Design method and device of fast response steam extraction check valve under high parameter conditions

Abstract

The present invention relates to the technical field of steam extraction check valves, and discloses a design method and device for a fast-response steam extraction check valve under high-parameter working conditions, the method comprising: constructing a valve working condition database and a valve body database; constructing a valve performance prediction model based on the valve working condition database and the valve body database, and obtaining a valve state quantity prediction value through real-time collected performance influencing factors; correcting the prediction value to obtain a valve state quantity correction value; establishing a valve control environment based on reinforcement learning, constructing a valve optimization control strategy model, and generating and executing a real-time optimal control action based on the valve state quantity correction value to achieve efficient optimization control of the valve; the present invention combines data-driven performance prediction with reinforcement learning optimization control to achieve fast response and efficient operation of the steam extraction check valve under high-parameter working conditions, thereby improving the reliability and energy-saving effect of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of steam extraction check valves, and more specifically, to a design method and device for a fast-response steam extraction check valve under high-parameter working conditions. Background Art

[0002] As a key control element in industrial pipeline systems, extraction check valves are widely used in power plants, chemical plants and other places where precise control of steam flow is required. Its main function is to prevent steam backflow in the pipeline and ensure safe and stable operation of the system. With the development of industrial automation and intelligent control technology, traditional extraction check valve control systems have gradually evolved towards online testing and intelligent control to improve the valve's response speed, sealing reliability and service life. However, existing technologies still have many deficiencies in meeting the demand for rapid response under high-parameter conditions.

[0003] The Chinese patent with publication number CN117606786A discloses an online test control system and test method for a steam extraction check valve. The system includes an air source assembly, a control valve assembly, an air-controlled reversing valve, a flow tank and a cylinder, wherein the control valve assembly is composed of a two-position three-way solenoid valve and a two-position three-way manual valve. The advantage of this design is that the control valve assembly, the air-controlled reversing valve and the flow tank are installed away from the cylinder, thereby avoiding the influence of high temperature on these components and extending their service life. In addition, the reversing of the air-controlled reversing valve is controlled by a manual valve, which effectively avoids the problem of the solenoid valve test button being stuck, and the safe opening of the cylinder is controlled by adjusting the capacity of the flow tank, thereby improving the safety of the test. However, the system relies on manually operated reversing valves, lacks automated control and intelligent optimization capabilities, and is difficult to achieve real-time dynamic adjustment and optimization control of the valve under complex working conditions, limiting its application effect under high parameter conditions.

[0004] The Chinese patent with publication number CN112050659A discloses a power plant condenser vacuum system and its control method. The system adds a condenser exhaust pipeline in parallel to the original water ring vacuum pump group, connects a pneumatic butterfly valve, a check valve, a Roots pump and a water ring pump in series, and is equipped with a pressure transmitter and a pressure switch to achieve effective vacuum control of the condenser. The system realizes remote monitoring and control of the vacuum process by connecting to the DCS remote distributed control system. However, this control method mainly relies on fixed pressure monitoring and mechanical control elements, lacks performance prediction and intelligent optimization control mechanism based on real-time data, and cannot adaptively adjust the control strategy according to changes in operating conditions, resulting in insufficient response speed and control accuracy when facing high-parameter conditions, and it is difficult to meet the requirements of efficient operation in complex environments.

[0005] In summary, the existing technology in the online testing and control system of the steam extraction check valve mainly relies on traditional mechanical control and fixed monitoring methods, lacks advanced data-driven performance prediction models and intelligent optimization control strategies, resulting in that under high parameter conditions, the valve response speed, sealing reliability and service life are difficult to be effectively improved, and the overall operating efficiency and safety of the system are limited. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a design method and device for a fast-response steam extraction check valve under high-parameter working conditions. By constructing a detailed valve working condition database and an ontology database, using a random forest algorithm to predict performance, and combining a reinforcement learning algorithm to optimize the control strategy, precise control and efficient response of the valve under complex working conditions are achieved, which significantly improves the reliability and energy-saving effect of the system.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The design method of the fast response steam extraction check valve under high parameter conditions includes:

[0009] Collect the first operating condition parameters of the target steam extraction check valve in the historical operation cycle to build a valve operating condition database; obtain valve structure parameters and material property parameters to build a valve body database;

[0010] Based on the valve working condition database and the valve body database, a valve performance prediction model is constructed; the real-time valve performance influencing factors are obtained, and the valve state quantity prediction value is obtained according to the real-time valve performance influencing factors and the valve performance prediction model; the theoretical valve state quantity under different working condition levels is calculated, and the theoretical valve state quantity is compared with the valve state quantity prediction value, and the valve state quantity deviation is calculated; according to the valve state quantity deviation, the valve state quantity prediction value is corrected to obtain the valve state quantity correction value;

[0011] A reinforcement learning environment for valve control is established. Based on the reinforcement learning environment for valve control, a valve optimization control strategy model is constructed. According to the valve state quantity correction value and the valve optimization control strategy model, the real-time optimal control action is obtained and valve optimization control is performed.

[0012] Furthermore, the valve operating status database includes n 1 Working condition records, each working condition record includes n 2 The attribute data of each sampling time point includes the sampling time stamp, pressure value, temperature value, flow value and valve state quantity; the valve state quantity includes valve opening and closing time, sealing pressure and valve opening degree;

[0013] According to the valve working condition database n 1The pressure, temperature and flow values ​​recorded in the working condition are 1 Cluster the n working condition records and cluster the n working condition records in the valve working condition database 1 The working condition records are divided into N 1 The operating condition levels are numbered in order from low to high according to the clustering results.

[0014] Furthermore, the pair n 1 The operating condition records are clustered as follows:

[0015] Step S1210, constructing an operating condition feature vector according to the pressure value, temperature value and flow value in the operating condition record; and 1 Cluster the working condition records and get N 1 Cluster centers; each cluster center corresponds to a working condition level, and each working condition record is divided into the working condition level to which the nearest cluster center belongs;

[0016] Step S1220, calculate the silhouette coefficient S, if S>S0, then output the clustering results and the working condition classification results; if S≤S0, then adjust the number of cluster centers N 1 , return to step S1210 to cluster again until S>S0; where S0 is a preset silhouette coefficient threshold.

[0017] Furthermore, the building of the valve performance prediction model includes:

[0018] The valve operating condition database and the valve body database are integrated to obtain a data set of valve performance influencing factors; the valve performance influencing factors include operating condition level, valve structure parameters and material property parameters;

[0019] A valve performance prediction model was constructed using random forest algorithm;

[0020] Select the valve performance influencing factors at m consecutive sampling time points within a working cycle as input, and the corresponding valve state quantity as output to construct a training sample set;

[0021] The valve performance prediction model is trained using the training sample set.

[0022] Furthermore, the valve state quantity prediction value includes the valve opening and closing prediction time, the sealing pressure prediction value and the valve opening prediction value; the theoretical valve state quantity includes the valve opening and closing theoretical time, the sealing pressure theoretical value and the valve opening theoretical value; the valve state quantity deviation includes the valve opening and closing time deviation, the sealing pressure deviation and the valve opening deviation;

[0023] The calculation of the valve state deviation comprises:

[0024] Calculate the valve opening and closing time deviation based on the valve opening and closing theoretical time and the valve opening and closing predicted time;

[0025] Calculate the sealing pressure deviation according to the sealing pressure theoretical value and the sealing pressure predicted value;

[0026] The valve opening deviation is calculated based on the valve opening theoretical value and the valve opening predicted value.

[0027] Furthermore, the valve state correction value includes valve opening and closing correction time, sealing pressure correction value and valve opening correction value;

[0028] The method of correcting the predicted value of the valve state quantity to obtain the corrected value of the valve state quantity includes:

[0029] According to the valve opening and closing time deviation, the valve opening and closing prediction time is corrected to obtain the valve opening and closing correction time;

[0030] According to the sealing pressure deviation, the sealing pressure prediction value is corrected to obtain the sealing pressure correction value;

[0031] According to the valve opening deviation, the valve opening prediction value is corrected to obtain the valve opening correction value.

[0032] Further, the reinforcement learning environment of the valve control includes a state space, an action space and a reward function;

[0033] The reinforcement learning environment for establishing valve control includes:

[0034] The valve state quantity and the working condition level of the valve constitute the state space of reinforcement learning; the dimension of the state space is the product of the number of valve state quantities and the working condition level;

[0035] Obtain the regulating variables of the valve control, and use the regulating variables of the valve control to form an action space of reinforcement learning; the dimension of the action space is the number of regulating variables;

[0036] Calculate valve performance indicators, and construct a reinforcement learning reward function R based on the valve performance indicators; the valve performance indicators include a rapid response performance indicator R1, a sealing reliability indicator R2, a service life indicator R3, and an energy consumption level indicator R4.

[0037] Furthermore, the reward function R for constructing reinforcement learning includes:

[0038] Calculate the weighted average of valve opening and closing time deviation, sealing pressure deviation and valve opening deviation as the rapid response performance index R1;

[0039] Obtain the yield strength σs of the sealing surface material from the valve body database, and calculate the contact stress σ of the valve disc under the sealing pressure; evaluate the sealing reliability index R2 based on the ratio of the contact stress σ to the yield strength σs;

[0040] Obtain the cumulative number of valve actions N, and evaluate the valve service life index R3 based on the cumulative number of valve actions N and the rated number of valve actions N0;

[0041] Obtain the energy consumption E of a single action of the valve, and evaluate the energy consumption level index R4 of the valve according to the energy consumption E of a single action of the valve;

[0042] The quick response performance index R1, sealing reliability index R2, service life index R3 and energy consumption level index R4 are weighted averaged to obtain the reward function R of reinforcement learning.

[0043] A design device for a fast-response steam extraction check valve under high-parameter working conditions, which is used to implement the design method for a fast-response steam extraction check valve under high-parameter working conditions, and the device comprises:

[0044] Database construction module: collect the first operating condition parameters of the target steam extraction check valve in the historical operation cycle to build a valve operating condition database; obtain valve structure parameters and material property parameters to build a valve body database;

[0045] Valve state prediction module: Based on the valve operating condition database and the valve body database, a valve performance prediction model is constructed; real-time valve performance influencing factors are obtained, and the valve state quantity prediction value is obtained according to the real-time valve performance influencing factors and the valve performance prediction model;

[0046] Valve state correction module: calculate the theoretical valve state quantity under different working conditions, compare the theoretical valve state quantity with the valve state quantity prediction value, and calculate the valve state quantity deviation; according to the valve state quantity deviation, correct the valve state quantity prediction value to obtain the valve state quantity correction value;

[0047] Valve optimization control module: Establish a reinforcement learning environment for valve control, and build a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtain the real-time optimal control action based on the valve state correction value and the valve optimization control strategy model, and execute valve optimization control.

[0048] An electronic device comprises a memory, a central processing unit and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the design method of a fast-response steam extraction check valve under high-parameter working conditions is implemented.

[0049] A computer-readable storage medium stores a computer program, which, when executed, implements the design method of a fast-response steam extraction check valve under high-parameter working conditions.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention realizes efficient operation and reliable control of the steam extraction check valve under high parameter conditions through systematic data collection and analysis, advanced machine learning model construction and intelligent control strategy optimization. First, a detailed valve operating condition database and ontology database are constructed, so that the operating status and structural characteristics of the valve under various operating conditions can be comprehensively recorded and analyzed, providing a solid data foundation for subsequent performance prediction and control optimization. Secondly, the performance prediction model constructed by the random forest algorithm can accurately capture the complex nonlinear relationship between the factors affecting valve performance and the state quantity, and improve the accuracy and robustness of the prediction. At the same time, the operating conditions are divided into different levels through cluster analysis, and the clustering results are optimized using the contour coefficient to ensure the scientificity and rationality of the operating condition classification, providing a clear basis for the formulation of customized control strategies. In addition, the optimization control strategy model based on reinforcement learning can respond to changes in operating conditions in real time, dynamically adjust valve control parameters, achieve rapid response and efficient operation, and significantly improve the overall performance and stability of the system. Finally, through the real-time correction of the valve state quantity and the continuous optimization of the control strategy, not only the response speed and sealing reliability of the valve are improved, but also the service life of the valve is effectively extended, energy consumption is reduced, and the energy saving effect and economy of the system are enhanced. Therefore, the present invention has significant advantages in improving the operating efficiency, reliability and energy saving of the steam extraction check valve, and meets the stringent requirements for valve performance under high parameter conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 It is a principle flow chart of the design method of the fast response steam extraction check valve under high parameter conditions in the present invention;

[0054] Figure 2 The method for designing a fast-response steam extraction check valve under high-parameter working conditions of the present invention is for n 1 Cluster the n working condition records and cluster the n working condition records in the valve working condition database 1 The working condition records are divided into N 1Flow chart of the method for each operating level;

[0055] Figure 3 A flow chart of a method for constructing and training a valve performance prediction model in a design method for a fast-response steam extraction check valve under high-parameter conditions of the present invention;

[0056] Figure 4 It is a flow chart of a method for calculating valve state quantity deviation in a design method of a fast-response steam extraction check valve under high-parameter working conditions of the present invention;

[0057] Figure 5 It is a flow chart of a method for correcting a predicted value of a valve state quantity in a design method of a fast-response steam extraction check valve under high-parameter working conditions of the present invention;

[0058] Figure 6 A flow chart of a method for establishing a reinforcement learning environment for valve control in a design method for a fast-response steam extraction check valve under high-parameter conditions of the present invention;

[0059] Figure 7 A flow chart of a method for constructing a reinforcement learning reward function R in a design method for a fast-response steam extraction check valve under high-parameter conditions of the present invention;

[0060] Figure 8 It is a functional module diagram of the design device of the fast response steam extraction check valve under high parameter conditions in the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Example 1

[0063] See also Figure 1 As shown, this embodiment provides a design method for a fast-response steam extraction check valve under high-parameter conditions, including:

[0064] Step S1000, collecting the first operating condition parameters of the target steam extraction check valve in the historical operation cycle, and constructing a valve operating condition database; the valve operating condition database includes n 1 Working condition records, each working condition record includes n 2 The attribute data of each sampling time point includes the sampling time stamp, pressure value, temperature value, flow value and valve status; according to the valve working condition database n 1 The pressure, temperature and flow values ​​recorded in the working condition are1 Cluster the n working condition records and cluster the n working condition records in the valve working condition database 1 The working condition records are divided into N 1 The working condition levels are numbered in order from low to high according to the clustering results; valve structure parameters and material property parameters are obtained to build a valve body database;

[0065] Furthermore, step S1000 includes:

[0066] Step S1100, collecting the first operating condition parameters of the target steam extraction check valve in the historical operation cycle, and constructing a valve operating condition database; the valve operating condition database includes n 1 Working condition records, each working condition record includes n 2 The attribute data of each sampling time point includes the sampling time stamp, pressure value, temperature value, flow value and valve state quantity; the valve state quantity includes valve opening and closing time, sealing pressure and valve opening degree;

[0067] Specifically, the first operating parameters of the target steam extraction check valve in a complete historical operation cycle are collected to build a valve operating database. The first operating parameters include the valve's pressure, temperature, flow, and other parameters that reflect the valve's operating status. By installing pressure sensors, temperature sensors, flow meters, and other instruments on the valve, the changes of these parameters over time can be continuously collected. The collected data is stored in time series to form a valve operating database.

[0068] Valve condition database includes n 1 Working condition records, each working condition record includes n 2 Attribute data of sampling time points. 1 Indicates the number of operating conditions recorded in one operating cycle, n 2 Indicates the number of data points contained in each operating record. For example, if data is collected every 1 minute and the operating cycle is 1 day, then n 2 =1440, that is, each working condition record contains 1440 data points. 1 The size of the value depends on the actual operation time of the valve and the sampling frequency.

[0069] The attribute data of each sampling time point includes sampling timestamp, pressure value, temperature value, flow value and valve status. The sampling timestamp records the time of data collection, the pressure value, temperature value and flow value reflect the current working parameters of the valve, and the valve status indicates the action state of the valve, such as valve opening and closing time, sealing pressure and valve opening. The valve opening and closing time records the time required for the valve to go from fully open to fully closed (or from fully closed to fully open); the sealing pressure indicates the contact pressure between the valve disc and the valve seat, reflecting the sealing performance of the valve; the valve opening describes the degree of opening of the valve disc, which can be a value of 0-100%, or a discrete state such as fully open, half open, slightly open, etc.

[0070] By collecting and storing historical operating data, a complete valve operating database is established. This database comprehensively records the operating characteristics and dynamic response of the valve under real working conditions, providing a basis for subsequent data analysis, performance prediction, and optimization control. Through big data analysis technology, valuable information can be extracted from massive operating data to reveal the factors and laws affecting valve performance, laying a data foundation for applications such as valve design optimization, fault diagnosis, and predictive maintenance.

[0071] Step S1200, according to the valve working condition database n 1 The pressure, temperature and flow values ​​recorded in the working condition are 1 Cluster the n working condition records and cluster the n working condition records in the valve working condition database 1 The working condition records are divided into N 1 A working condition level, recording the working condition level label of each working condition record; the working condition levels are numbered in order from low to high according to the clustering results;

[0072] Furthermore, if Figure 2 As shown, step S1200 includes:

[0073] Step S1210, constructing an operating condition feature vector according to the pressure value, temperature value and flow value in the operating condition record; and 1 Cluster the working condition records and get N 1 Cluster centers; each cluster center corresponds to a working condition level, and each working condition record is divided into the working condition level to which the nearest cluster center belongs;

[0074] Step S1220, calculate the silhouette coefficient S, if S>S0, then output the clustering results and the working condition classification results; if S≤S0, then adjust the number of cluster centers N 1 , return to step S1210 to cluster again until S>S0; where S0 is a preset silhouette coefficient threshold.

[0075] Specifically, each operating record is regarded as a sample, and the three parameters of pressure, temperature, and flow constitute the characteristic vector of the sample. Taking pressure as an example, the maximum value, minimum value, average value, standard deviation and other statistics in the record can be selected as the characteristic value of pressure. The characteristic value extraction method of temperature and flow is similar. In this way, each operating record is mapped to a high-dimensional feature space and becomes a data point. According to the operating characteristic vector, a clustering algorithm such as K-means is used to cluster n 1 Cluster the working condition records and get N 1 The K-means algorithm divides the data points into the categories to which the nearest cluster center belongs through iterative optimization, and continuously updates the positions of the cluster centers until convergence. 1 Indicates the number of clustering categories, which needs to be preset according to the distribution characteristics of the data and prior knowledge. Each cluster center corresponds to a working condition level, and the clustering result will be n 1 The working condition records are divided into N 1 The working condition levels are numbered from low to high according to the mean values ​​of pressure, temperature, and flow rate at the cluster center. The higher the pressure, temperature, and flow rate, the higher the working condition level, and the worse the corresponding working condition.

[0076] In order to evaluate the quality of clustering, the silhouette coefficient S is introduced as a measure of clustering quality. Given a threshold S0, if S is greater than S0, the current clustering result is considered acceptable, and the clustering result and working condition classification are output; if S is less than or equal to S0, the clustering effect is considered unsatisfactory, and the number of cluster centers N needs to be adjusted. 1 , return to step S1210 and perform clustering again until S>S0 is satisfied. The value of S0 needs to be set based on experience and actual needs, and is generally set to about 0.5.

[0077] Through the above steps, the operating records are automatically divided into different operating levels. This division makes full use of the statistical characteristics of the data itself, digs out the correlation pattern between the operating parameters, and is objective and adaptive. Through cluster analysis, we can find the typical operating conditions experienced by the valve in actual operation and understand the frequency and duration of different operating conditions. By specifically studying the performance characteristics of the valve under each operating level, more targeted control optimization measures can be taken. At the same time, reasonable operating condition division helps to simplify the operating space, reduce data redundancy, and reduce the complexity of subsequent analysis and modeling.

[0078] For example, suppose a cluster analysis is performed on 1,000 operating condition records of a steam trap, and three cluster centers are obtained, corresponding to the three operating conditions of low pressure, medium pressure, and high pressure. Through analysis, it is found that 80% of the operating condition records are low-pressure conditions, 15% are medium-pressure conditions, and 5% are high-pressure conditions, and high-pressure conditions usually occur at the moment of steam trap discharge. This result shows that the steam trap is in a low-pressure state most of the time and only experiences a higher pressure difference when discharging sewage. Therefore, when designing valves, the sealing performance and service life under low-pressure conditions should be focused on; when formulating maintenance strategies, the maintenance cycle under low-pressure conditions can be extended accordingly, while the inspection frequency of high-pressure conditions can be shortened, which can not only meet safety requirements but also reduce maintenance costs.

[0079] In short, the clustering and classification of working conditions can describe the actual working state of the valve from a data-driven perspective, extract typical working condition patterns, and optimize the spatial expression of working conditions. On this basis, performance evaluation, fault diagnosis, life prediction, etc. can be carried out, and the dynamic changes and mutual coupling of working conditions can be fully considered to obtain more practical and targeted results. Compared with the traditional single working condition evaluation method, working condition clustering has the advantages of comprehensiveness, objectivity, and adaptability.

[0080] The calculation of the silhouette coefficient S comprises:

[0081] First, define some necessary symbols:

[0082] :Indicates the Working condition records.

[0083] :Indicates the The cluster category to which the working condition record belongs.

[0084] : The total number of clusters.

[0085] : No. The number of cluster internal points of the working condition records.

[0086] :Indicates working condition record and The distance metric between them can be Euclidean distance, Manhattan distance, etc., depending on the specific situation. Indicates Working condition records.

[0087] : Indicates clustering categories The average distance of all points in the .

[0088] : Indicates clustering categories and The average distance between Indicates The cluster category to which the working condition record belongs.

[0089] : Indicates clustering categories and The density difference measure between is defined as the density difference between two classes.

[0090] Intra-class distance :The intra-class distance represents the The average distance between a working condition record and similar working condition records. The effect of record density on intra-class distance is added here to reflect the different contributions of high-density and low-density areas. The smaller the intra-class distance, the more compact the clustering and the better the clustering effect.

[0091] How to obtain: Calculate The distance between the working condition record and the similar working condition records, and weighted average:

[0092]

[0093] Distance between classes :The distance between classes refers to The average distance between a working condition record and the nearest record in other clusters. In order to consider the influence of density, the difference measure of cluster density is introduced to enhance the weight of inter-class distance. The larger the inter-class distance, the more obvious the difference between different clusters, and the better the clustering effect.

[0094] How to obtain: Calculate The average distance between the working condition record and all other clusters, and take the minimum value:

[0095]

[0096] in, Indicates Working condition records; Indicates The number of cluster internal points of the working condition records; Indicates working condition record and The distance measure between .

[0097] Density constraint: In order to introduce the density information of the data, a density difference measure is defined , which can be calculated by the difference in point density between clusters. Clusters with higher density have a greater impact on the calculation results, and vice versa.

[0098] Acquisition method: Calculate the average distance of points in different clusters:

[0099]

[0100] in, Represents cluster categories The average distance of all points in the .

[0101] The comprehensive silhouette coefficient :Comprehensively consider the distance within the class , the distance between classes and density constraints, we get the silhouette coefficient :

[0102]

[0103] The closer it is to 1, the better the clustering effect is. If it is close to 0 or negative, it means the clustering effect is poor.

[0104] Intra-class distance The smaller it is, the higher the similarity within the class is, and the better the clustering effect is. The molecular part will become relatively large, resulting in The value of increases. Reducing the intra-class distance will make the points in the class closer, thereby improving the quality of clustering and promoting the silhouette coefficient. Therefore, the intra-class distance The reduction of helps to improve the silhouette coefficient , so that its value tends to 1, indicating that the clustering effect is good. The smaller the intra-class distance, the greater the silhouette coefficient. The larger the value, the tighter the clustering and the higher the consistency within the class.

[0105] Distance between classes The bigger, The molecular part will increase, thereby increasing the silhouette coefficient The larger the distance between classes, the more distinct the classes in the clustering results will be, which will promote the silhouette coefficient. Develops towards 1. Therefore, the distance between classes The increase of helps to improve the silhouette coefficient , making its value close to 1, indicating a good clustering effect. The larger the distance between classes, the greater the silhouette coefficient. The higher it is, the greater the difference between clusters and the better the clustering effect.

[0106] Density Difference The larger the When the density difference is large, the clusters in the low-density area will be easier to identify, resulting in an improvement in the clustering effect. If the density difference between clusters is small, the evaluation of clustering effect will be affected, and the density term in the formula will not make much improvement on the silhouette coefficient, resulting in a relatively flat evaluation of clustering effect. Therefore, the introduction of density constraints allows the clustering effect in different density areas to be better reflected. When the density difference is large, the clustering effect will be better, and the silhouette coefficient Smaller density differences will affect the evaluation of the silhouette coefficient, making it more difficult to distinguish clusters of different densities.

[0107] By introducing density constraints, the silhouette coefficient The clustering results are evaluated more accurately. The clustering effect is not only determined by the intra-class distance and the inter-class distance, but also takes into account the density difference, making the clustering results more reasonable when the data is unevenly distributed. This formula helps to accurately evaluate the clustering quality in high-dimensional data by weighting the intra-class, inter-class distance and density difference. The introduction of density difference (density constraint) can reduce the negative impact of low-density areas on clustering and improve the stability of clustering.

[0108] Step S1300, obtaining valve structure parameters and material property parameters, and constructing a valve body database.

[0109] Specifically, valve structural parameters describe the geometric structure and size information of the valve, which is a key factor affecting valve performance. Common structural parameters include:

[0110] Valve seat diameter: The valve seat is part of the valve sealing surface, and its diameter directly determines the flow area and flow resistance characteristics.

[0111] Sealing surface cone angle: The sealing surface cone angle is the angle between the valve disc and the valve seat sealing surface, which affects the sealing effect and the opening and closing torque. The smaller the cone angle, the better the sealing performance, but the opening and closing torque is also greater.

[0112] Disc travel: Disc travel refers to the linear displacement distance of the disc from fully open to fully closed, which determines the adjustment range and control accuracy of the valve. Too small a stroke will limit the adjustment ability, while too large a stroke will affect the dynamic response.

[0113] Spring stiffness: Spring stiffness indicates the elastic force provided by the spring, which affects the valve disc's reset ability and sealing pressure. If the stiffness is too low, the valve disc is difficult to reset reliably; if the stiffness is too high, the opening and closing resistance increases.

[0114] Material property parameters reflect the material selection and mechanical properties of each valve component, which is crucial to the reliability and service life of the valve. They mainly include:

[0115] Material grade: Material grade indicates the type and grade of material used in various valve components, such as stainless steel 304, 316, chrome-molybdenum steel F11, F22, etc. The material grade determines the basic properties of the component, such as strength, hardness, and corrosion resistance.

[0116] Elastic modulus: The elastic modulus indicates the material's ability to resist elastic deformation and is an important parameter for evaluating component stiffness and stress levels. The larger the modulus, the smaller the component deformation and the greater the force transmitted.

[0117] Hardness: Hardness indicates the ability of a material to resist local plastic deformation, and has an important influence on the wear and scratch resistance of the valve. If the hardness is too low, wear and deformation will occur easily; if the hardness is too high, the processing performance will deteriorate.

[0118] Thermal conductivity: Thermal conductivity indicates the ability of a material to transfer heat, affecting the heat transfer characteristics and temperature distribution of the valve. The higher the thermal conductivity, the more uniform the temperature rise of the valve and the lower the thermal stress level.

[0119] Linear expansion coefficient: Linear expansion coefficient indicates the dimensional stability of the material under temperature changes, which has an important influence on the thermal deformation and sealing performance of the valve. If the expansion coefficient is too different, it is easy to cause relative displacement of components and stress concentration.

[0120] By collecting and organizing these structural parameters and material property data, a complete valve database is established. This database comprehensively describes the physical characteristics and mechanical behavior of the valve, and is an important basis for valve performance analysis and optimization design. When the valve structure or material changes, the corresponding parameters can be modified to quickly evaluate its impact on valve performance, providing data support for product selection and improvement.

[0121] In addition, the valve body database can also be combined with the valve operating database to achieve comprehensive prediction and diagnosis of valve performance. Taking the valve seat diameter and the sealing surface cone angle as an example, these two structural parameters directly determine the flow characteristics and sealing performance of the valve. By combining them with operating parameters (such as pressure difference and temperature), the leakage and adjustment accuracy under different operating conditions can be predicted. Similarly, by comprehensively considering parameters such as spring stiffness and valve disc quality, a valve dynamic characteristic model can be established to predict the response time and stability of the valve under transient conditions.

[0122] Valve structural parameters and material property parameters are the core factors that affect valve performance. Building a valve database can systematize and digitize the scattered physical parameters, provide data basis and decision support for valve analysis, diagnosis, design, selection and other work, and greatly improve the level of valve life cycle management. It is recommended that when designing and selecting valves, comprehensive consideration should be given to factors such as structure, materials, and working conditions, and various parameters should be optimized to balance the static performance and dynamic characteristics of the valve and achieve safe and economical operation.

[0123] Step S2000, based on the valve working condition database and the valve body database, construct a valve performance prediction model; obtain the real-time valve performance influencing factors, and obtain the valve state quantity prediction value according to the real-time valve performance influencing factors and the valve performance prediction model; calculate the theoretical valve state quantity under different working condition levels; compare the theoretical valve state quantity with the valve state quantity prediction value, and calculate the valve state quantity deviation; according to the valve state quantity deviation, correct the valve state quantity prediction value to obtain the valve state quantity correction value;

[0124] Furthermore, step S2000 includes:

[0125] Step S2100, integrating the valve working condition database and the valve body database to obtain a data set of valve performance influencing factors; the valve performance influencing factors include working condition level, valve structure parameters and material property parameters;

[0126] Specifically, factors affecting valve performance refer to various factors that have a significant impact on the static performance and dynamic characteristics of the valve, mainly including operating conditions, valve structure parameters and material property parameters.

[0127] The operating condition level reflects the external environmental conditions faced by the valve, such as the size and range of parameters such as pressure, temperature, and flow. Different operating condition levels correspond to different load levels and failure mechanisms. For example, high pressure and high temperature conditions will increase the risk of leakage and fatigue damage of the valve, while frequent opening and closing conditions will test the dynamic response and service life of the valve. Therefore, it is necessary to accurately divide the operating condition level based on the actual operating conditions of the valve and formulate targeted control strategies and maintenance plans.

[0128] The valve structural parameters and material property parameters reflect the design characteristics and physical properties of the valve body, and are key factors in determining the valve's pressure bearing capacity, sealing performance, flow resistance characteristics, etc. Structural parameters such as valve seat diameter and sealing surface cone angle directly affect the basic function and adjustment accuracy of the valve. Material properties such as elastic modulus and yield strength determine the valve's strength margin, deformation behavior, and failure mode. Optimizing the selection of valve structure and material grade can fundamentally improve the reliability and safety of the valve.

[0129] By fusing multi-source heterogeneous data such as operating condition level, structural parameters, and material properties, a unified data set of factors affecting valve performance is established. This data set comprehensively describes the influencing mechanism and law of valve performance and is an important basis for building a high-precision prediction model. Through data fusion, the coupling and interaction between different factors can be revealed, and the applicability and robustness of the prediction model can be improved.

[0130] Taking the selection of valve seat materials as an example, the elastic modulus, hardness, wear resistance and other properties of the material are crucial to the sealing reliability of the valve. However, material properties often vary with operating conditions such as temperature and stress state. If only the material properties at room temperature and pressure are considered, while ignoring the degradation of material properties under high temperature conditions, the sealing ability of the valve may be overestimated, resulting in leakage risks. Therefore, when constructing a performance prediction model, it is necessary to consider both the constitutive properties of the material and the influence of the operating conditions, and comprehensively evaluate the service performance of the valve.

[0131] There are complex interactions and constraints between factors such as operating conditions, body structure, and material properties. The weight of each factor on valve performance also varies depending on the operating conditions. When fusing data, it is necessary to reasonably set the importance level and influence coefficient of each factor according to the specific operating conditions. Through methods such as weighted fusion or hierarchical fusion, the influence of key factors can be highlighted, the interference of minor factors can be suppressed, and the accuracy and efficiency of the prediction model can be improved. In addition, when fusing heterogeneous data, attention should also be paid to the consistency and comparability of the data. Data from different sources and in different formats may have deviations or conflicts. It is necessary to eliminate data noise and outliers through pre-processing operations such as data cleaning, conversion, and alignment, and establish a standardized data set of performance influencing factors.

[0132] In summary, integrating valve operating data and body data to construct a data set of factors affecting valve performance is a key step in achieving valve performance prediction and optimal control. By systematically analyzing the influencing mechanism of valve performance, taking into account factors such as operating conditions, structural design, and material selection, and combining technologies such as big data analysis and machine learning, the accuracy and applicability of valve performance prediction can be comprehensively improved, laying the foundation for applications such as valve design optimization, condition monitoring, and fault diagnosis. It is recommended that when conducting valve performance prediction, full use should be made of multi-source heterogeneous data, and attention should be paid to the interactive coupling of various factors, in order to obtain reliable and comprehensive prediction results.

[0133] Step S2200, using valve performance influencing factors as input and valve state quantities as output, to construct and train a valve performance prediction model;

[0134] The valve performance prediction model is a data-driven model based on machine learning, which aims to establish a nonlinear mapping relationship between valve performance influencing factors and valve state quantities. By learning a large amount of historical operating data, the model can automatically extract the inherent laws and key features of valve performance and achieve accurate prediction of valve status.

[0135] Factors affecting valve performance refer to various parameters that affect the dynamic performance of valves, including operating conditions, valve structural parameters, and material property parameters. Among them, the operating conditions reflect the operating conditions of the valve, such as pressure, temperature, and flow; valve structural parameters such as valve seat diameter, sealing surface cone angle, and valve disc travel determine the inherent characteristics of the valve; material property parameters such as elastic modulus, hardness, and thermal conductivity affect the mechanical properties and thermal characteristics of various valve components. By taking these influencing factors as inputs to the model, the comprehensive impact of various factors on valve performance can be fully considered.

[0136] The valve state quantity refers to the key indicators that reflect the real-time dynamic performance of the valve, such as valve opening and closing time, sealing pressure, valve opening, etc. The valve opening and closing time reflects the rapid response capability of the valve action, the sealing pressure reflects the reliability of the valve seal, and the valve opening characterizes the accuracy of the valve adjustment. By predicting these state quantities, the dynamic characteristics of the valve under different working conditions can be mastered, providing a basis for optimized control.

[0137] Furthermore, if Figure 3 As shown, step S2200 includes:

[0138] Step S2210, using a random forest algorithm to construct a valve performance prediction model;

[0139] The random forest algorithm is a machine learning algorithm based on decision tree ensembles, which has strong nonlinear fitting ability and generalization performance. By constructing multiple decision trees and integrating their prediction results, the algorithm can effectively process high-dimensional, nonlinear and complex data and reduce the risk of overfitting.

[0140] When constructing a valve performance prediction model, it is first necessary to determine the input features and output targets of the model. The input features are the factors affecting valve performance, which can be extracted from the original data through feature engineering methods, such as calculating statistics such as mean, variance, peak, or performing signal processing operations such as frequency domain transformation and wavelet decomposition. The output target is the valve state quantity, which can be used directly as the original value or after preprocessing such as normalization and standardization. Then, multiple self-service sample sets and feature subsets are generated from the original data set through random sampling and feature randomization methods, and a decision tree is trained on each self-service sample set. The decision tree generation process adopts a recursive partitioning strategy, that is, the samples are divided into several subsets according to the value of a certain feature, so that the samples in each subset belong to the same category as much as possible, and the differences between subsets are as large as possible. By minimizing indicators such as the Gini index or information gain, the optimal partitioning features and thresholds can be selected.

[0141] When generating each decision tree, randomness can be introduced to further improve the diversity and robustness of the model. On the one hand, for each node, a certain number of features are randomly selected from all features as candidates, and the best features are selected from the candidate features for division; on the other hand, at the leaf node, a self-service sampling method can be used to generate multiple different training subsets, train multiple sub-models respectively, and average their prediction results as the output of the leaf node. This randomization strategy can effectively reduce the correlation between decision trees and improve the generalization ability of the integrated model.

[0142] Optimize model hyperparameters through grid search, including the number of decision trees, the maximum depth of the tree, the minimum number of samples for leaf nodes, etc. The number of decision trees refers to the number of basic decision trees in the random forest, and usually takes a larger value such as 100, 500, etc. The more trees there are, the better the integration effect, but the computational overhead will also increase. A balance needs to be struck between performance and efficiency. The maximum depth of the tree controls the complexity of a single decision tree. The greater the depth, the stronger the fitting ability, but the higher the risk of overfitting. It can usually be set to no more than 20 layers. The minimum number of samples for leaf nodes limits the minimum number of samples required for each leaf node, which plays a role in pruning and avoids overfitting caused by excessive subdivision. Common values ​​include 1, 5, 10, etc. Through the grid search method, different combinations can be traversed within the parameter value range, and the model performance can be evaluated by cross-validation to select the parameter combination with the best generalization performance. This hyperparameter optimization method is simple and effective, and can automatically find the optimal configuration of the model.

[0143] Step S2220, selecting valve performance influencing factors at m consecutive sampling time points within a working cycle as input, and the corresponding valve state quantity as output, to construct a training sample set;

[0144] In the actual operation of the valve, various state quantities and performance parameters are dynamically changing time series data. In order to establish a dynamic prediction model for valve performance, it is necessary to make full use of time series information. Selecting a complete operating cycle, such as a production batch or an equipment operation cycle, can cover the global change trend of valve performance. In this cycle, m time points are continuously sampled at a fixed time interval (such as 1 second) to obtain the valve performance influencing factors and corresponding state quantities at each time point to form a complete set of input-output pairs. Among them, m is the preset time window length, that is, the number of continuous time points sampled. The length of the time window needs to be reasonably set according to the time scale of the valve dynamic response. Usually, the action response time of the valve is in the order of milliseconds to seconds, so m can be between 100 and 1000. If the time window is too short, it may not capture the complete dynamic change process; if it is too long, it may introduce too much irrelevant information and increase the computational burden.

[0145] By constructing a sample from the data of m sampling points in each time window, a large number of training samples can be obtained. Each sample contains the input features of m time points and the corresponding target state quantity. This method of time window division can extract the dynamic change characteristics of valve performance and learn the time series evolution law of state quantity. At the same time, by setting an appropriate sliding step size, the number and diversity of samples can be increased, and the robustness of the model can be improved. Dividing time series data into time windows of equal length for modeling can avoid the risk of overfitting caused by too large a time span. When the time span is too long, the model may overfit some local fluctuations or noise, and the generalization performance will decrease. By dividing the time window, we can focus more on the local dynamic characteristics of valve performance while retaining certain contextual information, thereby achieving accurate state quantity prediction throughout the entire operating cycle.

[0146] Step S2230: training the valve performance prediction model using the training sample set.

[0147] Input the constructed training sample set into the random forest model and optimize the model parameters so that the predicted valve state quantity is as close to the true value as possible. In order to make full use of limited data and improve the efficiency and reliability of model training, methods such as cross-validation can be used to optimize the training process. Divide the training samples into training sets and test sets according to a certain ratio (such as 8:2), perform model learning on the training set, and evaluate the generalization performance of the model on the test set. Through multiple divisions and cross-validation, a stable estimate of the model performance can be obtained, reducing the random impact of data division.

[0148] Commonly used cross-validation methods include k-fold cross-validation and leave-one-out method. Taking 5-fold cross-validation as an example, the training samples are randomly divided into 5 equal parts, 4 of which are selected as training sets each time, and the remaining 1 is used as the validation set to train a sub-model. Repeat 5 times to obtain 5 sub-models. The prediction results of the 5 sub-models on the validation set are averaged as the performance indicator of cross-validation. Finally, all samples are used for training to obtain the final model. Cross-validation can effectively utilize data, objectively evaluate the generalization ability of the model, and guide model optimization and parameter selection.

[0149] During the training process, you can monitor the model's fit on the training set and validation set in real time, such as mean square error, coefficient of determination and other indicators. By visually analyzing the training curve, you can determine whether the model is overfitting or underfitting, and adjust the model complexity or training strategy as needed. At the same time, you can also evaluate the contribution of different performance influencing factors to the prediction results through feature importance analysis, identify key influencing factors, and guide further optimization design.

[0150] After sufficient training and verification, the valve performance prediction model obtained can establish a nonlinear mapping relationship between valve performance influencing factors and valve state quantities. The model can quickly and accurately predict the valve state quantity based on the valve performance parameters collected in real time, providing a reliable basis for subsequent control optimization. Compared with traditional physical modeling methods, data-driven machine learning models can adaptively fit the actual operating characteristics of the valve, overcoming the limitation that physical models are difficult to fully consider various influencing factors. Through continuous learning and optimization, the prediction model can continuously improve the prediction accuracy and scope of application, laying the foundation for intelligent management and control of the entire life cycle of the valve.

[0151] Using ensemble learning algorithms such as random forests for modeling can fully explore the inherent laws of valve performance data and capture complex nonlinear relationships. The integration of decision trees can effectively improve the robustness of predictions and reduce the risk of overfitting. Through methods such as randomization and self-service sampling, the diversity and generalization ability of the model are enhanced. Technologies such as hyperparameter optimization and cross-validation further improve the scientificity and reliability of model training. Valve performance prediction methods based on big data and machine learning provide new ideas and means for applications such as valve design optimization, condition monitoring, and life estimation.

[0152] Step S2300, obtaining real-time valve performance influencing factors, and obtaining valve state quantity prediction values ​​according to the real-time valve performance influencing factors and the valve performance prediction model; the valve state quantity prediction values ​​include valve opening and closing prediction time, sealing pressure prediction value and valve opening prediction value;

[0153] In the actual valve operation process, it is necessary to obtain various parameters reflecting the valve performance in real time as the input of the prediction model to obtain the instant prediction value of the valve state quantity. This process is usually achieved through online monitoring systems and data acquisition devices.

[0154] Pressure, temperature, flow, displacement and other sensors are arranged at key positions of the valve to collect various parameters reflecting the working conditions of the valve in real time. The pressure sensor can monitor the pressure changes at the inlet and outlet of the valve, the temperature sensor can measure the medium temperature and the valve body temperature, the flow sensor can record the medium flow through the valve, and the displacement sensor can measure the opening position of the valve disc. In addition to the parameters directly measured, some comprehensive performance parameters such as pressure difference, temperature difference, flow resistance, etc. can also be obtained through analysis and calculation.

[0155] Organize the collected parameters in time series to form a real-time valve performance data stream. Perform necessary preprocessing on the data, such as denoising, filtering, normalization, etc., to improve data quality. At the same time, obtain relevant static features from the valve body database based on the material properties and structural parameters of the valve. Combine dynamic performance parameters and static features in a predefined format to form a complete input of valve performance influencing factors.

[0156] The factors affecting valve performance acquired in real time are input into the trained valve performance prediction model. After the forward calculation of the model, the predicted value of the valve state quantity can be obtained. The predicted value includes key indicators such as the opening and closing time that reflects the valve action performance, the sealing pressure that reflects the sealing performance, and the opening degree that reflects the control accuracy. The valve opening and closing prediction time can estimate the time required for the valve to complete an opening or closing action under the current working conditions. The smaller the predicted value, the more agile the valve action. The sealing pressure prediction value can predict the contact stress level of the valve sealing surface under the current stress state. The closer the predicted value is to the yield strength of the material, the closer the sealing performance is to the limit. The valve opening prediction value can estimate the current opening degree of the valve. By comparing it with the given target opening degree, the accuracy of valve control can be judged.

[0157] By obtaining the real-time valve performance influencing factors and using the performance prediction model to predict the state quantity, it is helpful to timely discover potential performance anomalies and take measures in advance to prevent failures. Accurate state quantity prediction provides reliable data support for subsequent optimization control strategies, making control actions more accurate and efficient.

[0158] Step S2400, calculating theoretical valve state quantities under different working condition levels; the theoretical valve state quantities include valve opening and closing theoretical time, sealing pressure theoretical value and valve opening theoretical value;

[0159] Specifically, based on the theories of fluid dynamics and solid mechanics, a valve flow field model and stress-strain model are established to calculate the theoretical valve state quantities under different working conditions.

[0160] The specific calculation method is as follows:

[0161] (1) Calculation of the theoretical time for valve opening and closing

[0162] The opening and closing time of the valve depends on the movement speed and stroke length of the valve disc. According to the principles of fluid mechanics, the movement of the valve disc is affected by the fluid pressure difference and fluid resistance. Assuming that the valve disc is a rigid body, the resultant force it receives is the sum of the pressure difference force and the resistance.

[0163]

[0164] Among them, F is the resultant force on the valve disc, ΔP is the pressure difference before and after the valve, A is the cross-sectional area of ​​the valve disc, ρ is the fluid density, v is the flow velocity, and Cd is the resistance coefficient of the valve disc.

[0165] According to Newton's second law, the acceleration a of the valve disc is:

[0166]

[0167] in, The quality of the valve disc.

[0168] Assuming that the valve disc moves from the fully open position (x=0) to the fully closed position (x=L) with uniform acceleration and the initial velocity is 0, the relationship between the valve disc displacement x and time t is:

[0169]

[0170] Let x = L, and solve for the theoretical opening and closing time of the valve:

[0171]

[0172] Substituting the pressure difference, flow rate and other parameters under different working conditions into the above formula, the theoretical opening and closing time of the valve under the corresponding working conditions can be obtained. The shorter the opening and closing time, the better the rapid response performance of the valve.

[0173] (2) Calculation of the theoretical value of sealing pressure

[0174] When the valve is sealed, there is contact stress between the valve disc and the valve seat. According to Hertz contact theory, assuming that the contact surface between the valve disc and the valve seat is an ideal smooth elastic spherical surface, the contact stress satisfy:

[0175]

[0176] in, is the sealing pressure on the valve disc, is the equivalent elastic modulus, is the equivalent radius of curvature.

[0177] Equivalent elastic modulus The calculation formula is:

[0178]

[0179] Among them, E 1 is the elastic modulus of the valve disc material, E 2 is the elastic modulus of the valve seat material, ν 1 is the Poisson's ratio of the valve disc material, ν 2 is the Poisson’s ratio of the valve seat material.

[0180] The equivalent radius of curvature It can be expressed by the valve disc ball center beat h and the valve seat cone angle α:

[0181]

[0182] Where d is the valve seat diameter.

[0183] Substituting the sealing pressure corresponding to the working condition level into the above formula, the theoretical contact stress of the sealing surface can be obtained. If the contact stress exceeds the yield strength of the material, there may be a risk of sealing failure.

[0184] Reasonable sealing pressure can not only provide reliable sealing performance, but also reduce wear and extend service life.

[0185] (3) Calculation of the theoretical value of valve opening

[0186] The valve opening refers to the ratio of the actual displacement of the valve disc to the rated stroke, reflecting the relative size of the flow cross-sectional area. When the valve is fully open, the opening is 100%, and when it is fully closed, it is 0%.

[0187] Assuming that the valve disc moves in a straight line and the cross section of the valve seat is circular, the relationship between the opening θ and the valve disc displacement x is:

[0188]

[0189] in, is the rated travel of the valve disc.

[0190] Combined with the calculation results of valve opening and closing time, the disc displacement x is integrated to obtain the theoretical value of valve opening.

[0191]

[0192] Where v(t) is the velocity function of the valve disc.

[0193] The valve opening directly affects the flow rate and pressure drop of the fluid. Too small an opening will limit the flow rate, while too large an opening will increase energy consumption. Therefore, optimizing the valve opening according to the operating level is of great significance for improving system efficiency and saving energy and reducing consumption.

[0194] Through the above theoretical calculations, the ideal value range of valve state quantity under different working conditions can be quantitatively analyzed. The theoretical value is used as a benchmark to provide a reference for the evaluation and optimization control of valve performance. At the same time, the theoretical calculation results also provide guidance for valve design, such as optimizing structural parameters such as valve disc quality and valve seat angle to improve the rapid response and sealing reliability of the valve.

[0195] It should be noted that the above calculations are based on some simplified assumptions, such as the rigid body assumption of the valve disc and the ideal contact of the sealing surface. In practical applications, the influence of factors such as material properties, fluid properties, and processing errors also need to be considered. The theoretical model can be further modified through numerical simulation, experimental testing, and other means to improve the accuracy and applicability of the calculation.

[0196] Step S2500, comparing the theoretical valve state quantity with the valve state quantity prediction value, and calculating the valve state quantity deviation; the valve state quantity deviation includes valve opening and closing time deviation, sealing pressure deviation and valve opening deviation;

[0197] Furthermore, if Figure 4 As shown, step S2500 includes:

[0198] Step S2510, calculating the valve opening and closing time deviation according to the valve opening and closing theoretical time and the valve opening and closing predicted time;

[0199] Step S2520, calculating the sealing pressure deviation according to the sealing pressure theoretical value and the sealing pressure predicted value;

[0200] Step S2530, calculating the valve opening deviation according to the valve opening theoretical value and the valve opening predicted value.

[0201] Specifically, by comparing the theoretical value with the predicted value, the gap between the actual performance of the valve and the ideal performance, that is, the valve state deviation, can be quantitatively evaluated. The valve state deviation reflects the degree of deterioration of the dynamic response characteristics of the valve and is an important basis for valve state monitoring and fault diagnosis. According to the valve opening and closing theoretical time Tt and the valve opening and closing predicted time Tp, the valve opening and closing time deviation Δt is calculated; Δt=Tp-Tt; when Δt is a positive value, it indicates that the actual opening and closing time is longer than the theoretical value, and the valve action response is delayed; when Δt is a negative value, it indicates that the valve has unstable factors such as overshoot or oscillation. Ideally, Δt should be close to zero, that is, the actual opening and closing time is basically consistent with the theoretical value. For example, the theoretical opening and closing time of the valve under a certain working condition level is 1.5s, and the predicted opening and closing time is 1.8s, then the opening and closing time deviation Δt=1.8-1.5=0.3s. The deviation is a positive value, indicating that the actual opening process of the valve lags behind the theoretical value. If Δt exceeds the set threshold (such as 0.5s), it is necessary to analyze the cause of valve hysteresis, such as wear of the actuator, spring fatigue, etc., and perform maintenance or replace parts in time to ensure the normal operation of the valve. By real-time monitoring of the opening and closing time deviation, the control performance of the valve can be dynamically evaluated. When the deviation exceeds the warning value, an alarm message can be automatically generated to remind the operator to take corresponding measures. At the same time, the opening and closing time deviation also provides optimization directions for the design of the valve actuator, such as reducing the inertia moment, increasing the driving force, etc., to improve the rapid response capability of the valve.

[0202] According to the sealing pressure theoretical value Ps and the sealing pressure predicted value Pm, the sealing pressure deviation Δp is calculated; Δp=Pm-Ps; the sealing pressure deviation Δp characterizes the abnormality of the valve sealing performance. When Δp is a positive value, it indicates that the actual sealing pressure is higher than the theoretical value, the valve seal is too tight, which may lead to increased wear of the sealing surface; and when Δp is a negative value, it indicates that the sealing pressure is insufficient and there is a risk of leakage. Ideally, the predicted sealing pressure value should be basically consistent with the theoretical value, that is, Δp is close to zero. Taking a valve under a certain working condition as an example, the theoretical sealing pressure is 2MPa, and the sealing pressure predicted value obtained by the valve performance prediction model is 1.5MPa, then the sealing pressure deviation Δp=1.5-2=-0.5MPa. The deviation is a negative value, indicating that the actual sealing pressure of the valve is lower than the theoretical value, and there is a hidden danger of poor sealing. It is necessary to check whether there are defects or foreign matter on the sealing surface, and if necessary, it is necessary to repair or replace the sealing components such as the valve disc and valve seat. Monitoring of sealing pressure deviation can effectively prevent accidents such as valve leakage. According to the size and duration of the deviation, the severity and development trend of the leakage can be evaluated. When the deviation exceeds the preset safety threshold, the valve can be automatically cut off to prevent the situation from escalating. At the same time, by analyzing the seal pressure deviation data, the deterioration law of the valve sealing performance can be discovered, providing a basis for formulating preventive maintenance strategies.

[0203] According to the theoretical value xt of valve opening and the predicted value xp of valve opening, the valve opening deviation Δx is calculated; Δx=xp-xt; the valve opening deviation Δx reflects the difference between the actual valve opening and the target opening. When Δx is a positive value, it indicates that the actual valve opening is greater than the theoretical value, and there is a risk of overshoot; when Δx is a negative value, it indicates that the valve opening is insufficient, which may affect the accuracy of flow control. Ideally, the predicted opening value should be close to the theoretical value, that is, Δx tends to zero. For example, the theoretical opening value of a valve under a specific working condition is 60%, and the predicted opening value obtained according to the valve performance prediction model is 55%, then the opening deviation Δx=55%-60%=-5%. The deviation is a negative value, indicating that the actual valve opening cannot reach the target value, which may be caused by the actuator response lag or insufficient control accuracy. It is necessary to check the control loop, optimize the control algorithm, and improve the timeliness and accuracy of the opening control. By real-time monitoring of the valve opening deviation, the valve control performance can be evaluated. When the deviation exceeds the allowable range, the control parameters can be automatically adjusted, such as increasing the control gain, reducing the dead zone, etc., to reduce the deviation. At the same time, the opening deviation data also provides a reference for the selection of valve actuators, such as matching higher-precision electric devices, pneumatic devices, etc., to meet the control requirements under different working conditions.

[0204] In summary, the calculation and analysis of valve state deviation can comprehensively evaluate the actual operating status of the valve, including action performance, sealing performance, control performance and other aspects. By comparing the theoretical value with the predicted value, the degree of valve performance degradation can be quantified to provide data support for status monitoring, fault diagnosis, maintenance decisions, etc. By setting the early warning threshold based on the deviation size, abnormal conditions of the valve can be discovered in time to avoid the expansion of accidents. In addition, the statistical analysis of deviation data also provides direction for valve structure optimization, control strategy improvement, etc., to further improve the various performance indicators of the valve and realize intelligent management and operation and maintenance of the valve.

[0205] Step S2600, according to the valve state quantity deviation, the valve state quantity prediction value is corrected to obtain the valve state quantity correction value, wherein the valve state quantity correction value includes the valve opening and closing correction time, the sealing pressure correction value and the valve opening correction value.

[0206] Furthermore, if Figure 5 As shown, step S2600 includes:

[0207] Step S2610, correcting the predicted valve opening and closing time according to the valve opening and closing time deviation to obtain the valve opening and closing correction time;

[0208] Step S2620, correcting the predicted sealing pressure value according to the sealing pressure deviation to obtain a corrected sealing pressure value;

[0209] Step S2630: Correct the predicted valve opening value according to the valve opening deviation to obtain a valve opening correction value.

[0210] Specifically, according to the valve opening and closing time deviation Δt, the valve opening and closing prediction time Tp is corrected to obtain the valve opening and closing correction time Tc; Tc=Tp-λt×Δt; where λt is the opening and closing time correction coefficient, and the value range is 0 to 1. The larger the λt, the greater the correction amplitude of the predicted value. The selection of the correction coefficient λt needs to comprehensively consider the dynamic characteristics and working conditions of the valve. For valves with large inertia and slow response, λt can take a smaller value to smooth the correction process and avoid oscillation caused by excessive correction. For fast-response valves, λt can be appropriately increased to track the real-time changes in the valve state. At the same time, when the working conditions fluctuate violently, it is also necessary to increase the value of λt to cope with working condition disturbances. For example, the predicted opening and closing time value of a valve under the current working conditions is 2.0s, while the theoretical value is 1.8s, then the opening and closing time deviation Δt=2.0-1.8=0.2s. Taking λt=0.5, the corrected opening and closing time is: Tc=2.0-0.5×0.2=1.9s; it can be seen that the corrected opening and closing time is closer to the theoretical value, and the actual response time of the valve has been corrected.

[0211] According to the sealing pressure deviation Δp, the sealing pressure prediction value Pm is corrected to obtain the sealing pressure correction value Pc; Pc=Pm-λp×Δp; where λp is the sealing pressure correction coefficient, and the value range is 0 to 1. The larger the λp, the greater the correction amplitude of the predicted value. The selection of the sealing pressure correction coefficient λp requires a balance between sealing reliability and seal life. Although excessive sealing pressure is conducive to improving sealing performance, it will also aggravate the wear of the sealing surface and shorten the service life of the seal. Therefore, the value of λp needs to be compromised between reliability and economy. Usually, for harsh working conditions such as high pressure, high temperature, and corrosive media, λp can be appropriately increased to ensure sufficient sealing pressure. For general working conditions, λp can be moderately selected to meet the sealing requirements while taking into account the service life of the seal. For example, the sealing pressure prediction value of a valve under specific working conditions is 1.8MPa, while the theoretical value is 2.0MPa, then the sealing pressure deviation Δp=1.8-2.0=-0.2MPa. Taking λp=0.6, the corrected sealing pressure is: Pc=1.8-0.6×(-0.2)=1.92MPa; it can be seen that the corrected sealing pressure is closer to the theoretical value and the sealing performance of the valve is improved.

[0212] According to the valve opening deviation Δx, the valve opening prediction value xp is corrected to obtain the valve opening correction value xc; xc=xp-λx×Δx; where λx is the opening correction coefficient, and the value range is 0 to 1. The larger the λx, the greater the correction amplitude of the predicted value. The selection of the opening correction coefficient λx needs to comprehensively consider the flow control accuracy and stability of the valve. A larger λx helps to quickly track the opening setting value and improve the control accuracy, but it may also cause frequent adjustments and affect the control stability. Therefore, the value of λx needs to be balanced between control quality and energy saving efficiency. For occasions with high flow control requirements, such as high-precision metering valves, λx can be appropriately increased to achieve accurate opening control. For occasions with high requirements for control stability, such as large pipeline control valves, λx can be moderately selected to avoid over-frequency opening adjustment. For example, the opening prediction value of a valve under given working conditions is 60%, while the theoretical value is 55%, then the opening deviation Δx=60%-55%=5%. Taking λx=0.8, the corrected valve opening is: xc=60%-0.8×5%=56%; it can be seen that the corrected valve opening is closer to the theoretical value and the flow control accuracy is improved.

[0213] The correction of valve state is a key link in dynamic optimization control. By correcting the predicted value of the state in real time to make it as close to the theoretical optimal value as possible, the timeliness and accuracy of valve control can be guaranteed. At the same time, by reasonably selecting the correction coefficient, a compromise can be made between control performance and reliability and economy, thus improving the comprehensive performance of the valve.

[0214] The corrected valve state quantity provides reliable state feedback for subsequent optimization control. Based on the correction value, the current control effect can be accurately evaluated, providing a basis for real-time adjustment of the control strategy. In addition, state quantity correction also provides data support for valve fault diagnosis and predictive maintenance. By analyzing the change trend of the correction value, abnormal signs of valve performance can be discovered early, and preventive measures can be taken to avoid accidents.

[0215] In short, valve state correction is the basis and premise for realizing intelligent valve control. By dynamically correcting the predicted value, the influence of modeling error and measurement noise is eliminated, providing high-quality state feedback for optimal control. At the same time, a reasonable correction strategy also provides a guarantee for the health management and long-term operation of the valve. Looking at the overall situation, state correction, valve optimization control and fault diagnosis constitute a complementary technical closed loop, jointly promoting the intelligent development of valve control and providing a solid foundation for the safe and efficient operation of valves under high parameter conditions.

[0216] Step S3000, establish a reinforcement learning environment for valve control, and build a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtain real-time optimal control actions based on the valve state quantity correction value and the valve optimization control strategy model, and execute valve optimization control.

[0217] Furthermore, step S3000 includes:

[0218] Step S3100, establishing a reinforcement learning environment for valve control, including a state space, an action space, and a reward function;

[0219] Reinforcement learning is a learning method based on the interaction between agents and the environment. Agents change the state of the environment by taking actions and receive corresponding reward feedback. Its goal is to learn an optimal strategy so that the maximum cumulative reward can be obtained by selecting a specific action under a specific state. To apply reinforcement learning in the field of valve control, it is necessary to model the valve control problem as a Markov decision process (MDP), construct the state space, action space and reward function, and form a reinforcement learning environment. Each state in the environment is a complete description of the valve, including information such as the dynamic characteristics and working conditions of the valve; each action corresponds to a control decision, that is, adjusting the specific parameters of the valve; the reward reflects the degree of pros and cons of the valve in executing the current action under the current state, taking into account the comprehensive performance indicators of the valve. By learning the optimal strategy, the agent can dynamically adjust the control parameters according to the real-time state of the valve, quickly respond to changes in working conditions, stabilize valve operation, extend the service life, and minimize energy consumption. Compared with traditional valve control methods, the reinforcement learning-based method has the advantages of strong adaptability and real-time optimization, providing a new idea for realizing intelligent control of valves.

[0220] Furthermore, if Figure 6 As shown, step S3100 includes:

[0221] Step S3110, the valve state quantity and the working condition level of the valve constitute a state space of reinforcement learning; the dimension of the state space is the product of the number of valve state quantities and the working condition level;

[0222] Specifically, the state space defines the working state of the valve system at a certain moment. The valve state (such as valve opening, sealing pressure difference, opening and closing time, etc.) and the valve working condition level (reflecting the working pressure, temperature, flow rate, etc. of the valve) are used as state variables to fully describe the immediate operating state of the valve.

[0223] Selecting appropriate state variables requires balancing the completeness of information and the compactness of dimensions. On the one hand, the included state variables should be able to fully reflect the key characteristics of the valve and provide rich state information; on the other hand, too high a dimension of the state space will affect the efficiency of learning and increase the complexity of the algorithm. Therefore, constructing the state space requires professional knowledge and engineering experience to select the most representative and sensitive state variables.

[0224] The design of the state space directly determines the input layer structure of the strategy network. For example, if three state variables are selected: valve opening, sealing pressure difference, and opening and closing time, and the working condition level is divided into five levels, the state space of reinforcement learning is 3*5=15 dimensions, and the corresponding input layer of the strategy network has 15 neurons.

[0225] Reasonable state space design can improve the effect of valve optimization control in two aspects: first, incorporating the valve operating condition level variable can enable the strategy network to perceive the operating conditions of the valve, learn adaptive control strategies under different operating conditions, and improve the robustness of control; second, using continuous valve state quantities rather than discrete valve fully open / fully closed states can accurately characterize the dynamic process of the valve, learn more fine-grained control decisions, and achieve valve optimization control in a continuous action space.

[0226] Step S3120, obtaining the regulating variables of the valve control, and forming the action space of the reinforcement learning with the regulating variables of the valve control; the dimension of the action space is the number of regulating variables;

[0227] Specifically, the action space defines the optional execution actions of the valve controller, which is usually composed of a set of adjustment variables, such as continuous variables such as the amplitude, frequency, pulse width, slope of the control signal, or discrete variables such as valve opening / closing, fast opening / slow opening, etc. These adjustment variables can directly control the operating state of the valve actuator, such as the on / off of the electromagnet, the pressure of the pneumatic component, etc.

[0228] The action space determines the output layer structure of the policy network. For example, if two continuous adjustment variables, the amplitude and frequency of the control signal, are used, the action space is 2-dimensional, and the corresponding output layer of the policy network has 2 neurons, which output the instantaneous values ​​of the two adjustment variables respectively.

[0229] The design of the action space needs to fully consider the constraints of the valve hardware and the requirements of the control function. On the one hand, the physical characteristics of the actuator (such as voltage / current range, air pressure range, etc.) limit the range of the adjustment variable; on the other hand, achieving the expected control effect requires the adjustment variable to have sufficient adjustment accuracy and response speed. Under the premise of meeting the constraints, the action space should cover as many possible control actions as possible to provide sufficient exploration space for strategy optimization.

[0230] Although continuous action space brings greater flexibility to strategy search, it also increases the difficulty of optimization. Therefore, the adjustment variables can be further discretized in the continuous action space to simplify the action space to a certain extent and balance the optimization efficiency and control performance. In addition, adding appropriate exploration noise in the action space can also help the algorithm jump out of the local optimum and learn a more robust control strategy.

[0231] By systematically designing the state space and action space, the valve control problem is abstracted into a Markov decision process. Intelligent optimization algorithms such as reinforcement learning can be applied to solve the optimal control strategy and realize intelligent adaptive control of the valve. Among them, the state space is constructed with the valve state quantity and working condition level, which can describe the dynamic characteristics and working condition adaptability of the valve; the action space is constructed with continuous or discrete adjustment variables, which can flexibly control the valve actuator. The dimensions and quantization methods of the state space and action space directly affect the complexity and optimization efficiency of the control model. It is necessary to combine domain knowledge and engineering constraints for careful design, and strike a balance between information completeness and search efficiency to give full play to the potential of reinforcement learning in valve optimization control.

[0232] Step S3130, calculating valve performance indicators, and constructing a reinforcement learning reward function R based on the valve performance indicators; the valve performance indicators include a rapid response performance indicator R1, a sealing reliability indicator R2, a service life indicator R3, and an energy consumption level indicator R4.

[0233] Valve performance indicators are key factors in evaluating the comprehensive performance of valves, reflecting the control effect and operating status of valves. Quantifying various performance indicators into reward functions can guide reinforcement learning algorithms to optimize the dynamic control strategy of valves and achieve multi-objective balance.

[0234] The rapid response performance index R1 measures the rapidity of valve action and reflects the valve's ability to execute control instructions. The sealing reliability index R2 evaluates the integrity and durability of the valve seal and reflects the safe operation of the valve. The service life index R3 considers the durability and maintainability of the valve, which is related to the long-term stable operation of the valve. The energy consumption level index R4 measures the energy efficiency of the valve control process and reflects the economic cost of valve operation.

[0235] In order to integrate the above performance indicators into a reward function, it is necessary to comprehensively consider the importance and interaction of each indicator. The design of the reward function needs to weigh various indicators and focus on them according to actual needs. By properly configuring the indicator weights, the reinforcement learning algorithm can meet the key performance constraints while taking into account other secondary indicators, and ultimately achieve global optimization of the comprehensive performance of the valve.

[0236] Furthermore, if Figure 7 As shown, step S3130 includes:

[0237] Step S3131, calculating the weighted average of the valve opening and closing time deviation, the sealing pressure deviation and the valve opening deviation as the rapid response performance index R1;

[0238] Specifically, the weight coefficients of various deviations are set for different working conditions and control requirements. The larger the weight coefficient, the higher the importance attached to the corresponding indicator. When fast action is required, the weight of the opening and closing time deviation can be increased; when the sealing requirement is strict, the weight of the sealing pressure deviation should be appropriately increased. R1=w1*|Δt|+w2*|Δp|+w3*|Δx|; where Δt, Δp, and Δx are the valve opening and closing time deviation, sealing pressure deviation, and opening deviation, respectively; w1, w2, and w3 are the corresponding weight coefficients. Generally speaking, the opening and closing time deviation is more sensitive to the rapidity requirement, and a larger weight w1 should be given; while the sealing pressure deviation and opening deviation have relatively little effect on the rapid response, and the values ​​of w2 and w3 can be appropriately reduced. However, the specific weight distribution also needs to be adjusted according to the type of valve, working conditions, etc. The rapid response performance index R1 reflects the degree of deviation between the actual dynamic response characteristics of the valve and the ideal response characteristics. The larger the R1 value, the greater the deviation between the actual response speed, sealing performance and opening control accuracy of the valve and the expected value, and the worse the rapid response performance.

[0239] By calculating R1 by weighted average method, the response performance of the valve in terms of action speed, sealing reliability, control accuracy, etc. can be comprehensively evaluated, and the dynamic characteristics of the valve can be fully reflected. This provides a quantitative basis for valve status monitoring, fault diagnosis and control optimization. At the same time, the calculation of R1 index is simple and easy, and can be evaluated online in real time, providing feedback for valve intelligent control and realizing closed-loop optimization.

[0240] By evaluating the rapid response performance of the valve based on the R1 index, the design and production departments can be urged to improve the dynamic performance of the valve, such as optimizing the valve structure and selecting materials reasonably. The maintenance department can also be urged to strengthen the daily maintenance of the valve, such as regular calibration and replacement of wearing parts, so as to maximize the dynamic response capability of the valve and improve the safety and reliability of the system.

[0241] Step S3132, obtaining the yield strength σs of the sealing surface material from the valve body database, calculating the contact stress σ of the valve disc under the sealing pressure; and evaluating the sealing reliability index R2 according to the ratio of the contact stress σ to the yield strength σs;

[0242] Specifically, firstly, according to the valve structural parameters and force analysis, the contact stress σ of the valve disc under the action of the theoretical sealing pressure Ps is calculated. Taking the valve disc sealing surface as the research object, considering the pressure load and constraints it is subjected to, the mechanical equilibrium equation is established. Assuming that the sealing surface is an ideal rigid body and the stress concentration effect is not significant, the Hertz contact theory can be used to solve and obtain the contact stress σ. By comparing the obtained contact stress σ with the yield strength σs of the sealing surface material, the sealing reliability index R2 can be evaluated. R2=1-σ / σs; the value range of R2 is 0 to 1. When σ is much smaller than σs, R2 is close to 1, indicating that the sealing surface works in the elastic deformation stage under the current sealing pressure and the sealing reliability is high; when σ is close to or even exceeds σs, R2 will approach or even be less than 0, indicating that the sealing surface has undergone plastic deformation and there is a risk of sealing failure.

[0243] The significance of the R2 index is that it establishes a quantitative correlation between the mechanical properties of the sealing surface material and the valve sealing reliability. Through the online calculation and monitoring of the R2 index, the status and trend of the valve sealing performance can be evaluated in real time. On the one hand, the R2 index can be used for valve status monitoring and fault warning. When R2 drops below the preset threshold, it can automatically alarm to indicate that the sealing surface is at risk of failure, so that timely measures such as maintenance or replacement can be taken to avoid leakage accidents. On the other hand, the R2 index can also be used as a basis for valve design and material selection. By optimizing structural parameters such as the valve seat cone angle and curvature radius, and selecting high-strength and wear-resistant sealing materials, the service life and sealing performance of the valve can be maximized, and maintenance costs can be reduced while ensuring that the R2 index is qualified.

[0244] Step S3133, obtaining the cumulative number of valve actions N, and evaluating the service life index R3 of the valve according to the cumulative number of valve actions N and the rated number of valve actions N0;

[0245] The service life indicator R3 characterizes the remaining service life of the valve, reflecting the current health status and failure risk of the valve. R3 is equal to the ratio of the remaining number of valve operations to the rated number of operations, and its value range is [0,1]. When R3 is close to 1, it indicates that the valve is still in a healthy state and the remaining number of operations is sufficient; when R3 tends to 0, it means that the valve has approached or reached its design life, and the failure risk increases sharply.

[0246] The rated number of valve operations N0 is calculated based on the valve design requirements and applicable working conditions, combined with empirical data and reliability theory. It comprehensively considers factors such as the valve's structural form, material properties, and manufacturing process, and is the number of operations that the valve can reliably complete under specified conditions. The rated number of operations is usually provided by the valve manufacturer as an important indicator of product quality and service life.

[0247] The actual cumulative number of valve actuations N can be obtained in a variety of ways. One way is to install a counter on the valve drive device before the valve is put into use to record the number of valve actuations in real time. Another way is to determine the cumulative number of actuations by manual counting or detecting the change in the switch state during routine valve maintenance. With the development of smart instruments and industrial Internet of Things technology, wireless sensors can also be used to collect valve actuation signals in real time, and then count and upload them through edge computing devices.

[0248] After obtaining the rated number of actions N0 and the actual number of actions N, the service life index R3 can be calculated as follows: R3=(N0-N) / N0; R3 intuitively shows the proportion of the remaining number of valve actions to the rated number of actions. For example, if the rated number of actions of a valve is 5000 times and the current actual number of actions is 3000 times, then its service life index R3 is: R3=(5000-3000) / 5000=0.4; this means that the valve has consumed 60% of its service life and the remaining life is 40%. It is in the middle and late stages of use and needs to strengthen status monitoring and maintenance.

[0249] The significance of the service life index R3 is that it can help engineers scientifically formulate valve maintenance strategies and replacement plans, and reduce safety risks and economic losses caused by unexpected failures. By tracking the changing trend of the R3 index in real time, the health status and degradation law of the valve can be mastered. When R3 drops to the warning threshold, a work order can be automatically generated to prompt relevant personnel to carry out preventive maintenance, such as replacing seals and tightening connectors, to delay the degradation of valve performance. When R3 further drops to the replacement threshold, replacement of the valve should be considered to completely eliminate the potential safety hazards. At the same time, the R3 index is also an important basis for valve design and selection. By analyzing the R3 attenuation law of valves of different models and materials, the valve structure design can be optimized, the processing technology can be improved, and the durability and reliability of the valve can be improved. For specific working conditions, the appropriate valve type and material grade can be selected according to the R3 index requirements. Under the premise of ensuring safety and reliability, the service life of the valve can be maximized, the replacement frequency can be reduced, and the cost of the entire life cycle can be saved.

[0250] Step S3134, obtaining the energy consumption E of a single action of the valve, and evaluating the energy consumption level index R4 of the valve according to the energy consumption E of a single action of the valve;

[0251] Specifically, R4=Emax-E, where Emax is the maximum energy consumption allowed for a single action. The higher the energy consumption E, the lower the energy consumption level index R4. The energy consumption E of a single valve action can be calculated by measuring parameters such as current, voltage, and duration. For example, for electromagnetically driven valves, the voltage U, current I, and power-on time t of the control signal can be used to estimate the energy consumption of a single action: E=U*I*t; reducing the amplitude of the control signal and optimizing the signal excitation method can reduce energy loss while ensuring the control effect. For example, on the premise of meeting the action speed requirements, appropriately reduce the voltage and current to reduce the Joule heat loss of the transmission components; optimize the on-off time through PWM modulation and other methods to reduce the eddy current loss of the coil, etc.

[0252] The introduction of energy consumption level indicators enables the reinforcement learning algorithm to consider reducing energy costs while pursuing performance goals. In scenarios where energy is scarce or cost pressure is high, the algorithm can be guided to learn energy-saving control strategies by increasing the weight of the energy consumption level indicator. In the long run, reducing energy consumption can not only save operating costs, but also help extend the service life of valve components and improve the overall economy of the system.

[0253] At the same time, the collection and analysis of energy consumption data also provides data support for the energy efficiency management of valves. By tracking the energy consumption trend of valves, energy efficiency anomalies can be discovered, providing an entry point for energy-saving transformation. By comparing the energy consumption levels of different models and different working conditions, energy-saving potential can be tapped and valve selection and system design can be optimized. Based on energy consumption big data, a valve energy efficiency model can also be established to support the optimization and scheduling of the energy management system.

[0254] In short, the design of energy consumption level indicators enables the reinforcement learning algorithm to balance performance and cost and learn more economical, efficient and sustainable valve control strategies. The systematic collection and application of energy consumption data also lays the foundation for energy management throughout the valve life cycle, and promotes the green and low-carbon development of industrial valves.

[0255] Step S3135, taking a weighted average of the quick response performance index R1, the sealing reliability index R2, the service life index R3 and the energy consumption level index R4 to obtain a reward function R of reinforcement learning.

[0256] Specifically, R=λ1*R1+λ2*R2+λ3*R3+λ4*R4, where λ1, λ2, λ3, and λ4 are weight coefficients of each indicator, reflecting the emphasis of different performance requirements. The weight coefficients λ1~λ4 here are a set of hyperparameters that need to be adjusted according to the application scenario and demand preferences. For example, in situations where the dynamic response speed is required to be high (such as emergency shut-off valves), the weight λ1 of the rapid response performance indicator can be appropriately increased; in situations where the sealing performance is required to be strict (such as high pressure and high temperature or toxic media), the weight λ2 of the sealing reliability indicator needs to be increased; when the maintenance cost of the valve is very high or it is difficult to replace, the weight λ3 of the service life indicator should be higher; and in an environment where energy is scarce or cost pressure is high, the weight λ4 of the energy consumption level indicator should be increased.

[0257] The design of the reward function needs to take into account various performance requirements and make trade-offs based on actual needs. By properly configuring weights, the reinforcement learning algorithm can maximize the comprehensive performance of the valve while satisfying the constraints and achieve optimal control. As the core element of reinforcement learning, the rationality of the reward function design directly affects the convergence speed and control effect of the algorithm.

[0258] On the one hand, the reward function should be able to accurately evaluate the real-time control performance of the valve and sensitively reflect the changes in the valve state. The above design quantifies the key performance indicators of the valve into reward values, which can comprehensively examine the rapidity, reliability, durability and economy of the valve action, making the reward function more representative and discriminative.

[0259] On the other hand, the reward function should also guide the agent to learn to optimize long-term returns and weigh short-term benefits against future benefits. By adjusting the weight coefficients of various indicators, valve performance can be balanced on different time scales to avoid being too short-sighted or blindly pursuing a single goal. For example, excessive pursuit of fast action may damage seals, thereby reducing long-term reliability; short service life for temporary energy saving may increase replacement costs, etc.

[0260] The debugging of the reward function is also a gradual process, which requires continuous iterative optimization in simulation tests and physical verification. In the early stage, the weights can be set according to expert experience and adjusted appropriately during the training process. With the accumulation of working condition data and the maturity of the algorithm, the weight coefficients can be further optimized automatically using data-driven methods, and even the weight coefficients can be used as the output of the policy network to achieve adaptive design of the reward function and improve the universality and robustness of the algorithm.

[0261] In short, designing a reward function in the form of multi-objective weighting can balance the various performance requirements of valve control and maximize the overall benefits while satisfying the constraints. Systematically quantifying key performance indicators, reasonably weighing immediate benefits and long-term returns, and supplemented by data-driven optimization methods can further improve the rationality and effectiveness of the reward function, thereby accelerating the convergence of the reinforcement learning algorithm and improving the effect and reliability of valve control.

[0262] Step S3200, constructing a valve optimization control strategy model based on a valve control reinforcement learning environment;

[0263] Further, step S3200 includes:

[0264] Step S3210, using the DDPG algorithm as a deep reinforcement learning framework for valve control;

[0265] Step S3220, building a neural network structure of a policy network and a Q-value network;

[0266] Step S3230, sampling historical operating condition data from the valve operating condition database to construct a training data set for reinforcement learning;

[0267] Step S3240, using the training data set of reinforcement learning, training the strategy network through the DDPG algorithm to obtain the valve optimization control strategy model.

[0268] Specifically, the valve optimization control strategy model is an end-to-end control model based on reinforcement learning, which can directly generate the optimal control action according to the state of the valve and realize the intelligent adaptive control of the valve. Compared with traditional PID control, optimal control and other methods, this model has the following advantages: First, it can autonomously learn the dynamic characteristics and control laws of the valve from the data, without the need to establish an accurate mathematical model in advance, reducing the complexity of modeling; second, the model can adaptively adjust the control strategy to adapt to changes in working conditions through continuous trial and error and learning, and has stronger robustness and adaptability; third, the end-to-end control mode can simplify the design of the control system, reduce intermediate links, and improve the immediacy and accuracy of control. Therefore, the valve optimization control strategy model provides a new idea and method for realizing intelligent control of valves.

[0269] The construction of the valve optimization control strategy model mainly includes the following key steps: using a suitable reinforcement learning algorithm (such as DDPG) as the basic framework; designing the structure of the strategy network and the value function network according to the characteristics of the valve control problem; sampling the training data set from the historical operating data; using the training data set to train the optimization strategy model through the reinforcement learning algorithm.

[0270] In terms of algorithm selection, the DDPG algorithm is a deep reinforcement learning algorithm based on the actor-critic architecture, which is particularly suitable for control problems in continuous action spaces. Unlike traditional value function-based reinforcement learning algorithms such as DQN, DDPG introduces a deterministic policy gradient policy gradient algorithm to directly optimize the policy function, overcoming the difficulty of policy optimization in high-dimensional or continuous action spaces. At the same time, DDPG applies the experience replay and target network mechanisms in DQN to the actor-critic framework, effectively improving the stability and sample efficiency of training. Therefore, the DDPG algorithm is very suitable for continuous control scenarios such as valves.

[0271] In terms of network structure design, both the policy network (actor) and the value function network (critic) use multi-layer feedforward neural networks. With state variables as input, the policy network outputs deterministic continuous actions, and the value function network outputs the value of the state-action pair. The parameters such as the number of network layers, the number of neurons, and the activation function need to be adjusted and optimized according to the complexity of the problem and the scale of the data.

[0272] In terms of training data preparation, a batch of initial states can be sampled from the valve historical operating condition database, and then state-action transition data can be generated through random exploration or bootstrap. Using the valve performance prediction model, the next state and immediate reward after the transfer can be estimated. Collecting the real-time operating condition data of the valve online and incorporating it into the training set can enable the strategy model to continuously learn and optimize.

[0273] During the training process, the policy network and the value function network are alternately trained to achieve cyclic iterative optimization of policy evaluation and policy improvement. The DDPG algorithm uses the value function network (critic) to evaluate the action value generated by the policy network (actor), and then uses the deterministic policy gradient to increase the parameters of the policy network to optimize it in the direction of maximizing the value; at the same time, the TD error gradient is used to decrease the parameters of the value function network so that its estimated action value is close to the true value. In addition, mechanisms such as experience replay and soft update ensure the stability of the training process.

[0274] The constructed valve optimization control strategy model actually plays the role of an end-to-end controller, directly mapping the valve state to the control action. Compared with the traditional hierarchical control architecture, the end-to-end control method reduces the intermediate feature engineering and rule design links, avoiding the limitations of manual experience. The learned strategy model can adaptively adjust the optimization control strategy according to the change of state, overcoming the difficulty of PID parameter tuning. At the same time, because the strategy model can directly process continuous state and action space, it avoids the loss of accuracy caused by the discretization of actions in algorithms such as DQN.

[0275] The construction of the valve optimization control strategy model provides a data-driven intelligent optimization method for valve control under complex working conditions. By designing a reasonable reinforcement learning environment and network structure, and using historical and real-time working condition data for training and optimization, the end-to-end control strategy obtained can adaptively generate optimal control instructions, while meeting various constraints, maximizing the dynamic performance and operating efficiency of the valve, and providing an effective solution for achieving precise and intelligent optimization control of high-parameter control valves.

[0276] Step S3300, obtain the real-time optimal control action according to the valve state quantity correction value and the valve optimization control strategy model, and perform valve optimization control.

[0277] Specifically, the valve optimization control strategy model obtained through offline training can be deployed to the valve control unit to achieve real-time optimization control of the valve. Specifically, during the operation of the valve, the valve status information (such as valve position, pressure difference, temperature, etc.) is collected in real time, and the state is corrected according to the deviation between the valve state quantity and the theoretical value to obtain the corrected state quantity. The corrected state quantity is input into the valve optimization control strategy model, and the strategy model can directly output the optimal control action under the current state (such as the amplitude and frequency of the control signal and other regulating variables). The optimal control action is converted into a specific control instruction, and the valve opening, speed, etc. are adjusted in real time through the actuator (such as an electric head, a pneumatic head, etc.) to achieve optimal control of the valve.

[0278] The combination of valve state correction and valve optimization control strategy model can overcome the complexity and uncertainty of valve dynamic characteristics and realize precise control of working condition adaptation. State correction is to correct the deviation between the actual state and the ideal state of the valve to eliminate the influence of measurement error, linearization error, etc. The optimization control strategy model is an end-to-end mapping relationship, which can directly give the optimal control quantity according to the corrected state quantity, avoiding the complex parameter setting process in traditional PID control. The execution of valve optimization control can effectively improve the dynamic response characteristics of the valve, such as improving rapidity, reducing overshoot, suppressing oscillation, etc., thereby improving the control quality of the valve and the system efficiency.

[0279] In the process of real-time optimization control, it is necessary to continuously monitor the valve state quantity and evaluate the control effect. On the one hand, by comparing the changes in the valve state quantity before and after control, the effectiveness of the optimization control action can be judged; on the other hand, the new sample data generated by the optimization control process can be fed back to the valve condition database for offline updating and optimization of the strategy model to achieve continuous iterative optimization of the strategy. In addition, valve fault diagnosis and fault-tolerant control are also key points that need to be paid attention to in real-time control. By setting reasonable state quantity and control quantity thresholds, valve faults can be discovered and diagnosed in time; when a fault occurs, it can be switched to a pre-designed safety control mode to ensure the stable operation of the system.

[0280] In short, the organic combination of valve state correction and valve optimization control strategy model establishes a closed-loop real-time optimization control mechanism. Data-driven state correction eliminates the impact of valve nonlinearity, end-to-end intelligent strategy model improves control accuracy and efficiency, and real-time feedback optimization iteration improves control continuity and adaptability. This method provides an effective solution for the precise control of high-parameter control valves under complex working conditions, which is of great significance for improving the control performance and operating efficiency of valves and even the entire process system.

[0281] Example 2

[0282] This embodiment provides a design device for a fast-response steam extraction check valve under high-parameter conditions on the basis of embodiment 1, such as Figure 8 As shown, including:

[0283] Database construction module: collect the first operating condition parameters of the target steam extraction check valve in the historical operation cycle to build a valve operating condition database; obtain valve structure parameters and material property parameters to build a valve body database;

[0284] Valve state prediction module: Based on the valve operating condition database and the valve body database, a valve performance prediction model is constructed; real-time valve performance influencing factors are obtained, and the valve state quantity prediction value is obtained according to the real-time valve performance influencing factors and the valve performance prediction model;

[0285] Valve state correction module: calculate the theoretical valve state quantity under different working conditions, compare the theoretical valve state quantity with the valve state quantity prediction value, and calculate the valve state quantity deviation; according to the valve state quantity deviation, correct the valve state quantity prediction value to obtain the valve state quantity correction value;

[0286] Valve optimization control module: Establish a reinforcement learning environment for valve control, and build a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtain the real-time optimal control action based on the valve state correction value and the valve optimization control strategy model, and execute valve optimization control.

[0287] In the database construction module, the valve operating condition database includes n 1 Working condition records, each working condition record includes n 2 The attribute data of each sampling time point includes a sampling timestamp, a pressure value, a temperature value, a flow value and a valve state quantity; the valve state quantity includes a valve opening and closing time, a sealing pressure and a valve opening degree.

[0288] In the valve state prediction module, the construction of the valve performance prediction model includes:

[0289] Step S2100, integrating the valve working condition database and the valve body database to obtain a data set of valve performance influencing factors; the valve performance influencing factors include working condition level, valve structure parameters and material property parameters;

[0290] Step S2200, constructing and training a valve performance prediction model with valve performance influencing factors as input and valve state quantity as output.

[0291] The step S2200 includes:

[0292] Step S2210, using a random forest algorithm to construct a valve performance prediction model;

[0293] Step S2220, selecting valve performance influencing factors at m consecutive sampling time points within a working cycle as input, and the corresponding valve state quantity as output, to construct a training sample set;

[0294] Step S2230: training the valve performance prediction model using the training sample set.

[0295] In the valve state correction module, the calculation of the valve state deviation includes:

[0296] Step S2510, calculating the valve opening and closing time deviation according to the valve opening and closing theoretical time and the valve opening and closing predicted time;

[0297] Step S2520, calculating the sealing pressure deviation according to the sealing pressure theoretical value and the sealing pressure predicted value;

[0298] Step S2530, calculating the valve opening deviation according to the valve opening theoretical value and the valve opening predicted value.

[0299] In the valve state correction module, obtaining the valve state correction value includes:

[0300] Step S2610, correcting the predicted valve opening and closing time according to the valve opening and closing time deviation to obtain the valve opening and closing correction time;

[0301] Step S2620, correcting the predicted sealing pressure value according to the sealing pressure deviation to obtain a corrected sealing pressure value;

[0302] Step S2630: Correct the predicted valve opening value according to the valve opening deviation to obtain a valve opening correction value.

[0303] In the valve optimization control module, the reinforcement learning environment for establishing valve control includes:

[0304] Step S3110, the valve state quantity and the working condition level of the valve constitute a state space of reinforcement learning; the dimension of the state space is the product of the number of valve state quantities and the working condition level;

[0305] Step S3120, obtaining the regulating variables of the valve control, and forming the action space of the reinforcement learning with the regulating variables of the valve control; the dimension of the action space is the number of regulating variables;

[0306] Step S3130, calculating valve performance indicators, and constructing a reinforcement learning reward function R based on the valve performance indicators; the valve performance indicators include a rapid response performance indicator R1, a sealing reliability indicator R2, a service life indicator R3, and an energy consumption level indicator R4.

[0307] The step S3130 includes:

[0308] Step S3131, calculating the weighted average of the valve opening and closing time deviation, the sealing pressure deviation and the valve opening deviation as the rapid response performance index R1;

[0309] Step S3132, obtaining the yield strength σs of the sealing surface material from the valve body database, calculating the contact stress σ of the valve disc under the sealing pressure; and evaluating the sealing reliability index R2 according to the ratio of the contact stress σ to the yield strength σs;

[0310] Step S3133, obtaining the cumulative number of valve actions N, and evaluating the service life index R3 of the valve according to the cumulative number of valve actions N and the rated number of valve actions N0;

[0311] Step S3134, obtaining the energy consumption E of a single action of the valve, and evaluating the energy consumption level index R4 of the valve according to the energy consumption E of a single action of the valve;

[0312] Step S3135, taking a weighted average of the quick response performance index R1, the sealing reliability index R2, the service life index R3 and the energy consumption level index R4 to obtain a reward function R of reinforcement learning.

[0313] Example 3

[0314] This embodiment discloses an electronic device, which may include one or more processors and one or more memories, wherein the memories store computer readable codes, and when the computer readable codes are executed by the one or more processors, the design method of the fast response steam extraction check valve under high parameter conditions as described above may be executed.

[0315] The method or system according to the embodiment of the present application can also be implemented with the help of the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in the electronic device, such as a ROM or a hard disk, can store the design method of a fast-response steam extraction check valve under high parameter conditions provided in the present application. The design method of a fast-response steam extraction check valve under high-parameter working conditions may, for example, include: collecting the first working condition parameters of the target steam extraction check valve in the historical operating cycle to construct a valve working condition database; obtaining valve structural parameters and material property parameters to construct a valve body database; constructing a valve performance prediction model based on the valve working condition database and the valve body database; obtaining real-time valve performance influencing factors, and obtaining valve state quantity prediction values ​​based on the real-time valve performance influencing factors and the valve performance prediction model; calculating the theoretical valve state quantity under different working condition levels, comparing the theoretical valve state quantity with the valve state quantity prediction value, and calculating the valve state quantity deviation; correcting the valve state quantity prediction value based on the valve state quantity deviation to obtain the valve state quantity correction value; establishing a reinforcement learning environment for valve control, and constructing a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtaining real-time optimal control actions based on the valve state quantity correction value and the valve optimization control strategy model, and performing valve optimization control.

[0316] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in the present invention is only exemplary, and when implementing different devices, one or more components in the electronic device disclosed in the present invention may be omitted according to actual needs.

[0317] Example 4

[0318] This embodiment discloses a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the design method of a fast-response steam extraction check valve under high-parameter conditions of the embodiment of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0319] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart may be implemented as a computer software program. For example, the present application provides a non-temporary machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: collecting the first operating condition parameters of the target steam extraction check valve in the historical operation cycle, and constructing a valve operating condition database; obtaining valve structural parameters and material property parameters, and constructing a valve body database; constructing a valve performance prediction model based on the valve operating condition database and the valve body database; obtaining real-time valve performance influencing factors, and obtaining valve state quantity prediction values ​​based on the real-time valve performance influencing factors and the valve performance prediction model; calculating the theoretical valve state quantity under different operating condition levels, comparing the theoretical valve state quantity with the valve state quantity prediction value, and calculating the valve state quantity deviation; correcting the valve state quantity prediction value based on the valve state quantity deviation to obtain the valve state quantity correction value; establishing a reinforcement learning environment for valve control, and constructing a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtaining real-time optimal control actions based on the valve state quantity correction value and the valve optimization control strategy model, and performing valve optimization control. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed.

[0320] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0321] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0322] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A design method for a fast-response steam extraction check valve under high-parameter conditions, characterized in that: The method comprises: Collect the first operating condition parameters of the target steam extraction check valve in the historical operation cycle to build a valve operating condition database; obtain valve structure parameters and material property parameters to build a valve body database; Based on the valve working condition database and the valve body database, a valve performance prediction model is constructed; the real-time valve performance influencing factors are obtained, and the valve state quantity prediction value is obtained according to the real-time valve performance influencing factors and the valve performance prediction model; the theoretical valve state quantity under different working condition levels is calculated, and the theoretical valve state quantity is compared with the valve state quantity prediction value, and the valve state quantity deviation is calculated; according to the valve state quantity deviation, the valve state quantity prediction value is corrected to obtain the valve state quantity correction value; Establish a reinforcement learning environment for valve control, and build a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtain the real-time optimal control action based on the valve state correction value and the valve optimization control strategy model, and perform valve optimization control; The valve state deviation includes valve opening and closing time deviation, sealing pressure deviation and valve opening deviation; The establishment of the reinforcement learning environment for valve control includes: calculating valve performance indicators, and constructing a reinforcement learning reward function R according to the valve performance indicators; the valve performance indicators include a rapid response performance indicator R1, a sealing reliability indicator R2, a service life indicator R3, and an energy consumption level indicator R4; The reward function R for constructing reinforcement learning includes: Calculate the weighted average of valve opening and closing time deviation, sealing pressure deviation and valve opening deviation as the rapid response performance index R1; Obtain the yield strength σs of the sealing surface material from the valve body database, and calculate the contact stress σ of the valve disc under the sealing pressure; evaluate the sealing reliability index R2 based on the ratio of the contact stress σ to the yield strength σs; Obtain the cumulative number of valve actions N, and evaluate the valve service life index R3 based on the cumulative number of valve actions N and the rated number of valve actions N0; Obtain the energy consumption E of a single action of the valve, and evaluate the energy consumption level index R4 of the valve according to the energy consumption E of a single action of the valve; The quick response performance index R1, sealing reliability index R2, service life index R3 and energy consumption level index R4 are weighted averaged to obtain the reward function R of reinforcement learning.

2. The design method of the fast response steam extraction check valve under high parameter conditions according to claim 1 is characterized in that: The valve operating condition database includes n1 operating condition records, each of which includes attribute data of n2 sampling time points; the attribute data of each sampling time point includes sampling timestamp, pressure value, temperature value, flow value and valve state quantity; the valve state quantity includes valve opening and closing time, sealing pressure and valve opening; According to the pressure values, temperature values ​​and flow values ​​of n1 operating condition records in the valve operating condition database, the n1 operating condition records are clustered, and the n1 operating condition records in the valve operating condition database are divided into N1 operating condition levels; the operating condition levels are numbered in sequence from low to high according to the clustering results.

3. The design method of the fast response steam extraction check valve under high parameter conditions according to claim 2 is characterized in that: The clustering of n1 working condition records includes: Step S1210, constructing an operating condition feature vector according to the pressure value, temperature value and flow value in the operating condition record; clustering the n1 operating condition records according to the operating condition feature vector to obtain N1 cluster centers; each cluster center corresponds to an operating condition level, and each operating condition record is divided into the operating condition level to which the cluster center closest to it belongs; Step S1220, calculate the silhouette coefficient S, if S>S0, output the clustering results and the working condition classification results; if S≤S0, adjust the number of cluster centers N1, return to step S1210 and re-cluster until S>S0; wherein S0 is the preset silhouette coefficient threshold.

4. The design method of a fast-response steam extraction check valve under high-parameter conditions according to claim 2 is characterized in that: The construction of the valve performance prediction model comprises: The valve operating condition database and the valve body database are integrated to obtain a data set of valve performance influencing factors; the valve performance influencing factors include operating condition level, valve structure parameters and material property parameters; A valve performance prediction model was constructed using random forest algorithm; Select the valve performance influencing factors at m consecutive sampling time points within a working cycle as input, and the corresponding valve state quantity as output to construct a training sample set; The valve performance prediction model is trained using the training sample set.

5. The design method of a fast-response steam extraction check valve under high-parameter conditions according to claim 1 is characterized in that: The valve state quantity prediction value includes the valve opening and closing prediction time, the sealing pressure prediction value and the valve opening prediction value; the theoretical valve state quantity includes the valve opening and closing theoretical time, the sealing pressure theoretical value and the valve opening theoretical value; The calculation of the valve state deviation comprises: Calculate the valve opening and closing time deviation based on the valve opening and closing theoretical time and the valve opening and closing predicted time; Calculate the sealing pressure deviation according to the sealing pressure theoretical value and the sealing pressure predicted value; The valve opening deviation is calculated based on the valve opening theoretical value and the valve opening predicted value.

6. The design method of a fast-response steam extraction check valve under high-parameter conditions according to claim 5 is characterized in that: The valve state correction value includes valve opening and closing correction time, sealing pressure correction value and valve opening correction value; The method of correcting the predicted value of the valve state quantity to obtain the corrected value of the valve state quantity includes: According to the valve opening and closing time deviation, the valve opening and closing prediction time is corrected to obtain the valve opening and closing correction time; According to the sealing pressure deviation, the sealing pressure prediction value is corrected to obtain the sealing pressure correction value; According to the valve opening deviation, the valve opening prediction value is corrected to obtain the valve opening correction value.

7. The design method of a fast-response steam extraction check valve under high-parameter conditions according to claim 3 is characterized in that: The valve control reinforcement learning environment includes a state space, an action space and a reward function; The reinforcement learning environment for establishing valve control also includes: The valve state quantity and the working condition level of the valve constitute the state space of reinforcement learning; the dimension of the state space is the product of the number of valve state quantities and the working condition level; The regulating variables of the valve control are obtained, and the regulating variables of the valve control are used to form an action space of reinforcement learning; the dimension of the action space is the number of regulating variables.

8. A design device for a fast-response steam extraction check valve under high-parameter working conditions, which is used to implement a design method for a fast-response steam extraction check valve under high-parameter working conditions as claimed in any one of claims 1 to 7, characterized in that: The device comprises: Database construction module: collect the first operating condition parameters of the target steam extraction check valve in the historical operation cycle to build a valve operating condition database; obtain valve structure parameters and material property parameters to build a valve body database; Valve state prediction module: Based on the valve operating condition database and the valve body database, a valve performance prediction model is constructed; real-time valve performance influencing factors are obtained, and the valve state quantity prediction value is obtained according to the real-time valve performance influencing factors and the valve performance prediction model; Valve state correction module: calculate the theoretical valve state quantity under different working conditions, compare the theoretical valve state quantity with the valve state quantity prediction value, and calculate the valve state quantity deviation; according to the valve state quantity deviation, correct the valve state quantity prediction value to obtain the valve state quantity correction value; Valve optimization control module: Establish a reinforcement learning environment for valve control, and build a valve optimization control strategy model based on the reinforcement learning environment for valve control; obtain the real-time optimal control action based on the valve state correction value and the valve optimization control strategy model, and execute valve optimization control.

Images