, the inner loop PID proportional coefficient K p (t) is positively correlated with the medium viscosity:

[0102]

[0103] where μ(t) represents the real-time medium viscosity, and μ ref represents the preset standard viscosity; K p0 represents the preset inner loop PID proportional coefficient; obtained through the formula: when the viscosity increases, increase the inner loop PID proportional coefficient, which is the core parameter of the control response speed. Increase it when the viscosity increases to compensate for the delay and accelerate the valve response speed (compensate for the delay caused by viscous resistance); when the viscosity decreases, decrease the inner loop PID proportional coefficient to avoid overshoot oscillation;

[0104] Implementation steps: The viscosity sensor collects μ(t) in real time and eliminates noise through moving average filtering; calculate the inner loop PID proportional coefficient K p (t), and update it to the PID control register of the FPGA;

[0105] The outer loop pressure feedback interpolation layer synchronously adjusts the slope of the S-shaped curve to ensure that the pressure gradient does not exceed the preset threshold, for example, does not exceed the fixed threshold of 60% Pmax / m, where P_max / m represents the maximum allowable pressure gradient per meter of the pipeline;

[0106] In a possible embodiment, the fixed threshold is changed to be linked with the physical properties of the medium (viscosity, density). When the viscosity decreases, the threshold is automatically reduced.

[0107] Embodiment 2. The difference between the embodiment of the present invention and Embodiment 1 is that the influence of the pipe wall elasticity on the pressure is considered. Based on this, in the physical constraint term L phy add a pipe wall elasticity correction term to obtain the updated physical constraint term XL phy , and set the domain adaptation loss function based on the updated physical constraint term:

[0108]

[0109] where E(t) represents the real-time inverse pipe wall elastic modulus (measured by a strain sensor), ε(t) represents the pipe wall strain tensor (describing the degree of pipe wall deformation), is the tensor product, representing the coupling effect of the pressure gradient and the strain field, represents the feedback effect of the pipe wall elastic deformation on the pressure wave, and ρ represents the fluid density in the petrochemical pipe gallery;

[0110] In a possible embodiment, after adding the pipe wall elasticity correction term, when performing Kriging interpolation dynamic compensation, the following operations are added:

[0111] Data acquisition: Deploy strain sensors at key positions on the pipe wall (such as one group every 20 meters), measure the local strain ε(t) in real time and calculate the corresponding pipe wall elastic modulus E(t);

[0112] Interpolation inversion: Through the Kriging interpolation dynamic compensation algorithm (based on spatial correlation), interpolate the sparse strain sensor data into the elastic modulus distribution E(x, t) of the entire pipeline;

[0113] Dynamic update: Map the real-time changing pipe wall elastic modulus to the physical constraint layer of LSTM-PINN and correct the wave speed calculation model in real time.

[0114] In one possible embodiment, after adding the pipe wall elasticity correction term, during the training optimization process in step S13, pipe wall strain sensor data is added to the training data, and a fluid-structure coupling simulation data set (such as a CFD case of the interaction between pipe wall vibration and pressure waves) is introduced; using the domain adaptation loss function, the distribution differences between the CFD simulation data (including the elastic pipe wall assumption) and the real-time sensor data (including strain measurement) are aligned to improve the generalization ability of the prediction model.

[0115] Example 3, see Figure 3 The structural block diagram of the petrochemical pipeline corridor intelligent monitoring system based on the domain adaptation loss function is shown. The transfer learning mechanism is introduced into the LSTM-PINN. The full-order flow field data and the real-time data distribution are aligned through the domain adaptation loss function. The physical constraints in the domain adaptation loss function are: any one or a combination of the physical constraints based on dynamic viscosity (see Example 1), the physical constraints with the added pipe wall elasticity correction term (see Example 2), and the physical constraints considering multiphase mixing.

[0116] The difference between this embodiment of the present invention and embodiments 1 and 2 is that the system further includes:

[0117] The multiphase flow feature extraction module deploys dielectric constant sensors and microwave moisture meters in petrochemical pipeline corridors to monitor the volume fractions of oil, gas, and water phases in real time and construct a multiphase flow mixing parameter matrix M, which includes the volume fractions of oil, gas, and water phases, mixed density, and mixed viscosity.

[0118] In the pressure wave propagation prediction module of Example 1, by introducing the multiphase flow modified Navier-Stokes equations into the physical constraint terms of LSTM-PINN, full-order flow field data including phase distribution (such as gas phase volume fraction cloud map, oil-water interface tracking) are generated, and by introducing interphase force terms such as surface tension and interphase momentum exchange terms into the original fluid dynamics equations;

[0119] The physical constraint layer adds residual calculation of the multiphase flow equation to ensure that the predicted pressure field conforms to the multiphase flow conservation law; the VOF (Volume of Fluid) method is used to simulate the dynamics of the phase interface, and the data dimension is expanded to four dimensions (space + time + phase fraction);

[0120] The stratification trend is obtained by calculating the ratio of the maximum spatial gradient of the volume fractions of the oil, gas, and water phases to the arithmetic mean of the volume fractions of each phase. If the stratification trend of the petrochemical pipeline corridor is found to exceed the preset value, adaptive valve control is triggered to adjust the stratification trend by controlling the valve.

[0121] Stratification trend is an important indicator for dynamically quantifying stratification trend, which can accurately reflect the drastic volume change of the fluid in a local area and the overall mixing uniformity.

[0122] Finally, the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things, characterized in that Including: A petrochemical pipe gallery data acquisition module, which deploys a sensor array in the petrochemical pipe gallery, including at least a pressure sensor, an ultrasonic flowmeter, and a temperature sensor, and performs dynamic noise suppression and spatio-temporal alignment processing on the multi-modal sensor data to obtain a standardized time-series data stream; A pressure wave propagation prediction module, which is used to build a transient pressure wave propagation prediction model, extract the P matrix of the pressure field spatio-temporal from the standardized time-series data stream, perform singular value decomposition on the P matrix by proper orthogonal decomposition to obtain the dominant modal basis function; train the LSTM-PINN physics-informed neural network with the full-order flow field data generated by simulation and the standardized time-series data stream to obtain the transient pressure wave propagation prediction model; A valve control strategy generation module, based on the pressure wave prediction result of the transient pressure wave propagation prediction model, defines a state vector including pressure gradient, valve wear, and energy efficiency ratio; optimizes the weights of the multi-objective reward function by the NSGA-II algorithm, and trains a distributed Q-learning strategy library through a prioritized experience replay mechanism to generate a Pareto optimal control instruction set; A valve control adjustment module deploys the Pareto optimal control instruction set to the FPGA hardware, and solidifies the dominant modal basis function and the action value table query logic; based on the standardized time-series data stream, realizes the precision adjustment and anti-disturbance ability of the valve action through a double closed-loop control mechanism including inner-loop PID adjustment and outer-loop pressure feedback interpolation, and outputs a valve action signal.

2. The intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to claim 1, characterized in that, The acquisition method of the P matrix of the pressure field spatio-temporal includes: Arrange the preprocessed standardized time-series data stream in chronological order to form a matrix structure; the rows represent the positions of each sensor or grid point; the columns represent the pressure values at each sampling moment, including the pressure distribution of the entire pipeline; Decompose the P matrix into three matrices by proper orthogonal decomposition, which are respectively: The left singular vector, each column represents a spatial fluctuation mode, that is, the modal basis function; The diagonal matrix, the diagonal elements are singular values, indicating the importance of each modal basis function; The right singular vector, each row represents the time evolution law of the corresponding modal basis function.

3. The intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things according to claim 2, wherein, During the generation process of the P matrix of the pressure field spatio-temporal, perform Kriging interpolation dynamic compensation on the full-order flow field data and real-time sensing data to unify them to the same spatial resolution; the physical constraint ignores the gravity term; introduce a transfer learning mechanism in the LSTM-PINN, and adapt the full-order flow field data and real-time data distributions through the domain adaptation loss function.

4. The intelligent monitoring system for petrochemical pipe galleries based on the Internet of Things according to claim 3, characterized in that The acquisition method of the transient pressure wave propagation prediction model includes the following steps: Step S11, CFD full-order flow field simulation and data matrix construction: Use CFD numerical simulation to generate full-order flow field data covering typical working conditions, and the input includes pipe gallery geometric parameters and fluid physical property parameters; simulate the typical working conditions by solving the Navier-Stokes equation to obtain the simulation data of the typical working conditions; construct a pressure field snapshot matrix by extracting the spatial pressure distribution of the full-order flow field data; Step S12, POD modal basis function extraction and orthogonalization: Perform proper orthogonal decomposition on the pressure field snapshot matrix, disassemble it into several orthogonal modal basis functions, obtain the dominant modal basis function, and verify the modal orthogonality; Step S13: Construct an LSTM time-series encoder. By using the pressure, flow rate, and temperature data collected in real time as inputs and combining them with the dominant mode basis functions, construct an LSTM-PINN physics-informed neural network. The LSTM time-series encoder is used to process time-series data, and the physics constraint layer ensures that the output conforms to the Navier-Stokes equations; train based on the optimizer and training data to obtain a transient pressure wave propagation prediction model; Step S14: Online dynamic update and accuracy verification: At the set fixed frequency, perform incremental learning on the LSTM-PINN network by collecting new sensing data, and adjust the network parameters to adapt to real-time working condition changes; verify and optimize the transient pressure wave propagation prediction model in real time by calculating the global mean absolute error of the pressure wave and the pressure wave peak position error.

5. The intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to claim 4, characterized in that, The physical constraint term L of the domain adaptation loss function of the LSTM-PINN hybrid network phy is as follows: Among them, λ is the physical constraint weight coefficient, which is determined by cross-validation; P pred represents the predicted pressure field, u is the velocity vector field, ν is the dynamic viscosity coefficient, is the Laplace operator, α i (t) represents the time-dependent modal weight coefficient, which characterizes the contribution degree of the i-th order modal basis function to the pressure field at time t, k represents the number of dominant modal basis functions, and i represents the index; φ i (t) represents the POD modal basis function, which is the dominant pressure wave spatial distribution pattern extracted from the full-order flow field data.

6. The intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to claim 5, characterized in that, The domain adaptation loss function is constructed as follows: L total = ‖P pred - P preal ‖ + L phy Among them, the data fitting term ‖P pred -P preal ‖ represents forcing the prediction result to approximate the measured data, and the physical constraint term L phy is used to ensure that the prediction conforms to the hydrodynamic conservation law.

7. An intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to any one of claims 1-4, characterized in that, The multi-objective reward function is a weighted sum term for pressure fluctuation suppression, valve life optimization, and execution efficiency improvement. Among them, pressure fluctuation suppression is negatively correlated with the real-time pressure deviation, suppressing the risk of pressure overshoot; valve life optimization is negatively correlated with the cumulative mechanical wear of the valve, reducing equipment losses caused by frequent actions; execution efficiency improvement is positively correlated with the valve action torque efficiency, ensuring the balance between response speed and energy consumption. Determine the weight coefficients of each sub-item through a multi-objective optimization algorithm to generate a Pareto optimal control strategy set.

8. The intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to claim 5, characterized in that, In the physical constraint term L phy add a pipe wall elasticity correction term to obtain the updated physical constraint term XL phy : Among them, E(t) represents the elastic modulus of the pipe wall obtained by real-time inversion, which is obtained by deploying strain sensors at key positions on the pipe wall; ∈(t) represents the strain tensor of the pipe wall, is the tensor product, representing the coupling effect between the pressure gradient and the strain field, represents the feedback effect of the elastic deformation of the pipe wall on the pressure wave, and ρ represents the fluid density in the petrochemical pipe gallery.

9. An intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to any one of claims 1-4, characterized in that, The double closed-loop control mechanism includes: The inner loop control layer for high-frequency regulation: Adopt integral separation PID to dynamically amplify the control signal according to the deviation between the target and actual opening of the valve; if the deviation between the target and actual opening is greater than the preset value, the proportional term quickly outputs a strong correction signal; freeze the integral term and only retain the proportional and derivative actions to prevent the valve from being stuck at the limit position; The outer loop control layer for global planning: Based on the real-time pressure feedback, generate a smooth valve opening curve to avoid mechanical shocks caused by stepwise actions and limit the pressure gradient not to exceed the preset threshold.

10. The intelligent monitoring system for petrochemical pipe corridors based on the Internet of Things according to claim 8, characterized in that, During the inner loop PID regulation process, the inner loop PID proportional coefficient K p (t) is positively correlated with the medium viscosity. When the viscosity increases, the inner loop PID proportional coefficient is increased, which is the core parameter for controlling the response speed. When the viscosity increases, it is increased to compensate for the delay and accelerate the valve response speed to compensate for the delay caused by viscous resistance. When the viscosity decreases, the inner loop PID proportional coefficient is decreased to avoid overshoot oscillation.

Images