Double-gate valve opening and closing energy-saving control method based on intelligent algorithm

Abstract

The invention discloses a double-gate valve opening and closing energy-saving control method based on an intelligent algorithm, and the method comprises the following steps: S1, collecting the working condition data of a double-gate valve, and obtaining a working condition feature data vector; s2, inputting the working condition feature data vector into an improved physical information neural network model; s3, generating a sub-network through a control strategy, and outputting a valve opening and closing control track based on the working condition feature data vector; s4, inputting the control track and the working condition characteristic data vector into a physical field simulation sub-network to obtain a physical simulation result; s5, inputting the physical simulation result and the working condition feature data vector into the equipment health assessment sub-network to obtain a health assessment result; s6, constructing a multi-objective loss function to obtain an optimal control trajectory; and S7, valve opening and closing actions are carried out, feedback data are collected in real time, and parameters of the physical information neural network model are dynamically adjusted and improved. The improved physical information neural network is adopted, and intelligent and efficient control over the double-gate valve is achieved.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation and intelligent control technology, and in particular to an energy-saving control method for opening and closing a double-gate valve based on an intelligent algorithm. Background Art

[0002] In modern industrial automation systems, double-disc valves are widely used as fluid control components in processes across the petroleum, chemical, municipal, and metallurgical sectors. Their primary function is to control the opening and closing of fluids in pipelines and regulate flow. As industrial production demands increased energy conservation, emission reduction, and safe equipment operation, achieving efficient, energy-saving, and intelligent control of double-disc valves has become a key research area.

[0003] Currently, traditional methods for controlling the opening and closing of double-disc valves rely on preset opening and closing strategies or control logic based on empirical rules. These methods typically complete valve opening and closing operations by collecting information about pressure, flow, and opening, combined with manually set thresholds or logical judgments. Although basic automated control is achieved, complex operating conditions and the influence of multiple variables make it difficult to balance energy-saving optimization and equipment health maintenance, which can easily lead to increased energy consumption or increased equipment wear. Furthermore, some methods focus solely on the apparent state of the valve, lacking modeling and real-time assessment of internal physical processes and health conditions, resulting in insufficient energy efficiency and delayed fault prediction.

[0004] With the development of artificial intelligence and deep learning technologies, some research has introduced neural network models into the field of valve control, leveraging historical operating data to model and predict the opening and closing process. This approach has made progress in improving control accuracy and achieving certain levels of adaptability. However, traditional neural networks are mostly "black box" models with limited consideration of physical constraints, making it easy for model outputs to be inconsistent with the actual physical process. Furthermore, existing data-driven approaches lack generalization and robustness when faced with equipment anomalies, changing operating conditions, or a lack of historical data, making it difficult to meet the safety and reliability requirements of industrial scenarios.

[0005] When it comes to health status assessment and lifespan prediction, existing technologies often employ separate health monitoring and lifespan estimation methods, relying on offline data analysis or periodic manual inspections. This makes it difficult to accurately reflect equipment health and failure risks in a timely manner, limiting the dynamic adjustment of startup and shutdown strategies. Furthermore, some health assessment methods lack coupling with control strategies, making it difficult to implement health-based adaptive control, impacting system energy efficiency.

[0006] While recent advances in physical information neural networks have enabled the integration of physical laws with neural network-based data-physics coupling modeling, the integration of these systems in practical engineering applications remains immature. Existing systems commonly suffer from insufficient model coupling, imperfect optimization processes, and limited feedback loop capabilities, making it difficult to balance energy efficiency, equipment health, and control accuracy. Traditional methods struggle to achieve real-time adaptive adjustments in complex operating conditions and with aging equipment, impacting system operational efficiency and lifespan management.

[0007] Therefore, how to provide an energy-saving control method for opening and closing a double-gate valve based on an intelligent algorithm is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] One purpose of the present invention is to propose an energy-saving control method for the opening and closing of a double-gate valve based on an intelligent algorithm. The present invention fully integrates physical information neural networks, multi-objective joint optimization, equipment health assessment and real-time closed-loop adaptive control technology, and describes in detail the algorithm process of feature extraction, physical field simulation, health status assessment and optimal control trajectory generation for the opening and closing process of the double-gate valve. It has the advantages of significant energy-saving effect, high control accuracy, strong equipment life management capability and excellent adaptability.

[0009] According to an embodiment of the present invention, a method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm includes the following steps:

[0010] S1. Collecting the working condition data of the double gate valve, normalizing and extracting the working condition data, and obtaining the working condition characteristic data vector;

[0011] S2. Inputting the operating condition characteristic data vector into an improved physical information neural network model, wherein the improved physical information neural network model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork;

[0012] S3. Generate a subnetwork through the control strategy and output the valve opening and closing control trajectory based on the working condition characteristic data vector;

[0013] S4, inputting the control trajectory and working condition characteristic data vector into the physical field simulation sub-network, and obtaining the physical simulation results based on the coupling of the neural network and the physical equation;

[0014] S5. Input the physical simulation results and the working condition characteristic data vector into the equipment health assessment subnetwork to obtain a health assessment result;

[0015] S6. Based on the physical simulation results and health assessment results, a multi-objective loss function is constructed to obtain the optimal control trajectory by end-to-end joint optimization of the above sub-network parameters;

[0016] S7. Send the optimal control trajectory to the valve actuator to perform valve opening and closing actions, collect feedback data in real time, and dynamically adjust and improve the parameters of the physical information neural network model.

[0017] Optionally, the operating condition characteristic data vector X includes the inlet pressure P after normalization and feature extraction. in , outlet pressure P out , flow rate Q, valve opening A, opening and closing speed V, ambient temperature T env , equipment wear status D and historical operation data H.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Inputting the operating condition characteristic data vector X into an improved physical information neural network model, wherein the model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork;

[0020] S22. In the control strategy generation subnetwork, feature extraction and time series modeling are performed on the operating condition feature data vector X to generate and output the valve opening and closing control trajectory;

[0021] S23. In the physical field simulation subnetwork, the operating condition characteristic data vector X and the valve opening and closing control trajectory are used as inputs, physical information coupling simulation is performed in combination with physical constraints, and the physical simulation results are output;

[0022] S24. In the equipment health assessment subnetwork, the operating condition characteristic data vector X and the physical simulation results are used as input to perform state analysis and time series modeling, and output the health assessment results.

[0023] Optionally, the S3 specifically includes:

[0024] S31, inputting the operating condition feature data vector into the control strategy generation subnetwork, wherein the input into the control strategy generation subnetwork includes a feature extraction layer, a time series modeling layer, and a decision generation layer;

[0025] S32, performing feature encoding on the operating condition feature data vector at the feature extraction layer to obtain an operating condition feature representation;

[0026] S33, the working condition feature representation is input to the time series modeling layer, and the gated recurrent unit is used to perform time series modeling on the working condition feature to obtain the intermediate time series feature;

[0027] S34, input the intermediate time series feature S to the decision generation layer, and output the valve opening and closing control trajectory Γ=[A(t1),A(t2),…,A(t i ),…,A(t n )], where A(t i ) is the time t ivalve opening.

[0028] Optionally, the S4 specifically includes:

[0029] S41, inputting the valve opening and closing control trajectory Γ and the operating condition characteristic data vector X into the input layer of the physical field simulation subnetwork;

[0030] S42, through the feature fusion layer, the valve opening and closing control trajectory Γ is spliced ​​with the working condition feature data vector X to obtain a fused feature vector Z;

[0031] S43, inputting the fused feature vector Z into the deep neural network modeling layer, and performing nonlinear mapping on the fused feature vector Z using a multi-layer perceptron structure to obtain an intermediate feature representation;

[0032] S44. In the physical equation coupling layer, the intermediate feature representation is coupled with the physical equation, and the physical consistency constraint is imposed on the physical quantity prediction results output at each moment by introducing a physical constraint loss function.

[0033] S45, output the physical simulation result Y=[y1,y2,…,y i ,…,y n ], where y i is the physical simulation quantity obtained at the i-th moment, and the physical simulation quantity includes the valve flow, pressure distribution and energy consumption at the corresponding moment.

[0034] Optionally, the S44 specifically includes:

[0035] S441. Input the intermediate feature representation output by the deep neural network modeling layer into the physical equation coupling layer to obtain the physical quantity prediction result at each moment;

[0036] S442. Based on the valve opening and closing control trajectory and the operating condition characteristic data vector, a preset physical equation is used to calculate the theoretical reference value at each moment, wherein the theoretical reference value includes a theoretical valve flow rate, a theoretical pressure distribution, and a theoretical energy consumption;

[0037] S443. The theoretical valve flow rate is calculated by multiplying the valve flow coefficient by the valve opening at the i-th moment, and then multiplying it by the square root of the ratio of the difference between the inlet pressure and the outlet pressure to the fluid density ρ;

[0038] S444. The theoretical pressure distribution is calculated by subtracting the valve pressure drop determined by the valve resistance coefficient, the theoretical valve flow rate, and the valve opening, and the pipeline pressure drop determined by the pipeline friction factor f, length L, diameter D, fluid density ρ, and flow velocity V, from the inlet pressure.

[0039] S445, Theoretical energy consumption is calculated by multiplying the theoretical valve flow by the difference between the inlet pressure and the outlet pressure, and then multiplying it by the time step t s Multiply and finally divide by the energy conversion efficiency η;

[0040] S446. Introduce the physical simulation loss function into the physical equation coupling layer to impose physical consistency constraints on the physical quantity prediction results at each moment:

[0041]

[0042] Among them, L phy is the physical constraint loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, Q i is the valve flow output at the i-th moment, Q i,ref is the theoretical valve flow at the i-th moment, P i is the pressure distribution output at the i-th moment, P i,ref is the theoretical pressure distribution at the i-th moment, E i is the energy consumption output at the i-th moment, E i,ref is the theoretical energy consumption at the i-th moment.

[0043] Optionally, the S5 specifically includes:

[0044] S51. Splicing the physical simulation results with the working condition characteristic data vector, inputting the results into the equipment health assessment sub-network, performing feature extraction through a multi-layer neural network structure, and obtaining feature extraction results;

[0045] S52, input the feature extraction result into the equipment health status classification branch, use the Softmax activation function to classify and judge the equipment health status, and obtain the probability of each health status category [p i,1 ,p i,2 ,...,p i,C ], where C is the number of health status categories;

[0046] S53. Based on the probabilities of the health status categories, the category label corresponding to the health status category with the largest probability value is used as the device health status indicator;

[0047] S54: Input the feature extraction result into the equipment remaining life prediction branch to obtain the equipment remaining life prediction value and combine it with the equipment health status indicator to output as the health assessment result.

[0048] Optionally, the S6 specifically includes:

[0049] S61. Using the physical simulation results output by the physical simulation subnetwork and the health assessment results output by the device health assessment subnetwork as inputs to the multi-objective loss function;

[0050] S62. Define the health assessment loss function:

[0051]

[0052] Among them, L health is the health assessment loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, β1 is the weight coefficient of the equipment health status indicator loss term, β2 is the weight coefficient of the equipment remaining life prediction value loss term, S i is the equipment health status indicator at the i-th moment, is the true label of the health status of the device at the i-th moment, is the cross entropy loss between the device health status indicator and the true label of the device health status, L i is the predicted value of the remaining life of the equipment at the i-th moment, is the true value of the remaining life of the equipment at the i-th moment;

[0053] S63. Construct a multi-objective loss function:

[0054] L total =λ1·L phy +λ2·L health ;

[0055] Among them, L total is the multi-objective loss function, λ1 is the weight coefficient of the physical simulation loss function, L phy is the physical simulation loss function, λ2 is the weight coefficient of the health assessment loss function;

[0056] S64. Based on the multi-objective loss function, an end-to-end joint optimization method is used to jointly optimize the parameters of the physical field simulation subnetwork and the equipment health assessment subnetwork to output the optimal control trajectory.

[0057] Optionally, the joint optimization includes inputting initial state parameters and initial trajectories of control variables, and iteratively optimizing the trajectories of the control variables based on a multi-objective loss function;

[0058] In each iteration, the physical simulation sub-network and the equipment health assessment sub-network are used to calculate the physical simulation loss function and the health assessment loss function, and the control variable trajectory is updated. When the convergence condition is met, the iteration is stopped and the current control variable trajectory is taken as the optimal control trajectory, where ∈ is the preset convergence threshold and k is the number of iterations.

[0059] Optionally, the S7 specifically includes:

[0060] S71. Send the obtained optimal control trajectory to the valve actuator, driving the valve to open and close according to the optimal control trajectory;

[0061] S72. Collect feedback data from the valve system in real time through sensors, wherein the feedback data includes valve status parameters, equipment health status indicators, and equipment remaining life prediction values;

[0062] S73. Based on the collected feedback data, dynamically update the parameters of the physical information neural network model, and recalculate the physical simulation loss function and the health assessment loss function;

[0063] S74. According to the updated physical information neural network model and the actual operating status of the system, the optimal control trajectory is adjusted in real time and sent to the valve actuator again to form a closed-loop control.

[0064] The beneficial effects of the present invention are:

[0065] The present invention achieves efficient and precise management of the opening and closing process of a double-disc valve by organically combining a physical information neural network, a multi-objective optimization algorithm, equipment health assessment, and closed-loop adaptive control technology. Unlike traditional control methods based on empirical formulas or single data drives, the present invention is able to integrate the physical laws of the valve opening and closing process with real-time operating data to establish a control model that better conforms to actual operating conditions. The introduction of a physical information neural network effectively improves the model's ability to fit the dynamic process of valve opening and closing, while enhancing the model's interpretability and stability, allowing the system to maintain high control accuracy even under complex operating conditions.

[0066] This invention demonstrates significant advantages in multi-objective optimization. By jointly optimizing key performance indicators (KPIs) such as energy consumption, valve response speed, and equipment lifespan, it can dynamically adjust the valve's opening and closing trajectory to meet varying application requirements, balancing energy conservation and equipment protection. The system also collects and analyzes valve operating status in real time, adaptively adjusting control parameters based on health assessment results, effectively reducing increased energy consumption and the risk of equipment failures due to aging or abnormal operating conditions. This mechanism extends valve lifespan, improving equipment reliability and maintenance efficiency.

[0067] Furthermore, the closed-loop adaptive feedback control mechanism proposed in this invention enables the system to continuously optimize its control strategy based on real-time on-site feedback, enabling rapid response and intelligent adjustment to dynamic operating conditions. This allows the valve to maintain stable and efficient operation under a variety of complex and changing environmental conditions, effectively improving the control system's energy efficiency and operational safety. Overall, this invention provides strong technical support for enhancing the intelligent control level and operational reliability of double-disc valves, helping to promote the development of industrial automation systems towards high efficiency, energy conservation, and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0069] Figure 1 This is a flow chart of an energy-saving control method for opening and closing a double-gate valve based on an intelligent algorithm proposed by the present invention;

[0070] Figure 2 This is a structural diagram of the physical information neural network modeling and multi-objective optimization control used in the energy-saving control method for opening and closing of a double-gate valve based on an intelligent algorithm proposed in the present invention;

[0071] Figure 3 This is a schematic diagram of the functional modules for equipment health assessment and adaptive control parameter adjustment in the energy-saving control method for opening and closing of a double-gate valve based on an intelligent algorithm proposed in the present invention. DETAILED DESCRIPTION

[0072] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0073] refer to Figure 1-3 A method for controlling the opening and closing of a double-disc valve with energy saving based on an intelligent algorithm comprises the following steps:

[0074] S1. Collecting the working condition data of the double gate valve, normalizing and extracting the working condition data, and obtaining the working condition characteristic data vector;

[0075] S2. Inputting the operating condition characteristic data vector into an improved physical information neural network model, wherein the model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork;

[0076] S3. Generate a subnetwork through the control strategy and output the valve opening and closing control trajectory based on the working condition characteristic data vector;

[0077] S4, inputting the control trajectory and working condition characteristic data vector into the physical field simulation sub-network, and obtaining the physical simulation results based on the coupling of the neural network and the physical equation;

[0078] S5. Input the physical simulation results and the working condition characteristic data vector into the equipment health assessment subnetwork to obtain a health assessment result;

[0079] S6. Based on the physical simulation results and health assessment results, a multi-objective loss function is constructed to obtain the optimal control trajectory by end-to-end joint optimization of the above sub-network parameters;

[0080] S7. Send the optimal control trajectory to the valve actuator to perform valve opening and closing actions, collect feedback data in real time, and dynamically adjust and improve the parameters of the physical information neural network model.

[0081] The present invention proposes an energy-saving control method for the opening and closing of a double-gate valve based on an intelligent algorithm. By introducing an improved physical information neural network, intelligent and adaptive control of the opening and closing process of the double-gate valve is achieved. The accuracy of data processing is improved by normalizing operating condition data and extracting features. By coupling the neural network with physical equations, the physical response of the valve under different operating conditions can be fully simulated, and the health status of the equipment can be accurately assessed. The end-to-end joint optimization of the multi-objective loss function effectively takes into account key indicators such as energy consumption, response speed, and equipment life, thereby outputting the optimal control trajectory. Practical application results show that the present invention not only significantly reduces the energy consumption of the valve and improves the opening and closing efficiency, but also improves the accuracy of equipment abnormality warnings and fault detection rates, extends the maintenance cycle, and reduces operation and maintenance costs. Through real-time data collection and feedback adjustment, the system has continuous self-learning and adaptive capabilities, can dynamically respond to complex and changing operating conditions, and significantly improves the safety, reliability, and economy of the double-gate valve operation.

[0082] In this embodiment, the working condition characteristic data vector X includes the inlet pressure P after normalization and feature extraction. in , outlet pressure P out , flow rate Q, valve opening A, opening and closing speed V, ambient temperature T env , equipment wear status D and historical operation data H.

[0083] The present invention improves the precision of double-disc valve control and the accuracy of equipment health assessment by extracting and normalizing features of inlet pressure, outlet pressure, flow, valve opening, opening and closing speed, ambient temperature, equipment wear status and historical operating data, thereby achieving efficient and safe intelligent adaptive control.

[0084] In this embodiment, S2 specifically includes:

[0085] S21. Inputting the operating condition characteristic data vector X into an improved physical information neural network model, wherein the model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork;

[0086] S22. In the control strategy generation subnetwork, feature extraction and time series modeling are performed on the operating condition feature data vector X to generate and output the valve opening and closing control trajectory;

[0087] S23. In the physical field simulation subnetwork, the operating condition characteristic data vector X and the valve opening and closing control trajectory are used as inputs, physical information coupling simulation is performed in combination with physical constraints, and the physical simulation results are output;

[0088] S24. In the equipment health assessment subnetwork, the operating condition characteristic data vector X and the physical simulation results are used as input to perform state analysis and time series modeling, and output the health assessment results.

[0089] The present invention improves the physical information neural network model by inputting the operating condition characteristic data vector, and uses the three sub-networks of control strategy generation, physical field simulation, and equipment health assessment to work together to achieve intelligent optimization of the valve opening and closing process. The control strategy generation sub-network performs deep feature extraction and time series modeling, and outputs efficient opening and closing trajectories; the physical field simulation sub-network achieves accurate simulation of actual operating conditions by introducing physical constraints, effectively improving the reliability and adaptability of the control strategy; the equipment health assessment sub-network realizes dynamic monitoring and accurate assessment of equipment status based on multi-source data. Overall, the present invention improves the operating efficiency and safety of the double-gate valve, and enhances the system's adaptability and health management level.

[0090] In this embodiment, S3 specifically includes:

[0091] S31, inputting the operating condition feature data vector into the control strategy generation subnetwork, wherein the input into the control strategy generation subnetwork includes a feature extraction layer, a time series modeling layer, and a decision generation layer;

[0092] S32, performing feature encoding on the operating condition feature data vector at the feature extraction layer to obtain an operating condition feature representation;

[0093] S33, the working condition feature representation is input to the time series modeling layer, and the gated recurrent unit is used to perform time series modeling on the working condition feature to obtain the intermediate time series feature;

[0094] S34, input the intermediate time series feature S to the decision generation layer, and output the valve opening and closing control trajectory Γ=[A(t1),A(t2),…,A(t i ),…,A(t n )], where A(t i ) is the time t i valve opening.

[0095] The present invention achieves deep encoding of operating condition feature data and temporal relationship modeling by setting a feature extraction layer, a timing modeling layer, and a decision generation layer in the control strategy generation subnetwork. A gated recurrent unit is used to perform temporal analysis of operating condition features, effectively capturing the dynamic impact of operating condition changes on valve control and improving the model's adaptability to complex operating environments. A multi-layer perceptron is used to generate high-precision valve opening and closing control trajectories to ensure the smoothness and response speed of valve action. Overall, the present invention optimizes the accuracy and real-time performance of valve control, helping to improve the system's operating efficiency and safety assurance level.

[0096] In this embodiment, the S4 specifically includes:

[0097] S41, inputting the valve opening and closing control trajectory Γ and the operating condition characteristic data vector X into the input layer of the physical field simulation subnetwork;

[0098] S42, through the feature fusion layer, the valve opening and closing control trajectory Γ is spliced ​​with the working condition feature data vector X to obtain a fused feature vector Z;

[0099] S43, inputting the fused feature vector Z into the deep neural network modeling layer, and performing nonlinear mapping on the fused feature vector Z using a multi-layer perceptron structure to obtain an intermediate feature representation;

[0100] S44. In the physical equation coupling layer, the intermediate feature representation is coupled with the physical equation, and the physical consistency constraint is imposed on the physical quantity prediction results output at each moment by introducing a physical constraint loss function.

[0101] S45, output the physical simulation result Y=[y1,y2,…,y i ,…,y n ], where y i is the physical simulation quantity obtained at the i-th moment, and the physical simulation quantity includes the valve flow, pressure distribution and energy consumption at the corresponding moment.

[0102] In the physical field simulation subnetwork, this invention achieves effective integration of multi-source information by fusing valve opening and closing control trajectories with operating condition characteristics. A multi-layer perceptron is used to perform deep nonlinear modeling of the fused features, improving the model's ability to represent complex operating conditions. The introduction of a physical equation coupling layer and a physical constraint loss function imposes physical consistency constraints on the physical simulation quantities, improving the accuracy and reliability of the simulation results. The resulting outputs of valve flow, pressure distribution, and energy consumption provide a high-quality physical foundation for subsequent equipment health assessment and control strategy optimization, significantly enhancing the overall system performance and operational safety.

[0103] In this embodiment, the S44 specifically includes:

[0104] S441. Input the intermediate feature representation output by the deep neural network modeling layer into the physical equation coupling layer to obtain the physical quantity prediction result at each moment;

[0105] S442. Based on the valve opening and closing control trajectory and the operating condition characteristic data vector, a preset physical equation is used to calculate the theoretical reference value at each moment, wherein the theoretical reference value includes a theoretical valve flow rate, a theoretical pressure distribution, and a theoretical energy consumption;

[0106] S443. The theoretical valve flow rate is calculated by multiplying the valve flow coefficient by the valve opening at the i-th moment, and then multiplying it by the square root of the ratio of the difference between the inlet pressure and the outlet pressure to the fluid density ρ;

[0107] S444. The theoretical pressure distribution is calculated by subtracting the valve pressure drop determined by the valve resistance coefficient, the theoretical valve flow rate, and the valve opening, and the pipeline pressure drop determined by the pipeline friction factor f, length L, diameter D, fluid density ρ, and flow velocity V, from the inlet pressure.

[0108] S445, Theoretical energy consumption is calculated by multiplying the theoretical valve flow by the difference between the inlet pressure and the outlet pressure, and then multiplying it by the time step t s Multiply and finally divide by the energy conversion efficiency η;

[0109] S446. Introduce the physical simulation loss function into the physical equation coupling layer to impose physical consistency constraints on the physical quantity prediction results at each moment:

[0110]

[0111] Among them, L phy is the physical constraint loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, Q i is the valve flow output at the i-th moment, Q i,ref is the theoretical valve flow at the i-th moment, P i is the pressure distribution output at the i-th moment, P i,ref is the theoretical pressure distribution at the i-th moment, E i is the energy consumption output at the i-th moment, E i,ref is the theoretical energy consumption at the i-th moment.

[0112] The present invention achieves high-precision prediction of the physical quantities of valve operating conditions by coupling the intermediate feature representation output by the deep neural network with the physical equation. At the physical equation coupling layer, based on the operating condition characteristics and control trajectory, the theoretical flow rate, pressure distribution and energy consumption are calculated as reference values. The flow rate is calculated using the valve flow coefficient combined with the real-time opening and pressure difference. The pressure distribution is estimated based on the principles of fluid dynamics and pipeline parameters. The energy consumption is integrated in the time domain by multiplying the pressure difference and the flow rate to reflect the energy conversion efficiency. In order to ensure that the physical quantity output by the neural network is consistent with the theoretical benchmark, the system introduces a physical constraint loss function. This loss function quantifies the difference between the model prediction value and the theoretical reference value at each moment, and measures the physical consistency of the flow rate, pressure and energy consumption by accumulating the square difference. The introduction of this loss function effectively constrains the model learning process, improves the accuracy and physical reliability of the simulation results, provides solid data support for subsequent health assessment and intelligent control, and significantly enhances the practicality and safety assurance capabilities of the system.

[0113] In this embodiment, the S5 specifically includes:

[0114] S51. Splicing the physical simulation results with the working condition characteristic data vector, inputting the results into the equipment health assessment sub-network, performing feature extraction through a multi-layer neural network structure, and obtaining feature extraction results;

[0115] S52, input the feature extraction result into the equipment health status classification branch, use the Softmax activation function to classify and judge the equipment health status, and obtain the probability of each health status category [p i,1 ,p i,2 ,...,p i,C ], where C is the number of health status categories;

[0116] S53. Based on the probabilities of the health status categories, the category label corresponding to the health status category with the largest probability value is used as the device health status indicator;

[0117] S54: Input the feature extraction result into the equipment remaining life prediction branch to obtain the equipment remaining life prediction value and combine it with the equipment health status indicator to output as the health assessment result.

[0118] This method fuses physical simulation results with operating condition characteristic data vectors and inputs them into the device health assessment subnetwork, leveraging a multi-layer neural network to deeply extract and analyze device status characteristics. Combining health status classification and remaining life prediction, this approach accurately identifies device health status categories and effectively predicts remaining life. The resulting health assessment provides a quantitative and scientific basis for decision-making in device operation and maintenance, helping to identify potential failure risks in advance, optimize maintenance plans, extend equipment life, and improve the safety and reliability of system operations.

[0119] In this embodiment, S6 specifically includes:

[0120] S61. Using the physical simulation results output by the physical simulation subnetwork and the health assessment results output by the device health assessment subnetwork as inputs to the multi-objective loss function;

[0121] S62. Define the health assessment loss function:

[0122]

[0123] Among them, L health is the health assessment loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, β1 is the weight coefficient of the equipment health status indicator loss term, β2 is the weight coefficient of the equipment remaining life prediction value loss term, S i is the equipment health status indicator at the i-th moment, is the true label of the health status of the device at the i-th moment, is the cross entropy loss between the device health status indicator and the true label of the device health status, L i is the predicted value of the remaining life of the equipment at the i-th moment, is the true value of the remaining life of the equipment at the i-th moment;

[0124] S63. Construct a multi-objective loss function:

[0125] L total =λ1·L phy +λ2·L health ;

[0126] Among them, L total is the multi-objective loss function, λ1 is the weight coefficient of the physical simulation loss function, L phy is the physical simulation loss function, λ2 is the weight coefficient of the health assessment loss function;

[0127] S64. Based on the multi-objective loss function, an end-to-end joint optimization method is used to jointly optimize the parameters of the physical field simulation subnetwork and the equipment health assessment subnetwork to output the optimal control trajectory.

[0128] The present invention realizes comprehensive constraints on the model training process by taking the physical simulation results output by the physical simulation subnetwork and the health assessment results output by the equipment health assessment subnetwork as the input of the multi-objective loss function. The health assessment loss function consists of two parts: health status classification loss and remaining life regression loss. The health status classification loss adopts the cross entropy function to measure the difference between the equipment health status prediction and the true label, thereby improving the classification accuracy; the remaining life regression loss adopts the mean square error to measure the deviation between the equipment remaining life prediction value and the actual value, thereby enhancing the accuracy of life prediction. The multi-objective loss function comprehensively optimizes the physical simulation loss and the health assessment loss by introducing weight coefficients, taking into account the performance of both physical consistency and health assessment. This method improves the generalization ability and prediction reliability of the model, provides theoretical support for intelligent equipment management and preventive maintenance, and promotes the safety and economy of system operation.

[0129] In this embodiment, the joint optimization includes inputting initial state parameters and initial trajectories of control variables, and iteratively optimizing the trajectories of control variables based on a multi-objective loss function;

[0130] In each iteration, the physical simulation sub-network and the equipment health assessment sub-network are used to calculate the physical simulation loss function and the health assessment loss function, and the control variable trajectory is updated. When the convergence condition is met, the iteration is stopped and the current control variable trajectory is taken as the optimal control trajectory, where ∈ is the preset convergence threshold and k is the number of iterations.

[0131] This paper uses a joint optimization method, taking the initial state parameters and initial control variable trajectories as inputs and iteratively optimizing the control variable trajectories based on a multi-objective loss function. In each iteration, the loss function is calculated using a physical simulation network and an equipment health assessment network, respectively, and the control variable trajectories are updated. When the change in the multi-objective loss function falls below a set threshold, the iteration is terminated and the current optimal control trajectory is output. This method achieves a coordinated optimization of system performance and equipment health, providing a scientific basis for precise control and intelligent maintenance.

[0132] In this embodiment, the S7 specifically includes:

[0133] S71. Send the obtained optimal control trajectory to the valve actuator, driving the valve to open and close according to the optimal control trajectory;

[0134] S72. Collect feedback data from the valve system in real time through sensors, wherein the feedback data includes valve status parameters, equipment health status indicators, and equipment remaining life prediction values;

[0135] S73. Based on the collected feedback data, dynamically update the parameters of the physical information neural network model and recalculate the physical simulation loss function and the health assessment loss function. The parameters of the physical information neural network model include network weight parameters, network bias parameters, physical constraint-related parameters, loss function weight parameters, regularization parameters, network layer number parameters, number of neurons in each layer parameters, activation function parameters, physical field modeling parameters, and health assessment modeling parameters.

[0136] S74. According to the updated physical information neural network model and the actual operating status of the system, the optimal control trajectory is adjusted in real time and sent to the valve actuator again to form a closed-loop control.

[0137] This invention achieves adaptive model updates by dynamically adjusting the parameters of the physical information neural network model by sending the optimal control trajectory to the valve actuator in real time and combining it with feedback data collected by sensors. Based on the feedback data, the system can correct the physical simulation and health assessment loss functions in real time to ensure that the model accurately reflects the actual operating conditions. Through closed-loop control, the system can continuously optimize the control strategy during actual operation, improve the accuracy and response speed of valve control, extend the service life of the equipment, enhance the safety and economic efficiency of operation, and realize intelligent and autonomous equipment management.

[0138] Example 1:

[0139] To verify the feasibility of the present invention, it was applied to the automatic control system for the main oil pipeline of a large petrochemical enterprise. During the operation of this system, the double-gate valve, as a key fluid control device, is frequently opened and closed, and the operating conditions are complex. Long-term operation is prone to excessive energy consumption, increased equipment wear, and difficulty in timely detection of abnormal operating conditions. Traditional valve opening and closing control methods mainly rely on preset opening and closing curves, and are unable to dynamically adjust the control strategy according to real-time operating conditions. This leads to high energy consumption during certain periods, shortened valve life, increased equipment maintenance costs, and the inability to provide timely warnings of abnormal health conditions.

[0140] In this embodiment, in response to the above-mentioned problems, the intelligent control method based on the improved physical information neural network proposed in the present invention is used to optimize the control of the double gate valve. The system first collects multiple operating condition data such as the number of valve opening and closing times, opening and closing time, pressure, flow, temperature, and vibration frequency in real time through the data acquisition unit, and uses data cleaning and normalization preprocessing. Subsequently, these operating condition characteristic data are input into the improved physical information neural network model. The model includes three sub-networks: control strategy generation, physical field simulation, and equipment health assessment, which are responsible for intelligently generating valve opening and closing trajectories, simulating physical field responses, and evaluating valve health status. The output results of each sub-network are jointly input into the multi-objective loss function for end-to-end optimization, thereby achieving continuous adaptive adjustment of the control strategy.

[0141] During actual operation, the system automatically collects and processes various valve operating data daily, using model reasoning to optimize valve opening and closing trajectories in real time. Experimental data shows that, after implementing this method, the maximum instantaneous current during valve opening and closing decreased by approximately 18%, the average opening and closing duration was shortened to 78% of the original value, and the annual overall energy consumption decreased by 14.6%. Furthermore, through real-time monitoring by the health assessment subnetwork, the accuracy of equipment abnormality warnings increased to over 97%, and the incidence of equipment failures was reduced by 35% compared to traditional control methods. For example, during peak pipeline load periods, the system automatically adjusts the valve opening and closing speed and amplitude, avoiding unexpected shutdowns caused by overload or sticking. Compared with historical manual control data, the average valve response time under automated intelligent control was shortened from 2.6 seconds to 1.9 seconds, and the temperature fluctuation during valve opening and closing was reduced from ±4.2°C to ±2.5°C, significantly improving system safety and stability. After one year of operation, the valve maintenance cycle was extended from 7 months to 14 months, reducing maintenance costs by 41%, significantly alleviating maintenance pressure.

[0142] The following is a summary table of some experimental data, titled "Comparison Table of Application Effects of Intelligent Double-Screen Valve Control System":

[0143] Table 1 Comparison of application effects of intelligent double gate valve control system

[0144]

[0145] As can be seen from Table 1 above, the present invention shows obvious advantages over traditional manual control methods in terms of automated control and health management of double-gate valves. First, in terms of energy efficiency, after adopting the intelligent control system of the present invention, the annual comprehensive energy consumption of the valve dropped from 19560kWh to 16710kWh, a decrease of 14.6%, and the maximum instantaneous current also decreased by 18%, which shows that the system has significantly improved the energy consumption management of the opening and closing actions and the electrical shock suppression ability of the actuator. Secondly, the average opening and closing time of the valve was shortened from 2.6 seconds to 1.9 seconds, a shortening of 26.9%, which not only improved the response speed of the production system, but also helped to improve the efficiency and safety of pipeline scheduling.

[0146] In terms of equipment health and safety, the health assessment and abnormal warning mechanism introduced in this invention has increased the accuracy of equipment abnormality warning from 78.2% to 97.3%, and the health assessment fault detection rate from 72.5% to 96.8%. This means that the system can identify potential risks of equipment earlier and more accurately, significantly reducing sudden downtime caused by faults. In actual statistics, the number of faults occurred within the year decreased from 23 to 15, a decrease of 34.8%, which fully demonstrates the proactive prevention effect of the intelligent system. In addition, through intelligent control of the temperature during the valve opening and closing process, the temperature fluctuation range is reduced from ±4.2°C to ±2.5°C, which greatly reduces the risk of damage to the equipment due to thermal shock and also plays a positive role in extending the life of the equipment.

[0147] In terms of maintenance and operation, the advantages brought by the present invention are also outstanding. The valve maintenance cycle has been extended from 7 months to 14 months, doubling the growth, and the annual maintenance cost has also dropped from 87,000 yuan to 51,000 yuan, a decrease of 41.4%. This not only reduces the workload of on-site maintenance personnel, but also effectively saves the company's operation and maintenance expenses. Overall, the present invention has achieved comprehensive optimization in key indicators such as equipment energy consumption, operational safety, and maintenance efficiency through intelligent data acquisition, physical modeling, and adaptive control methods, significantly improving the operational reliability and economic benefits of the double-gate valve, and has strong industrial promotion value.

[0148] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A double-disc valve opening and closing energy-saving control method based on intelligent algorithm, characterized in that: The steps include: S1. Collecting the working condition data of the double gate valve, normalizing and extracting the working condition data, and obtaining the working condition characteristic data vector; S2. Inputting the operating condition characteristic data vector into an improved physical information neural network model, wherein the improved physical information neural network model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork; S3. Generate a subnetwork through the control strategy and output the valve opening and closing control trajectory based on the working condition characteristic data vector; S4, inputting the control trajectory and working condition characteristic data vector into the physical field simulation sub-network, and obtaining the physical simulation results based on the coupling of the neural network and the physical equation; S5. Input the physical simulation results and the working condition characteristic data vector into the equipment health assessment subnetwork to obtain a health assessment result; S6. Based on the physical simulation results and health assessment results, a multi-objective loss function is constructed to obtain the optimal control trajectory by end-to-end joint optimization of the above sub-network parameters; S7. Send the optimal control trajectory to the valve actuator to perform valve opening and closing actions, collect feedback data in real time, and dynamically adjust and improve the parameters of the physical information neural network model.

2. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The working condition characteristic data vector X includes the inlet pressure P after normalization and feature extraction in , outlet pressure P out , flow rate Q, valve opening A, opening and closing speed V, ambient temperature T env , equipment wear status D and historical operation data H.

3. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The S2 specifically includes: S21. Inputting the operating condition characteristic data vector X into an improved physical information neural network model, wherein the model includes a control strategy generation subnetwork, a physical field simulation subnetwork, and an equipment health assessment subnetwork; S22. In the control strategy generation subnetwork, feature extraction and time series modeling are performed on the operating condition feature data vector X to generate and output the valve opening and closing control trajectory; S23. In the physical field simulation subnetwork, the operating condition characteristic data vector X and the valve opening and closing control trajectory are used as inputs, physical information coupling simulation is performed in combination with physical constraints, and the physical simulation results are output; S24. In the equipment health assessment subnetwork, the operating condition characteristic data vector X and the physical simulation results are used as input to perform state analysis and time series modeling, and output the health assessment results.

4. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The S3 specifically includes: S31, inputting the operating condition feature data vector into the control strategy generation subnetwork, wherein the input into the control strategy generation subnetwork includes a feature extraction layer, a time series modeling layer, and a decision generation layer; S32, performing feature encoding on the operating condition feature data vector at the feature extraction layer to obtain an operating condition feature representation; S33, the working condition feature representation is input to the time series modeling layer, and the gated recurrent unit is used to perform time series modeling on the working condition feature to obtain the intermediate time series feature; S34, input the intermediate time series feature S to the decision generation layer, and output the valve opening and closing control trajectory Γ=[A(t1),A(t2),…,A(t i ),…,A(t n )], where A(t i ) is the time t i valve opening.

5. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The S4 specifically includes: S41, inputting the valve opening and closing control trajectory Γ and the operating condition characteristic data vector X into the input layer of the physical field simulation subnetwork; S42, through the feature fusion layer, the valve opening and closing control trajectory Γ is spliced ​​with the working condition feature data vector X to obtain a fused feature vector Z; S43, inputting the fused feature vector Z into the deep neural network modeling layer, and performing nonlinear mapping on the fused feature vector Z using a multi-layer perceptron structure to obtain an intermediate feature representation; S44. In the physical equation coupling layer, the intermediate feature representation is coupled with the physical equation, and the physical consistency constraint is imposed on the physical quantity prediction results output at each moment by introducing a physical constraint loss function. S45, output the physical simulation result Y=[y1,y2,…,y i ,…,y n ], where y i is the physical simulation quantity obtained at the i-th moment, and the physical simulation quantity includes the valve flow, pressure distribution and energy consumption at the corresponding moment.

6. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 5 is characterized in that: The S44 specifically includes: S441. Input the intermediate feature representation output by the deep neural network modeling layer into the physical equation coupling layer to obtain the physical quantity prediction result at each moment; S442. Based on the valve opening and closing control trajectory and the operating condition characteristic data vector, a preset physical equation is used to calculate the theoretical reference value at each moment, wherein the theoretical reference value includes a theoretical valve flow rate, a theoretical pressure distribution, and a theoretical energy consumption; S443. The theoretical valve flow rate is calculated by multiplying the valve flow coefficient by the valve opening at the i-th moment, and then multiplying it by the square root of the ratio of the difference between the inlet pressure and the outlet pressure to the fluid density ρ; S444. The theoretical pressure distribution is calculated by subtracting the valve pressure drop determined by the valve resistance coefficient, the theoretical valve flow rate, and the valve opening, and the pipeline pressure drop determined by the pipeline friction factor f, length L, diameter D, fluid density ρ, and flow velocity V, from the inlet pressure. S445, Theoretical energy consumption is calculated by multiplying the theoretical valve flow by the difference between the inlet pressure and the outlet pressure, and then multiplying it by the time step t s Multiply and finally divide by the energy conversion efficiency η; S446. Introduce the physical simulation loss function into the physical equation coupling layer to impose physical consistency constraints on the physical quantity prediction results at each moment: Among them, L phy is the physical constraint loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, Q i is the valve flow output at the i-th moment, Q i,ref is the theoretical valve flow at the i-th moment, P i is the pressure distribution output at the i-th moment, P i,ref is the theoretical pressure distribution at the i-th moment, E i is the energy consumption output at the i-th moment, E i,ref is the theoretical energy consumption at the i-th moment.

7. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The S5 specifically includes: S51. Splicing the physical simulation results with the working condition characteristic data vector, inputting the results into the equipment health assessment sub-network, performing feature extraction through a multi-layer neural network structure, and obtaining feature extraction results; S52, input the feature extraction result into the equipment health status classification branch, use the Softmax activation function to classify and judge the equipment health status, and obtain the probability of each health status category [p i,1 ,p i,2 ,...,p i,C ], where C is the number of health status categories; S53. Based on the probabilities of the health status categories, the category label corresponding to the health status category with the largest probability value is used as the device health status indicator; S54: Input the feature extraction result into the equipment remaining life prediction branch to obtain the equipment remaining life prediction value and combine it with the equipment health status indicator to output as the health assessment result.

8. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1 is characterized in that: The S6 specifically includes: S61. Using the physical simulation results output by the physical simulation subnetwork and the health assessment results output by the device health assessment subnetwork as inputs to the multi-objective loss function; S62. Define the health assessment loss function: Among them, L health is the health assessment loss function, n is the number of moments in the valve opening and closing control trajectory, i is the moment index, β1 is the weight coefficient of the equipment health status indicator loss term, β2 is the weight coefficient of the equipment remaining life prediction value loss term, S i is the equipment health status indicator at the i-th moment, is the true label of the health status of the device at the i-th moment, is the cross entropy loss between the device health status indicator and the true label of the device health status, L i is the predicted value of the remaining life of the equipment at the i-th moment, is the true value of the remaining life of the equipment at the i-th moment; S63. Construct a multi-objective loss function: L total =λ1·L phy +λ2·L health ; Among them, L total is the multi-objective loss function, λ1 is the weight coefficient of the physical simulation loss function, L phy is the physical simulation loss function, λ2 is the weight coefficient of the health assessment loss function; S64. Based on the multi-objective loss function, an end-to-end joint optimization method is used to jointly optimize the parameters of the physical field simulation subnetwork and the equipment health assessment subnetwork to output the optimal control trajectory.

9. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 8 is characterized in that: The joint optimization includes inputting initial state parameters and initial trajectories of control variables, and iteratively optimizing the trajectories of control variables based on a multi-objective loss function; In each iteration, the physical simulation sub-network and the equipment health assessment sub-network are used to calculate the physical simulation loss function and the health assessment loss function, and the control variable trajectory is updated. When the convergence condition is met, the iteration is stopped and the current control variable trajectory is taken as the optimal control trajectory, where ∈ is the preset convergence threshold and k is the number of iterations.

10. The method for energy-saving control of opening and closing of a double-disc valve based on an intelligent algorithm according to claim 1, characterized in that: The S7 specifically includes: S71. Send the obtained optimal control trajectory to the valve actuator, driving the valve to open and close according to the optimal control trajectory; S72. Collect feedback data from the valve system in real time through sensors, wherein the feedback data includes valve status parameters, equipment health status indicators, and equipment remaining life prediction values; S73. Based on the collected feedback data, dynamically update the parameters of the physical information neural network model, and recalculate the physical simulation loss function and the health assessment loss function; S74. According to the updated physical information neural network model and the actual operating status of the system, the optimal control trajectory is adjusted in real time and sent to the valve actuator again to form a closed-loop control.

Images