Device operating state control method and apparatus based on physical constraint inversion operator

Abstract

The application discloses a device operation state control method and device based on a physical constraint inversion operator, and relates to the technical field of industrial process control. The method comprises the following steps: inputting obtained target sparse measurement data, a target executed control amount, target physical condition data and a reconstructed state field coordinate into a target physical constraint inversion operator to obtain a target reconstructed state field; determining four kinds of residuals based on the target reconstructed state field, the target sparse measurement data, a physical residual operator and a reduced basis matrix; generating a reduced state error upper bound based on a domain parameter set and the four kinds of residuals; determining a risk index based on a plurality of sensitivity upper bounds, the reduced state error upper bound and an obtained low-dimensional representation vector; and determining industrial state data based on the domain parameter set, the target sparse measurement data and the risk index to control the target device to be in a normal operation state, so that the technical problem that it is difficult to accurately control the device operation state is solved.

Description

Technical Field

[0001] This application relates to the field of industrial process control technology, and in particular to a method and apparatus for controlling the operating state of equipment based on physical constraint inversion operators. Background Technology

[0002] The core states of industrial processes such as heat conduction and fluid transport (such as temperature field, pressure field, and velocity field) are continuously distributed in time and space, which can directly determine product quality and equipment safety. However, due to limitations in sensor installation space, cost, and high temperature and high pressure environment, only a limited number of measurement points can be set up on site. The measurement data generally suffers from problems such as sparseness, asynchrony, easy loss, and noise, making it impossible to fully obtain the entire state of the process. Therefore, state reconstruction technology must be relied upon.

[0003] Currently, there are two main reconstruction schemes: state reconstruction methods based on mechanistic models and data-driven state reconstruction methods. Mechanism-based inversion methods require solving the inverse problem of partial differential equations in each cycle, resulting in high computational costs and difficulty meeting real-time requirements. Furthermore, order reduction introduces additional errors, and there is a lack of error quantification methods. While data-driven methods offer fast online speeds, their accuracy is significantly affected by sample coverage and operating condition deviations. Even with physical constraints, error quantification cannot be guaranteed. If the error estimate is not independently separated from the training samples, the output confidence level will deviate, making it difficult to support safe decision-making.

[0004] The state variables, physical equations, sensor configurations, and operating boundaries of different physical processes (such as heat conduction and fluid transport) vary significantly. Existing technologies simply replace the state estimation model without simultaneously verifying the consistency of observation operators, parameter units, etc., which can easily lead to parameter mismatch. At the same time, the lack of a binding mechanism between control commands, state data, and validity periods can easily lead to the misforwarding of old or mismatched commands, resulting in technical problems that make it difficult to accurately control the operating status of equipment. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method and apparatus for controlling the operating state of a device based on a physical constraint inversion operator, so as to at least solve the technical problem of difficulty in accurately controlling the operating state of a device.

[0006] According to one aspect of this application, a device operation state control method based on a physical constraint inversion operator is provided. The method includes: acquiring the domain parameter set and reconstructed state field coordinates of the target device in a target industrial process, as well as the target sparse measurement data, the target executed control variables, and the target physical condition data of the target device within the current measurement time window; inputting the target sparse measurement data, the target executed control variables, the target physical condition data, and the reconstructed state field coordinates into a target physical constraint inversion operator to obtain the target reconstructed state field; and determining the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set; and using the reduced-order basis matrix to perform dimensionality reduction and compression processing on the target reconstructed state field to obtain a low-dimensional table. The system generates a reduced-order state error upper bound using an error calculation method based on the domain parameter set, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals. For each initial operating boundary margin in the set of operating boundaries within the domain parameter set, the system corrects the margin based on the sensitivity upper bound, the reduced-order state error upper bound, and the low-dimensional representation vector, resulting in multiple operating boundary risk components. These risk components are then aggregated to obtain a risk index. Based on the domain parameter set, target sparse measurement data, and the risk index, the system determines the industrial state data of the target equipment in the current control cycle and binds this industrial state data to the control commands to be forwarded. This allows the system to control the target equipment to operate normally in the target industrial process using the forwarded control commands.

[0007] Optionally, the target physical constraint inversion operator includes: acquiring an initial physical constraint inversion operator and inversion training samples, and constructing a training measurement feature vector based on the inversion training samples, wherein the initial physical constraint inversion operator includes at least a backbone network and branch networks; acquiring observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss, and constructing a training loss function based on the observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss; inputting the training measurement feature vector into the branch network to obtain a first feature vector, and inputting the spatial coordinates of the target location in the target device and the target time into the backbone network to obtain a second feature vector, so as to determine the predicted reconstructed state field based on the first feature vector and the second feature vector; and training the initial physical constraint inversion operator using the training loss function and the predicted reconstructed state field to obtain the target physical constraint inversion operator.

[0008] Optionally, based on the target reconstructed state field, target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined, including: determining the observation residual based on the predicted value at the target measurement location in the target reconstructed state field and the measured value at the target measurement location in the target sparse measurement data; obtaining physical parameters and inputting the target reconstructed state field, physical parameters, target executed control variables, and target physical condition data into the physical residual operator to obtain the physical residual; and determining the reduced-order projection residual and sparse interpolation residual based on the target reconstructed state field and the reduced-order basis matrix.

[0009] Optionally, based on the target reconstructed state field and the reduced basis matrix, the reduced-order projection residual and the sparse interpolation residual are determined, including: obtaining the state error weighting matrix, and determining the reduced-order projection residual based on the reduced basis matrix, the transpose matrix corresponding to the reduced basis matrix, the target reconstructed state field, and the state error weighting matrix; obtaining the residual snapshot matrix, performing singular value decomposition on the residual snapshot matrix to obtain the residual basis matrix, and generating a selection matrix based on the residual basis matrix using a matrix column selection algorithm; obtaining the current residual vector, and determining the sparse residual vector based on the residual basis matrix, the transpose matrix corresponding to the selection matrix, and the current residual vector; obtaining the residual error weighting matrix, and determining the sparse interpolation residual based on the current residual vector, the sparse residual vector, and the residual error weighting matrix.

[0010] Optionally, the domain parameter set includes at least: a deterministic set of applicable operating conditions, a lower bound for the combined residual stability constant, observation residual error transformation coefficients, physical residual error transformation coefficients, projection residual error transformation coefficients, interpolation residual error transformation coefficients, an upper bound for physical condition disturbance error, and an upper bound for numerical calculation error. Based on the domain parameter set, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, an error calculation method is used to generate a reduced-order state error upper bound, including: in response to the error calculation method being a deterministic error calculation method, obtaining the current operating condition of the target equipment and determining whether the current operating condition belongs to the deterministic set of applicable operating conditions; in response to the current operating condition belonging to the deterministic set of applicable operating conditions, calculating the deterministic state error upper bound using the following formula: , in, Used to represent the upper bound of the error in a deterministic state; Used to represent the conversion coefficient of observation residual error; Used to represent observation residuals; Used to represent the conversion coefficient of physical residual error; Used to represent physical residuals; Used to denote the lower bound of the combined residual stability constant; Used to represent the conversion coefficient of projection residual error; Used to represent the reduced-order projection residual; Used to represent the conversion coefficient of interpolation residual error; Used to represent sparse interpolation residuals; Used for the upper bound of physical condition disturbance error; Used to represent the upper bound of numerical calculation error; in response to the error calculation method being statistical error calculation method, the upper bound of statistical state error is calculated using the following formula: , in, Used to represent the upper bound of statistical state error; Used to represent the basic error estimation function; Used to represent quantile thresholds; converts the upper bound of deterministic state error or statistical state error into a reduced-order state error upper bound.

[0011] Optionally, using the following formula, for each operational boundary in the set of operational boundaries in the domain parameter set, the initial operational boundary margin is modified based on the upper bound of sensitivity, the upper bound of reduced-order state error, and the low-dimensional representation vector of each operational boundary, to obtain multiple operational boundary risk components, including: , , in, Used to represent the first in the set of running boundaries j The initial running boundary margin corresponding to each running boundary; Used to represent the operating boundary margin after correcting the initial operating boundary margin; Used to indicate the first j The upper bound of sensitivity corresponding to each operational boundary; Used to represent the upper bound of the reduced-order state error; Used to indicate the first j Margin scale corresponding to each operating boundary; Used to indicate the first j Each operational boundary corresponds to an operational boundary risk component; Used to represent a restricted function; when the first j When the running boundary is the upper limit boundary This can be expressed by the following formula: , In the above formula, Used to represent boundary thresholds; Used to represent boundary functions, ,and , Used to represent the transpose of the weight vector. Used to represent bias terms; when the first term... j When each operating boundary is a lower limit boundary This can be expressed by the following formula: , In the above formula, .

[0012] Optionally, the domain parameter set may further include at least: observation residual hard threshold and physical residual hard threshold. Based on the domain parameter set, target sparse measurement data, and risk index, the industrial state data of the target equipment in the current control cycle is determined, including: acquiring the training condition feature vectors of the target equipment under multiple normal training conditions, and the current condition feature vector of the target equipment under the current condition in multiple control cycles, and performing a mean operation on the multiple training condition feature vectors to obtain a feature mean vector; determining the regularized covariance matrix based on the multiple training condition feature vectors and the feature mean vector, and determining the regularized covariance matrix based on the multiple current condition feature vectors, the feature mean vector, and the regularized covariance matrix. Define multiple distribution offset distances; acquire the observation residuals, physical residuals, and the number of effective measurements in the target sparse measurement data under multiple control cycles; and determine the target anomaly indication quantity under multiple control cycles based on multiple observation residuals, multiple physical residuals, multiple effective measurement quantities, and multiple distribution offset distances; determine the failure flag information in response to the target anomaly indication quantity being greater than or equal to the anomaly indication threshold, or satisfying any of the following conditions: the observation residual is greater than the observation residual hard threshold, the physical residual is greater than the physical residual hard threshold, and the number of effective measurements is 0; and determine the industrial status data based on the target anomaly indication quantity, distribution offset distance, failure flag information, and risk index.

[0013] Optionally, the regularized covariance matrix is ​​determined based on multiple training condition feature vectors and feature mean vectors using the following formula, including: , in, Used to represent the regularized covariance matrix; Used to indicate the number of training samples; Used to indicate the target device in the... i A training condition feature vector under normal training conditions; Used to represent the feature mean vector; Used to represent regularization weights; Used to represent the identity matrix; The following formula is used to determine multiple distribution offset distances based on multiple current operating condition feature vectors, feature mean vectors, and regularized covariance matrix, including: , in, Used to indicate the current control cycle t The distribution offset distance below; Used to indicate the current control cycle t The current operating condition feature vector; The matrix used to represent the inverse of the regularized covariance matrix; The domain parameter set also includes at least: observation residual threshold, physical residual threshold, distribution offset threshold, and effective measurement quantity threshold. Based on multiple observation residuals, multiple physical residuals, multiple effective measurement quantities, and multiple distribution offset distances, the target anomaly indication quantity under multiple control periods is determined using the following formula, including: , , , in, Used to indicate the current control cycle t The observation residuals below; Used to represent the observation residual threshold, and the observation residual threshold is less than the observation residual hard threshold; Used to indicate the current control cycle t The observed residual abnormality indicator quantity; This is used to represent a conditional judgment function. When the condition inside the parentheses is true, it outputs 1; otherwise, it outputs 0. Used to indicate the current control cycle t The physical residuals below; Used to represent the physical residual threshold, and the physical residual threshold is less than the physical residual hard threshold; Used to indicate the current control cycle t The physical residual abnormality indicator; Used to represent the distribution offset threshold; Used to indicate the current control cycle t The offset distance is an abnormal indicator; Used to indicate the current control cycle t The number of effective measurements below; Used to indicate the threshold of valid measurement quantity; Used to indicate the current control cycle t The number of valid measurements is abnormal. Used to indicate the current control cycle t The target anomaly indicator quantity.

[0014] Optionally, using industrial status data to control the target equipment to be in normal operation during the target industrial process includes: receiving a control instruction to be forwarded generated by the controller based on the industrial status data, wherein the control instruction to be forwarded carries the instruction validity period; determining whether the current time of generating the industrial status data is within the instruction validity period; if the current time is within the instruction validity period and the risk index is less than the stop control threshold, forwarding the control instruction to be forwarded to the target equipment to control the target equipment to be in normal operation during the target industrial process.

[0015] According to another aspect of this application, a device for controlling the operating state based on a physical constraint inversion operator is provided. The device includes: an acquisition unit, configured to acquire the domain parameter set and reconstructed state field coordinates of the target device in a target industrial process, as well as the target sparse measurement data, the target executed control quantity, and the target physical condition data of the target device within the current measurement time window; a determination unit, configured to input the target sparse measurement data, the target executed control quantity, the target physical condition data, and the reconstructed state field coordinates into the target physical constraint inversion operator to obtain the target reconstructed state field, and, based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and the reduced-order basis matrix in the domain parameter set, determine the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual; and a generation unit, configured to perform dimensionality reduction and compression processing on the target reconstructed state field using the reduced-order basis matrix to obtain... The system generates a low-dimensional representation vector and, based on the domain parameter set, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, uses an error calculation method to generate a reduced-order state error upper bound. A correction unit is used to correct the initial operating boundary margin for each operating boundary in the set of operating boundaries in the domain parameter set, based on the sensitivity upper bound, the reduced-order state error upper bound, and the low-dimensional representation vector of each operating boundary, obtaining multiple operating boundary risk components. These multiple operating boundary risk components are then aggregated to obtain a risk index. A control unit is used to determine the industrial state data of the target equipment in the current control cycle based on the domain parameter set, target sparse measurement data, and the risk index. This industrial state data is then bound to the control commands to be forwarded, so that the target equipment can be controlled to operate normally in the target industrial process through the forwarded control commands.

[0016] According to another aspect of this application, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described device operation state control method based on physical constraint inversion operators.

[0017] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the program, it implements the above-described device operation state control method based on physical constraint inversion operators.

[0018] By means of the above technical solution, this application provides a method and apparatus for controlling the operating state of equipment based on a physical constraint inversion operator. The method includes: firstly, acquiring the domain parameter set of the target equipment in the target industrial process, the reconstructed state field coordinates, and the target sparse measurement data, executed control variables, and target physical condition data within the current measurement window; then, inputting the above data into the target physical constraint inversion operator to obtain the target reconstructed state field; and determining the observation residual and physical residual based on the state field, the target sparse measurement data, and the physical residual operator; subsequently, using a reduced-order basis matrix to reduce the dimensionality of the target reconstructed state field to obtain... A low-dimensional representation vector is used to determine the reduced-order projection residual and the sparse interpolation residual. Combining the domain parameter set, observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual, an upper bound for the reduced-order state error is generated through error calculation. Then, for each initial operating boundary margin of the operating boundary set in the domain parameter set, it is corrected based on the upper bound of the sensitivity of the corresponding operating boundary and the upper bound of the reduced-order state error to obtain the operating boundary risk components and aggregate them into a risk index. Finally, based on the domain parameter set, the target sparse measurement data, and the risk index, the industrial state data of the current control cycle is determined, thereby controlling the target equipment to operate normally in the target industrial process. In other words, in this embodiment, by combining physical constraint inversion, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, a quantifiable upper bound for the reduced-order state error is generated. Furthermore, the operating boundary margin is corrected based on the sensitivity upper bound, and risk components are aggregated to obtain a risk index. This overcomes the problems of unquantifiable state reconstruction errors and lack of dynamic assessment of operating boundary risks in existing technologies. Simultaneously, based on the domain parameter set, sparse measurement data, and risk index, industrial state data is determined, compensating for the issues of misforwarding old or mismatched instructions due to the lack of binding between the control instructions to be forwarded and the industrial state data, as well as the lack of validity period binding. This solves the technical problem of difficulty in accurately controlling the operating state of equipment and achieves the technical effect of accurately controlling the operating state of equipment.

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a device operation state control method based on a physical constraint inversion operator provided in an embodiment of this application is shown. Figure 2 A schematic flowchart of a cross-physical domain industrial control method provided in an embodiment of this application is shown; Figure 3 This illustration shows an offline training diagram of a physical constraint inversion operator provided in an embodiment of this application; Figure 4 This illustration shows a schematic diagram of the construction of a reduced-order basis matrix and sparse interpolation points according to an embodiment of this application; Figure 5 A flowchart illustrating a method for calculating the upper bound of a state error according to an embodiment of this application is shown. Figure 6 This illustration shows a structural diagram of the propagation of state error to operational boundary risk according to an embodiment of this application; Figure 7 This illustration shows a conversion diagram of domain mismatch detection and failure flag conversion provided in an embodiment of this application; Figure 8 A flowchart illustrating a domain parameter switching method provided in an embodiment of this application is shown. Figure 9 A flowchart illustrating a method for binding and verifying status data and control commands according to an embodiment of this application is shown. Figure 10 This illustration shows a control diagram of a risk classification and degradation control method provided in an embodiment of this application; Figure 11 This illustration shows a schematic diagram of the application relationship between heat conduction and fluid transport processes according to an embodiment of this application; Figure 12 This illustration shows a schematic diagram of a device operation status control device based on a physical constraint inversion operator according to an embodiment of this application; Figure 13 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0021] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0022] In this embodiment, Figure 1 This paper illustrates a flowchart of a device operation state control method based on a physical constraint inversion operator, as provided in an embodiment of this application. Figure 1 As shown, the method includes: Step S101: Obtain the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control variables, and target physical condition data of the target device within the current measurement time window.

[0023] In the technical solution provided by step S101 of this application, it is necessary to obtain the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control quantities and target physical condition data of the target device within the current measurement time window.

[0024] Optionally, the target industrial process corresponding to the target equipment is determined, and the domain parameter set corresponding to the target industrial process is loaded. The domain parameter set includes at least the domain identifier, domain parameter version, applicable equipment identifier, applicable operating condition range, state variable definition, spatial coordinate definition, measurement point mapping relationship, observation operator, physical residual operator, physical parameters, control condition field definition, normalization parameter, model parameters of physical constraint inversion operator, reduced-order basis matrix, error upper bound parameter, maximum permissible state error, and operating boundary set.

[0025] Optionally, the observation operator is passed through Representation; physical residual operators are used to... Representation; physical parameters are expressed through Representation; reduced-order basis matrix through This is represented; the maximum permissible state error is expressed through... Representation; running boundary set through To express.

[0026] Optionally, the initial sparse measurement data, initial executed control variables, and initial physical condition data of the target device within the current measurement time window are acquired. Based on the domain parameter set, the initial sparse measurement data, initial executed control variables, and initial physical condition data are normalized to obtain the target sparse measurement data, target executed control variables, and target physical condition data. The initial sparse measurement data can be simply referred to as sparse measurement data. The initial executed control variables can be called the already executed control variables. The initial physical condition data can be called known physical condition data.

[0027] For example, the system acquires sparse measurement data generated by the target device within the current measurement time window, executed control variables, and known physical condition data. The sparse measurement data includes at least sensor identifiers, measured values, sampling times, measurement locations, and quality flags. The known physical condition data includes at least one of source term parameters, initial condition parameters, and boundary condition parameters. Using the reference time of the current control cycle as a benchmark, the sparse measurement data and executed control variables are time-aligned. Missing values ​​or those with invalid quality flags are marked with missing information. Based on the measurement point mapping relationship and normalization parameters in the domain parameter set, the valid measured values, measurement locations, relative sampling times, executed control variables, and known physical condition data are arranged and normalized to obtain the arranged and normalized valid measured values, missing flags, relative sampling times, executed control variables, and known physical condition data.

[0028] Step S102: Input the target sparse measurement data, the target executed control variables, the target physical condition data, and the reconstructed state field coordinates into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, determine the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual.

[0029] In the technical solution provided by step S102 of this application, the target sparse measurement data, target executed control variables, target physical condition data, and reconstructed state field coordinates obtained in the above steps are input into the target physical constraint inversion operator to obtain the target reconstructed state field. The reconstructed state field coordinates can be referred to as the coordinates of the state field to be reconstructed, or the spatial coordinates of the position to be reconstructed. The target reconstructed state field is the reconstructed state field output by the physical constraint inversion operator, which is the estimated state vector, and is obtained through... To express.

[0030] Optionally, the reconstructed state field refers to the estimated state of an industrial process covering the entire space region (such as the temperature distribution and velocity distribution of the entire space) derived from limited sensor data and physical laws. It serves as the basis for subsequent error analysis and safety control decisions.

[0031] Optionally, the target sparse measurement data, the target executed control variables, and the target physical condition data are constructed into an input vector, and then the input vector and the reconstructed state field coordinates are input into the target physical constraint inversion operator to obtain the predicted value, i.e. the target reconstructed state field.

[0032] For example, the valid measurements after permutation and normalization, missing markers, relative sampling time, executed control variables, known physical condition data, and the state field to be reconstructed are input into the physical constraint inversion operator to obtain the target reconstructed state field. .

[0033] Optionally, based on the target reconstructed state field, sparse target measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined. The observation residual is determined by... Representation is performed. Physical residuals are expressed through... Representation is performed. The reduced-order projection residual is obtained through... This is represented. The sparse interpolation residuals are expressed through... To express.

[0034] For example, based on the target reconstructed state field, the observation residual is calculated by the difference between the predicted value and the actual measured value at the effective measurement location; the target reconstructed state field, physical parameters, executed control variables, source term parameters, initial conditions and boundary conditions are input into the physical residual operator corresponding to the target industrial process to obtain the physical residual.

[0035] Optionally, the reduced-order projection residual and sparse interpolation residual can be determined using the reduced-order basis matrix and the target reconstructed state field. For example, the reduced-order basis matrix in the domain parameter set can be used to map the target reconstructed state field to obtain the reduced-order projection residual and sparse interpolation residual.

[0036] Step S103: Using the reduced-order basis matrix, the target reconstructed state field is subjected to dimensionality reduction and compression processing to obtain a low-dimensional representation vector. Based on the domain parameter set, observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual, an upper bound of the reduced-order state error is generated using an error calculation method.

[0037] In the technical solution provided in step S103 of this application, the target reconstructed state field obtained above is subjected to dimensionality reduction and compression processing using the reduced-order basis matrix to obtain a low-dimensional representation vector, so as to determine the reduced-order projection residual and the sparse interpolation residual. Then, based on the domain parameter set, the observation residual, the physical residual, the reduced-order projection residual and the sparse interpolation residual, the upper bound of the reduced-order state error is generated by using the error calculation method.

[0038] Alternatively, the low-dimensional representation vector can be obtained through... This is used to represent the reduced-order state. The upper bound of the reduced-order state error passes through... The error calculation methods include at least: deterministic error bound calculation methods and independently calibrated statistical error bound calculation methods.

[0039] Optionally, the target reconstructed state field can be reduced in dimension and compressed using the reduced-order basis matrix according to the following formula to obtain a low-dimensional representation vector: , in, This is used to represent the transpose of a reduced-order basis matrix. That is, it is derived from the reduced-order basis matrix using the formula above. Project the target reconstructed state field into a reduced-order state.

[0040] For example, based on the domain parameter set, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, a deterministic error bound calculation method or an independently calibrated statistical error bound calculation method is used to generate the upper bound of the state error and the type of the upper bound, and then the upper bound of the state error is converted into a reduced-order state error upper bound.

[0041] Optionally, when using the deterministic error bound calculation method, the corresponding applicable operating condition identifier is output; when using the statistical error bound calculation method, the corresponding confidence level is output; the upper bound of the state error ( ) is converted into a reduced-order state error upper bound ( ), and according to Does it exceed the preset threshold? The validity flag is determined by analyzing the newly generated state data in the current control cycle and whether the applicable conditions for the corresponding error bounds are met. The validity flag is calculated immediately to determine whether the obtained state data and error upper bounds meet the applicable conditions and can be used for subsequent control decisions.

[0042] In this embodiment, a reduced-order basis matrix and sparse interpolation points are constructed for each target industrial process. First, according to the node order, state variable arrangement order, and normalization method of the output state field of the physical constraint inversion operator, the state field samples under multiple reference conditions are vectorized to obtain state snapshot vectors, which are then used to form a state snapshot matrix, namely: , For a target industrial process containing two or more state variables, after dimensionless processing using the normalized parameters corresponding to each state variable, a state snapshot vector is formed according to a preset state variable order. The resulting state snapshot matrix... Performing singular value decomposition yields: , , in, Used to represent a left singular vector matrix; Used to represent diagonal singular value matrices; Used to represent a right singular vector matrix; Used to represent reduced dimension; Used to indicate the first There are 10 singular values, and these singular values ​​are arranged in descending order. Select The former Columns form a reduced-order basis matrix .

[0043] Step S104: For the initial operating boundary margin of each operating boundary in the set of operating boundaries in the domain parameter set, based on the upper bound of sensitivity, the upper bound of reduced state error, and the low-dimensional representation of each operating boundary, the initial operating boundary margin is corrected to obtain multiple operating boundary risk components. The multiple operating boundary risk components are then aggregated to obtain the risk index.

[0044] In the technical solution provided in step S104 of this application, for the initial operating boundary margin of each operating boundary in the set of operating boundaries in the domain parameter set, the initial operating boundary margin is corrected according to the upper bound of sensitivity, the upper bound of reduced-order state error, and the low-dimensional representation vector of each operating boundary, to obtain multiple operating boundary risk components. These multiple operating boundary risk components are then aggregated to obtain a risk index. The initial operating boundary margin can be referred to as the normalized operating boundary margin without considering state error, and is obtained through... The risk index is represented by R.

[0045] For example, for each operating boundary in the set of operating boundaries, the operating boundary margin without considering state errors is calculated based on the reduced-order state. Then, based on the upper bound of sensitivity and the upper bound of reduced-order state error corresponding to the operating boundary, multiple operating boundary margins are contracted to obtain the error-corrected operating boundary margin. The corresponding boundary risk components are calculated based on the error-corrected margins of each operating boundary, and multiple boundary risk components are aggregated to obtain the risk index for the current control cycle.

[0046] In this embodiment of the application, multiple operational boundary risk components are aggregated using the following formula to obtain a risk index, including: , in, Used to indicate the first j Each operational boundary corresponds to an operational boundary risk component; M 1 represents the total number of residual terms participating in the aggregation. In other words, any single term exceeding the boundary in each operating boundary may cause a safety risk to the target device. Therefore, the risk aggregation type in the domain parameter set is pre-set to the maximum risk aggregation, and the risk index is calculated according to the above formula. The risk aggregation type is determined before the domain parameter set is enabled and is not switched during online control. When the risk aggregation type changes, a new version of the domain parameters should be generated and each risk threshold should be redefined.

[0047] Furthermore, the risk index can also be calculated using a weighted average method, specifically using the following formula: , in, To and The corresponding weights, and the weighted average method, are suitable for evaluating overall operational deviations. Alternatively, the risk index can be calculated by maximizing the smoothness, using the following formula: , in, The smoothing coefficient used to represent risk aggregation, and the calculation method that maximizes this smoothness, is suitable for optimization control requiring differentiable risk functions, and requires pre-storage. Simultaneously, the domain parameter set stores the risk threshold corresponding to each risk aggregation type. Different risk aggregation types do not share risk thresholds that have not been redefined; that is, they are configured according to... Output the main risk boundary number ( The boundary identifier, error correction margin, and risk components are written into the state data. When the upper bound of the state error is invalid, the dimension of the boundary function input is inconsistent, or the current reduced state exceeds the applicable state set of the upper bound of sensitivity, the state validity flag is set to invalid.

[0048] It should be noted that this is only a preferred implementation method for calculating the risk index, and no specific limitations are made on the calculation method and process of the risk index.

[0049] Step S105: Based on the domain parameter set, target sparse measurement data and risk index, determine the industrial status data of the target equipment in the current control cycle, and bind the industrial status data with the control instructions to be forwarded, so as to control the target equipment to be in normal operation in the target industrial process through the control instructions to be forwarded.

[0050] In the technical solution provided in step S105 of this application, based on the obtained domain parameter set, target sparse measurement data, and risk index, the industrial status data of the target device in the current control cycle is determined. Then, the industrial status data is bound to the control command to be forwarded, so that the target device can be controlled to be in normal operating condition through the control command to be forwarded. The industrial status data can be simply referred to as status data, and in addition to the data mentioned above, the status data also includes at least a status data number.

[0051] Optionally, a monotonically increasing and non-repeating cycle number is generated for each control cycle of the target device. ), and generate a status data number based on the target device's device identifier, domain identifier, domain parameter version, and cycle number: , in, Used to indicate field concatenation; Used to represent status data numbers. The status data includes at least a status data number. State generation time State validity deadline Device identifier, domain identifier, domain parameter version, demotion status Upper bound of reduced-order state error Risk Index and status validity flags, where, , Used to indicate the longest valid duration of status data.

[0052] Optionally, when the controller generates a control command to be forwarded based on the status data, it includes the status data number, control cycle number, device identifier, domain identifier, domain parameter version, and command generation time. and the duration of the instruction Write the control instruction to be forwarded. The control instruction to be forwarded also includes the target execution device identifier, control variable name, control variable unit, and control variable set value. The timing clock used by the controller is synchronized with the timing clock used to control the target device to be in normal operation according to the control instruction to be forwarded; when a clock deviation is detected that exceeds the maximum allowable clock deviation, the time validity verification result is determined to be unsuccessful.

[0053] Optionally, based on the status data number carried by the control instruction to be forwarded, the corresponding status data record is searched, and it is verified that the record exists and has not been marked as used or expired; the control cycle number, device identifier, domain identifier, and domain parameter version are verified to be consistent; the target execution device identifier is verified to belong to the target device corresponding to the status data; and the control variable name and unit are verified to be consistent with the execution variable definition in the current domain parameter set.

[0054] For example, based on the domain parameter set, target sparse measurement data, and risk index, status data corresponding to the current control cycle is generated. This status data is then bound to the control commands to be forwarded to control the operating status of the target device. The status data includes at least the status data number, status generation time, validity period, device identifier, domain identifier, domain parameter version, degraded status, upper bound of status error, type of upper bound error, confidence level or applicable operating condition identifier, risk index, and status validity flag.

[0055] In the technical solution provided by steps S101 to S105 of this application, the domain parameter set of the target equipment in the target industrial process, the reconstructed state field coordinates, and the target sparse measurement data, executed control variables, and target physical condition data within the current measurement window are first obtained. Then, the above data are input into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on this state field, the target sparse measurement data, and the physical residual operator, the observation residual and physical residual are determined. Subsequently, the target reconstructed state field is dimensionality reduced using a reduced-order basis matrix to obtain a low-dimensional representation vector to determine the reduced-order basis. The projection residual and sparse interpolation residual, combined with the domain parameter set, observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual, generate a reduced-order state error upper bound through error calculation. Then, for each initial operating boundary margin of the operating boundary set in the domain parameter set, it is corrected based on the sensitivity upper bound and the reduced-order state error upper bound of the corresponding operating boundary to obtain the operating boundary risk component and aggregate it into a risk index. Finally, based on the domain parameter set, target sparse measurement data, and risk index, the industrial state data of the current control cycle is determined, thereby controlling the target equipment to operate normally in the target industrial process. In other words, in this embodiment, by combining physical constraint inversion, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, a quantifiable upper bound for the reduced-order state error is generated. Furthermore, the operating boundary margin is corrected based on the sensitivity upper bound, and risk components are aggregated to obtain a risk index. This overcomes the problems of unquantifiable state reconstruction errors and lack of dynamic assessment of operating boundary risks in existing technologies. Simultaneously, based on the domain parameter set, sparse measurement data, and risk index, industrial state data is determined, compensating for the issues of misforwarding old or mismatched instructions due to the lack of binding between the control instructions to be forwarded and the industrial state data, as well as the lack of validity period binding. This solves the technical problem of difficulty in accurately controlling the operating state of equipment and achieves the technical effect of accurately controlling the operating state of equipment.

[0056] The method described in this embodiment will be further described below.

[0057] As an optional embodiment, the target physical constraint inversion operator includes: acquiring an initial physical constraint inversion operator and inversion training samples, and constructing a training measurement feature vector based on the inversion training samples, wherein the initial physical constraint inversion operator includes at least a backbone network and a branch network; acquiring observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss, and constructing a training loss function based on the observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss; inputting the training measurement feature vector into the branch network to obtain a first feature vector, and inputting the spatial coordinates of the target location in the target device and the target time into the backbone network to obtain a second feature vector, so as to determine the predicted reconstructed state field based on the first feature vector and the second feature vector; and training the initial physical constraint inversion operator using the training loss function and the predicted reconstructed state field to obtain the target physical constraint inversion operator.

[0058] In this embodiment, an initial physical constraint inversion operator and inversion training samples are first obtained. Based on the inversion training samples, a training measurement feature vector is constructed. Then, observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss are obtained. Based on the observation consistency loss, control equation residual loss, initial boundary condition loss, and state field difference loss, a training loss function is constructed. The training measurement feature vector is then input into the branch network of the initial physical constraint inversion operator to obtain a first feature vector. The spatial coordinates of the target location in the target device and the target time are input into the backbone network to obtain a second feature vector. Based on the first and second feature vectors obtained above, the predicted reconstructed state field is determined. Finally, the initial physical constraint inversion operator is trained using the training loss function and the predicted reconstructed state field to obtain a pre-trained physical constraint inversion operator, i.e., the target physical constraint inversion operator.

[0059] Optionally, the training measurement feature vector can be achieved through... The initial physical constraint inversion operator is an untrained physical constraint inversion operator. The spatial coordinates of the target location can be referred to as the spatial coordinates of the location to be reconstructed, and are represented by... The target time is represented by t. The predicted reconstructed state field can be called the predicted state field, and is represented by t. To express.

[0060] Optionally, a training measurement feature vector is constructed based on the inverted training samples using the following formula, including: , in, Used to indicate the first There are several state variables, and a training measurement feature vector can be constructed using multiple state variables; The number of feature terms; Used to represent the feature quantity index; Used to indicate the branch network for the first The output of the nth state variable Characteristic coefficients of measurement and control conditions; Used to indicate that the backbone network is for the first The output of the nth state variable Spatiotemporal coordinate characteristic coefficients; Used to indicate the first The bias parameter corresponding to the nth state variable. That is, for the nth state variable... With each state variable, the predicted state value at the location to be reconstructed can be generated according to the above formula, so as to determine the predicted state field.

[0061] Optionally, a training loss function is constructed based on the observation consistency loss, the governing equation residual loss, the initial boundary condition loss, and the state field difference loss using the following formula, including: , in, Used to represent the training loss function; All are loss weights greater than 0; Used to represent the loss of observation consistency; It is used to represent the residual loss of the control equation, which is obtained by substituting the predicted state field, physical parameters, executed control variables, source term parameters and known physical conditions into the physical residual operator; Used to represent the loss under initial boundary conditions; This is used to represent the state field difference loss between the predicted state field and the reference state field; that is, the training loss function of the physical constraint inversion operator can be constructed using the above formula.

[0062] Furthermore, the observation consistency loss is calculated using the following formula: , in, Used to indicate the quantity of observation data; Used to indicate the first i The mask of each observation data, and To be related to the measurement vector Effective measurement mask vector of the same dimension; Used to indicate the first i The observation operator for each observation data point. The residual loss of the governing equation is calculated using the following formula: , in, Used to represent the number of physical constraints. Used to represent physical configuration points; Used to represent physical parameters; For control variables that have already been executed, The known physical conditions consist of source term parameters, initial condition parameters, and boundary condition parameters. The time and spatial derivatives required to calculate the physical residuals are obtained through automatic differentiation or a preset numerical difference method.

[0063] For example, the training process for each target industrial process using the physical constraint inversion operator is as follows: Inversion training samples are obtained. Each inversion training sample includes at least the reference state field of the target industrial process, sparse measurement values ​​obtained from preset measurement locations in the reference state field, measurement locations, sampling times, quality flags, executed control variables, physical parameters, source term parameters, initial condition parameters, and boundary condition parameters. Sparse measurement values ​​are arranged according to the same measurement point mapping relationship as in the online control phase. Valid measurement markers are generated based on the quality flags. Preset placeholder values ​​are filled into missing measurement fields. The relative sampling times of each sparse measurement value and the executed control variable are calculated based on the reference time corresponding to the reference state field.

[0064] Furthermore, based on the normalization parameters of the corresponding domain parameter set, the sparse measurement values, measurement locations, relative sampling times, executed control variables, physical parameters, source term parameters, initial condition parameters, and boundary condition parameters are normalized respectively. The normalized sparse measurement values, valid measurement markers, quality indicators, relative sampling times, executed control variables, physical parameters, source term parameters, initial condition parameters, and boundary condition parameters are then combined in a fixed field order to form a training measurement feature vector. Establish a physical constraint inversion operator that includes a branch network, a backbone network, and output heads corresponding to each state variable; and train the measurement feature vector. Input the branch network and input the spatial coordinates of the location to be reconstructed. and target time Input the backbone network and generate the predicted state field through the output headers corresponding to each state variable. The inversion training samples are divided into non-overlapping model training samples and model validation samples. The model parameters of the physical constraint inversion operator are updated using the model training samples, while the loss weights, number of feature terms, and training stopping time are selected using the model validation samples. After training, the model parameters of the physical constraint inversion operator are fixed, and the model parameters, input field order, control condition field definition, state variable output order, input-output dimension, normalization parameters, and model version are written into the corresponding domain parameter set. During the online control phase, the order of measurement fields, control condition fields, normalization parameters, state variable output order, and input-output dimension of the input physical constraint inversion operator remain consistent with those during the training phase. Error estimation training samples and independent calibration samples are not used to update the model parameters of the physical constraint inversion operator. When the model parameters, input fields, control condition fields, normalization parameters, state variable order, or input-output dimension change, the original error estimation model, calibration quantiles, and statistical confidence levels are invalidated, and error estimation training and independent calibration are re-executed.

[0065] As an optional implementation, based on the target reconstructed state field, target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined, including: determining the observation residual based on the predicted value at the target measurement location in the target reconstructed state field and the measured value at the target measurement location in the target sparse measurement data; obtaining physical parameters and inputting the target reconstructed state field, physical parameters, target executed control variables, and target physical condition data into the physical residual operator to obtain the physical residual; and determining the reduced-order projection residual and sparse interpolation residual based on the target reconstructed state field and the reduced-order basis matrix.

[0066] In this embodiment, the observation residual is first determined based on the predicted value at the target measurement location in the target reconstructed state field and the measured value at the target measurement location in the target sparse measurement data. Then, the physical parameters are obtained, and the target reconstructed state field, physical parameters, target executed control variables, and target physical condition data are input into the physical residual operator to obtain the physical residual. Finally, the reduced-order projection residual and sparse interpolation residual can be determined based on the target reconstructed state field and the reduced-order basis matrix.

[0067] Optionally, the observation residual is determined based on the predicted value at the target measurement location in the target reconstruction state field and the measured value at the target measurement location in the target sparse measurement data, using the following formula: , in, Used to represent observation residuals, To represent the weighted norm, it should be noted that the vector norms here all adopt the Euclidean norm, that is, the weighted norm is defined as: ; Used to represent the observation residual weighting matrix, which is used to adjust the error weights of different observation data to achieve weighted quantization of the observation residuals; M Used to represent a valid measurement mask vector with the same dimension as the measurement vector y, its elements are either 0 or 1; when an element is 1, it indicates that the corresponding measurement data is valid, and when an element is 0, it indicates that the corresponding measurement is missing or the quality is invalid. Used to represent the predicted value at the target measurement location in the target reconstructed state field, i.e., the observation operator's relationship to the reconstructed state. The mapping result; Used to indicate element-wise multiplication; y The measurement value at the target measurement location in the sparse measurement data is used to represent the measurement vector (actual observation data) arranged according to the mapping relationship of measurement points. In other words, the observation residual is calculated according to the above formula based on the difference between the predicted value and the sparse measurement value at the effective measurement location of the reconstructed state field. .

[0068] Optionally, the target reconstructed state field, physical parameters, target executed control variables, and target physical condition data are input into the physical residual operator using the following formula to obtain the physical residual, which includes: , in, Used to represent physical residuals; Used to represent the weighted matrix of physical residuals; Used to represent physical parameters; Used to indicate control variables that have been executed, i.e., target control variables that have been executed; This is used to represent known physical condition data, i.e., target physical condition data. In other words, it reconstructs the target's state field and physical parameters. Control variables that have already been executed and known physical condition data Input physical residual operator Calculate the physical residuals as described above. .

[0069] As an optional implementation, the reduced-order projection residual and sparse interpolation residual are determined based on the target reconstructed state field and the reduced-order basis matrix, including: obtaining a state error weighting matrix, and determining the reduced-order projection residual based on the reduced-order basis matrix, the transpose matrix corresponding to the reduced-order basis matrix, the target reconstructed state field, and the state error weighting matrix; obtaining a residual snapshot matrix, performing singular value decomposition on the residual snapshot matrix to obtain a residual basis matrix, and generating a selection matrix based on the residual basis matrix using a matrix column selection algorithm; obtaining the current residual vector, and determining the sparse residual vector based on the residual basis matrix, the transpose matrix corresponding to the selection matrix, and the current residual vector; obtaining a residual error weighting matrix, and determining the sparse interpolation residual based on the current residual vector, the sparse residual vector, and the residual error weighting matrix.

[0070] In this embodiment, the state error weighting matrix is ​​first obtained, and the reduced-order projection residual is determined based on the reduced-order basis matrix, the transpose matrix corresponding to the reduced-order basis matrix, the target reconstructed state field, and the state error weighting matrix. Then, singular value decomposition is performed on the obtained residual snapshot matrix to obtain the residual basis matrix. Based on the residual basis matrix, a matrix column selection algorithm is used to generate a selection matrix. Then, the current residual vector needs to be obtained, and the sparse residual vector is determined based on the residual basis matrix, the transpose matrix corresponding to the selection matrix, and the current residual vector. Finally, the sparse interpolation residual is determined based on the current residual vector, the sparse residual vector, and the obtained residual error weighting matrix.

[0071] Optionally, the reduced-order projection residual is determined using the following formula, based on the reduced-order basis matrix, the transpose matrix corresponding to the reduced-order basis matrix, the target reconstructed state field, and the state error weighting matrix, including: , in, Used to represent the reduced-order projection residual; Used to represent the target reconstructed state field; Used to represent reduced-order basis matrices; Used to represent the transpose of the reduced-order basis matrix; Used to represent the state error weighting matrix.

[0072] Optionally, the residual snapshot matrix is ​​represented by G. The residual basis matrix is ​​represented by... Representation. Selection matrix via... The matrix column selection algorithm can be a discrete empirical interpolation algorithm or a decomposition algorithm with column pivoting (QR decomposition method with column pivoting).

[0073] For example, the physical residuals, nonlinear terms, or source terms of each reference condition are vectorized to obtain residual snapshot vectors and form residual snapshot matrices. And then Perform singular value decomposition and select the first [value] according to the preset residual cumulative energy ratio. The left singular vectors form the residual basis matrix. Then, based on the residual basis matrix Discrete empirical interpolation or QR decomposition with column pivoting can be used to select... 1 sparse interpolation position, and generate a selection matrix. ,make Reversible.

[0074] Optionally, the sparse residual vector is determined using the following formula, based on the residual basis matrix, the transpose matrix corresponding to the selection matrix, and the current residual vector, including: , in, Used to represent the physical residual approximation vector obtained by sparse interpolation, i.e., the sparse residual vector, also known as the sparse interpolation approximation; This is used to represent the current residual vector, also known as the current physical residual vector. In other words, for the current residual vector... The sparse interpolation approximation value is calculated using the above formula.

[0075] Furthermore, the reconstructed state field of the current cycle Physical parameters Executed control quantities and known physical conditions Input physical residual operator This yields the current residual vector arranged in the predetermined order of residual components, as shown in the following calculation formula: , The residual vector includes at least the residual components of the governing equations and the residual components of the boundary conditions, and selectively includes the residual components of the initial conditions depending on the target industrial process.

[0076] Optionally, the sparse interpolation residuals are determined based on the current residual vector, the sparse residual vector, and the residual error weighting matrix using the following formula, including: , in, Used to represent sparse interpolation residuals; Used to represent the residual error weighting matrix.

[0077] Furthermore, reducing the order of the basis matrix Reduced dimensionality State error weighting matrix residual basis matrix Selection matrix Residual error weighting matrix The order of state variables, the order of nodes, and their corresponding versions are written into the domain parameter set of the target industrial process. When the definition of state variables, the node discretization method, the normalization parameter, the physical residual operator, or the state error weighting matrix changes, the reduced-order basis matrix and the selection matrix are reconstructed, and the original reduced-order basis version is invalidated.

[0078] As an optional implementation, the domain parameter set includes at least: a deterministic applicable operating condition set, a lower bound for the combined residual stability constant, observation residual error conversion coefficients, physical residual error conversion coefficients, projection residual error conversion coefficients, interpolation residual error conversion coefficients, an upper bound for physical condition disturbance error, and an upper bound for numerical calculation error. Based on the domain parameter set, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, an error calculation method is used to generate a reduced-order state error upper bound, including: in response to the error calculation method being a deterministic error calculation method, obtaining the current operating condition of the target equipment and determining whether the current operating condition belongs to the deterministic applicable operating condition set; in response to the current operating condition belonging to the deterministic applicable operating condition set, calculating the deterministic state error upper bound using the following formula: , in, Used to represent the upper bound of the error in a deterministic state; Used to represent the conversion coefficient of observation residual error; Used to represent observation residuals; Used to represent the conversion coefficient of physical residual error; Used to represent physical residuals; Used to denote the lower bound of the combined residual stability constant; Used to represent the conversion coefficient of projection residual error; Used to represent the reduced-order projection residual; Used to represent the conversion coefficient of interpolation residual error; Used to represent sparse interpolation residuals; Used for the upper bound of physical condition disturbance error; Used to represent the upper bound of numerical calculation error; in response to the error calculation method being statistical error calculation method, the upper bound of statistical state error is calculated using the following formula: , in, Used to represent the upper bound of statistical state error; Used to represent the basic error estimation function; Used to represent quantile thresholds; converts the upper bound of deterministic state error or statistical state error into a reduced-order state error upper bound.

[0079] In this embodiment, the error calculation method includes multiple calculation methods. When the error calculation method is a deterministic error calculation method, the current operating condition of the target device is obtained, and it is determined whether the current operating condition belongs to the set of deterministic applicable operating conditions. If the current operating condition belongs to the set of deterministic applicable operating conditions, the upper bound of the deterministic state error is calculated using the following formula: , Furthermore, when the error calculation method is the statistical error calculation method, the upper bound of the statistical state error can be calculated using the following formula: , Furthermore, the aforementioned deterministic or statistical state error upper bounds are transformed into reduced-order state error upper bounds. Specifically, valid deterministic or statistical state error upper bounds are uniformly denoted as the state error upper bound, and the state error upper bound is determined by... This is represented. The upper bound of the reduced-order state error passes through... This is represented. The set of deterministic applicable operating conditions is expressed through... The representation should include, at a minimum, the ranges of state variables, physical parameters, executed control variables, source parameters, initial conditions, and boundary conditions.

[0080] For example, upper bounds on state errors include deterministic upper bounds on state errors and statistical upper bounds on state errors. Each domain parameter set pre-records the types of error upper bounds that can be used, their corresponding applicable conditions, and the maximum permissible state error. The applicable operating condition identifier or confidence level for the corresponding error upper bound is output only when the corresponding applicable conditions are met. For deterministic state error upper bounds, a set of deterministic applicable operating conditions is pre-recorded in the domain parameter set. Lower bound of combined residual stability constant Observation residual error conversion factor Physical residual error conversion factor Projection residual error conversion factor Interpolation residual error conversion coefficient Upper bound of physical condition disturbance error Upper bound of numerical calculation error The above data satisfies the following conditions: .

[0081] Furthermore, the combined residual mapping consisting of the observation operator and the physical residual operator is denoted as... Within the set of deterministic applicable operating conditions, the lower bound of the stability constant is determined based on the mandatory constant, minimum eigenvalue, minimum singular value, or a verified lower bound of the interval of the combined residual mapping. For nonlinear combined residual mappings, based on their linearization operator ( )Sure , making , in, Used to represent the infimum, that is, the minimum value in a given set; Used to represent the minimum singular value of a matrix; The weight matrix used to represent physical constraints; Jacobian matrix used to represent physical constraints; The weight matrix is ​​used to represent the state variables. That is, the minimum singular value obtained from only a finite number of samples is not directly used as a lower bound for the deterministic stability constant unless the lower bound is proven through interval calculations, energy estimation, or verified eigenvalue calculations. Upper bound for physical condition perturbation error. It includes at least measurement noise, physical parameter deviations, deviations in the recorded control variables, source parameter deviations, initial condition deviations, and the upper bound of the error formed after the boundary condition deviations propagate to the state error scale. When the corresponding applicable conditions are met, the deterministic state error upper bound is calculated according to the following formula: , Simultaneously, it outputs the deterministic error boundary type and applicable operating condition identifier.

[0082] For another example, regarding the upper bound of statistical state error, after fixing the model parameters of the physical constraint inversion operator, a training sample set for error estimation is prepared ( ) and independent calibration sample set ( The error estimation training sample set, the independent calibration sample set, and the physical constraint inversion operator model training sample set are mutually exclusive; the independent calibration samples do not participate in the parameter updates of the physical constraint inversion operator or the error estimation model. For the first... Each sample undergoes the same measurement preprocessing, control condition processing, state reconstruction, and residual calculation as the online phase, resulting in... , in, Used to indicate the first The observation residuals of each sample; Used to indicate the first The physical residuals of each sample; Used to indicate the first The reduced-order projected residuals of each sample; Used to indicate the first The sparse interpolation residuals of each sample. It should be noted that, using the above formula, ... For input, Train a base error estimation function with non-negative output for supervised values. After training, the parameters of the error estimation model are fixed. A calibration score is calculated for each sample in the independent calibration sample set. , in, Used to indicate the first The calibration score for each sample. Based on the formula above, multiple calibration scores can be obtained, and the preset value ( The calibration scores are arranged in ascending order. For those that meet the requirements... significance level ,Sure , in, Used to represent the current error feature vector. The current error feature vector is obtained during the online phase. Then, calculate the upper bound of the statistical state error according to the following formula: , If the current operating condition is within the calibration validity range covered by the independent calibration samples, the current error characteristic does not exceed the preset distribution offset threshold, and the current sample and the independent calibration samples meet the commutativity condition, the error upper bound type is marked as statistical error bound, and the confidence level is output. .

[0083] Optionally, when the corresponding applicable conditions for the upper bound of the deterministic state error or the upper bound of the statistical state error are not met, the corresponding applicable operating condition identifier or confidence level is not output; when the applicable conditions for both upper bounds of the state error are not met, the state validity flag is set to invalid.

[0084] Furthermore, if the current operating condition exceeds the effective range of statistical calibration, the error characteristics exhibit a distribution shift exceeding a threshold, or any of the following changes occur: physical constraint inversion operator, error estimation model, input field, control condition field, normalization parameter, or residual definition, the original statistical confidence level is invalidated. When the conditions for the deterministic error bound are simultaneously met, the system switches to the deterministic state error upper bound; when neither the conditions for the deterministic error bound nor the statistical error bound are met, the state validity flag is invalidated. For valid deterministic state error upper bounds or statistical state error upper bounds, they are uniformly denoted as the state error upper bound. Based on the reduced-order basis matrix Weighted matrix of state error The error propagation coefficient is predetermined using the following formula: , in, Used to represent the gain factor; sup represents the supremum, taking the maximum value for all non-zero vectors a to ensure coverage of all possible perturbation directions; a represents the non-zero state perturbation vector; when When this happens, the status validity flag is set to invalid.

[0085] As an optional implementation, the initial operating boundary margin of each operating boundary in the set of operating boundaries in the domain parameter set is corrected based on the upper bound of sensitivity, the upper bound of reduced-order state error, and the low-dimensional representation vector of each operating boundary using the following formula, resulting in multiple operating boundary risk components, including: , , in, Used to represent the first in the set of running boundaries j The initial running boundary margin corresponding to each running boundary; Used to represent the operating boundary margin after correcting the initial operating boundary margin; Used to indicate the first j The upper bound of sensitivity corresponding to each operational boundary; Used to represent the upper bound of the reduced-order state error; Used to indicate the first j Margin scale corresponding to each operating boundary; Used to indicate the first j Each operational boundary corresponds to an operational boundary risk component; Used to represent restrictive functions, such as vector truncation operations. Indicates the vector Each component is restricted to its corresponding lower limit. and upper limit Between; when the j When the running boundary is the upper limit boundary This can be expressed by the following formula: , In the above formula, Used to represent boundary thresholds; Used to represent boundary functions, ,and , Used to represent the transpose of the weight vector. Used to represent bias terms; when the first term... j When each operating boundary is a lower limit boundary This can be expressed by the following formula: , In the above formula, .

[0086] In this embodiment, the set of running boundaries includes Each runtime boundary record must include at least a boundary identifier, a boundary type, and a boundary function. Boundary threshold Positive margin scale Upper limit of sensitivity Sum of positive risk weights Among them, the positive margin scale can be called the margin scale; the boundary function is used to map the reduced-order state to at least one monitored process variable among the highest temperature, maximum temperature gradient, pressure, flow rate, velocity, or concentration. When the first... When the operating boundary is the upper limit boundary, its safety condition is: and in accordance with: , Calculate the normalized operational boundary margin; when it is the lower boundary, the safety condition is: and in accordance with: , Calculate the normalized operating boundary margin. Within the set of applicable states corresponding to the operating boundary ( Within ), the boundary function satisfies: , in, Used to indicate the current estimated price reduction status; This represents the error between the actual reduced-order state and the estimated value; This represents the upper bound of the sensitivity of the boundary function to changes in the reduced-order state. For a differentiable boundary function, the upper bound of sensitivity is determined according to the following formula: , in, This is used to represent the supremum operator, which finds the minimum upper bound of the function gradient norm within a given region. For non-differentiable boundary functions, the sensitivity upper bound is determined through interval calculation, piecewise linear upper bound, or conservative envelope calculation. The reduced-order state error upper bound is also considered. The error-corrected operating boundary margin is obtained using the following formula: , when When this occurs, it indicates that the state error range has reached or exceeded the specified threshold. Even if the process variables estimated based on the reduced-order state points have not yet crossed the boundary, the operating boundary is still considered to have a risk of exceeding the boundary. Multiple operating boundary risk components are calculated based on the error-corrected operating boundary margin: , when hour, ;when hour, It increases as the safety margin decreases; when hour, .

[0087] As an optional implementation, the domain parameter set may at least include: observation residual hard threshold and physical residual hard threshold. Based on the domain parameter set, target sparse measurement data, and risk index, the industrial state data of the target equipment in the current control cycle is determined, including: acquiring training condition feature vectors of the target equipment under multiple normal training conditions, and current condition feature vectors of the target equipment under the current condition in multiple control cycles; and performing a mean operation on the multiple training condition feature vectors to obtain a feature mean vector; determining a regularized covariance matrix based on the multiple training condition feature vectors and the feature mean vector; and determining a regularized covariance matrix based on the multiple current condition feature vectors, the feature mean vector, and the regularized covariance matrix. The difference matrix is ​​used to determine multiple distribution offset distances; the observation residuals, physical residuals, and the number of effective measurements in the target sparse measurement data are obtained under multiple control cycles, and the target anomaly indication is determined under multiple control cycles based on multiple observation residuals, multiple physical residuals, multiple effective measurement numbers, and multiple distribution offset distances; in response to the target anomaly indication being greater than or equal to the anomaly indication threshold, or satisfying any of the following conditions: the observation residual is greater than the observation residual hard threshold, the physical residual is greater than the physical residual hard threshold, and the number of effective measurements is 0, the failure flag information is determined; based on the target anomaly indication, distribution offset distance, failure flag information, and risk index, the industrial status data is determined.

[0088] In this embodiment, the training condition feature vectors of the target device under multiple normal training conditions and the current condition feature vector of the target device under the current condition under multiple control cycles are first obtained. The mean vector of the multiple training condition feature vectors is then obtained by averaging the multiple training condition feature vectors. Then, the regularized covariance matrix is ​​determined based on the multiple training condition feature vectors and the feature mean vector. Based on the multiple current condition feature vectors, the feature mean vector, and the regularized covariance matrix, multiple distribution offset distances are determined. Then, the observation residuals, physical residuals, and the number of effective measurements in the target sparse measurement data under multiple control cycles are obtained. Based on the multiple observation residuals, multiple physical residuals, multiple effective measurement numbers, and multiple distribution offset distances, the target anomaly indications under multiple control cycles are determined. If the multiple anomaly indications are all greater than the anomaly indication threshold, or if any of the following conditions are met: the observation residual is greater than the observation residual hard threshold, the physical residual is greater than the physical residual hard threshold, and the number of effective measurements is 0, the failure flag information is determined. Finally, the industrial status data can be determined based on the multiple anomaly indications, distribution offset distances, failure flag information, and risk index. The domain parameter set includes at least: observation residual hard threshold and physical residual hard threshold.

[0089] Optionally, the observation residual hard threshold is obtained by... This is represented. The physical residual hard threshold is achieved through... Representation is performed. The training condition feature vector is obtained through... Representation is performed. The current operating condition feature vector is represented through... Representation is performed. The feature mean vector is represented by... This is represented. The distribution offset distance is expressed through... This can be represented. The target anomaly indicator can be expressed through... This is used for representation. The anomaly indication threshold can be a preset value, for example, 2. The number of valid measurements is determined by... To express.

[0090] Optionally, the number of valid measurements is obtained as follows: after acquiring sparse measurement data of the target device within the current measurement time window, the number of sensors whose quality indicators (or quality status) are determined to be valid is counted. Specifically, for each measurement record, if its "quality indicator" indicates that the measurement value is reliable and valid (e.g., the sensor is working normally, the data is within the range, no abnormality is detected, etc.), then the measurement record will be included in the "number of valid measurements" count; if the quality indicator indicates that the measurement is invalid (e.g., sensor failure, missing data, abnormal measurement value, etc.), then it will not be included in the count.

[0091] Optionally, the failure flag information, which may be simply referred to as the failure flag, is a system-level latched flag that persists across cycles. It indicates that the system has entered a severe or persistent abnormal state, requiring the triggering of degradation control. Even if the state is temporarily valid in the next cycle, the failure flag remains at 1. Only newly generated state data continuously for a preset period (…) It will only recover to 0 when the condition is effective for 10 cycles, the risk is below the recovery threshold, and there is no hard threshold exceeding the limit.

[0092] Optionally, when all target anomaly indications are greater than the anomaly indication threshold, or when any of the following conditions are met: the observed residual is greater than the observed residual hard threshold, the physical residual is greater than the physical residual hard threshold, and the number of effective measurements is 0, the failure flag information can be determined to be 1, where the failure flag information can be simply referred to as the failure flag. For example, when ,make And continuously reach During each control cycle, the failure flag is set to 1; otherwise, it is... Set to 0; or, when , and If any of the conditions in the code are met, the failure flag is set to 1 directly.

[0093] As an optional implementation method, the regularized covariance matrix is ​​determined based on multiple training condition feature vectors and feature mean vectors using the following formula, including: , in, Used to represent the regularized covariance matrix; Used to indicate the number of training samples; Used to indicate the target device in the... i A training condition feature vector under normal training conditions; Used to represent the feature mean vector; Used to represent regularization weights; Used to represent the identity matrix; multiple distribution offset distances are determined based on multiple current operating condition feature vectors, feature mean vectors, and regularized covariance matrix using the following formulas, including: , in, Used to indicate the current control cycle t The distribution offset distance below; Used to indicate the current control cycle t The current operating condition feature vector; The inverse matrix used to represent the regularized covariance matrix; the domain parameter set also includes at least: observation residual threshold, physical residual threshold, distribution offset threshold, and effective measurement quantity threshold. Based on multiple observation residuals, multiple physical residuals, multiple effective measurement quantities, and multiple distribution offset distances, the target anomaly indication quantity under multiple control periods is determined using the following formula, including: , , , in, Used to indicate the current control cycle t The observation residuals below; Used to represent the observation residual threshold, and the observation residual threshold is less than the observation residual hard threshold; Used to indicate the current control cycle t The observed residual abnormality indicator quantity; This is used to represent a conditional judgment function. When the condition inside the parentheses is true, it outputs 1; otherwise, it outputs 0. Used to indicate the current control cycle t The physical residuals below; Used to represent the physical residual threshold, and the physical residual threshold is less than the physical residual hard threshold; Used to indicate the current control cycle t The physical residual abnormality indicator; Used to represent the distribution offset threshold; Used to indicate the current control cycle t The offset distance is an abnormal indicator; Used to indicate the current control cycle t The number of effective measurements below; Used to indicate the threshold of valid measurement quantity; Used to indicate the current control cycle t The number of valid measurements is abnormal. Used to indicate the current control cycle t The target anomaly indicator quantity.

[0094] In this embodiment of the application, the domain parameter set also includes an observation residual threshold. Physical residual threshold Distribution offset threshold Minimum number of effective measurements Number of failure confirmation cycles ( ), number of recovery confirmation cycles ( ), observation residual hard threshold and physical residual hard threshold ,in, ,and .

[0095] Optionally, for each target industrial process, operating condition characteristic statistics are constructed using training samples within the normal operating range. Following the same field order and normalization method as the online phase, the normalized measurements, missing data markers, quality indicators, relative sampling time, executed control variables, physical parameters, source term parameters, and boundary condition parameters of each training sample are combined into a fixed-dimensional training operating condition feature vector. The regularized covariance matrix is ​​calculated using the following formula: , , In the current control cycle Following the same field order and normalization method as in the training phase, the current measurement value, missing data marker, quality flag, relative sampling time, executed control variables, physical parameters, source term parameters, and boundary condition parameters are combined into a current operating condition feature vector. And calculate the distribution offset distance: , Then count the number of measurement data that have valid quality marks and passed preprocessing verification, and the number of measurement data that pass... The abnormal indication quantities are represented using the following formulas: , , The target anomaly indicator is obtained by summing the four indicators obtained above, i.e., the calculation formula is: .

[0096] As an optional implementation, industrial status data is bound to a control command to be forwarded, so as to control the target equipment to be in normal operation in the target industrial process through the control command to be forwarded. This includes: receiving a control command to be forwarded generated by the controller based on the industrial status data, wherein the control command to be forwarded carries the validity period of the command; determining whether the current time of generating the industrial status data is within the validity period of the command; if the current time is within the validity period of the command and the risk index is less than the stop control threshold, forwarding the control command to be forwarded to the target equipment to control the target equipment to be in normal operation in the target industrial process.

[0097] In this embodiment, after obtaining industrial status data, the system provides this data to the controller and receives a control command to be forwarded generated by the controller based on the industrial status data. It then determines whether the current time of generating the industrial status data is within the validity period of the control command to be forwarded. If the current time is within the validity period and the risk index is less than the stop control threshold, the control command to be forwarded is forwarded to the target device to control the target device to operate normally in the target industrial process. The stop control threshold can be determined by... The control instruction to be forwarded is referred to as the control instruction to be executed. The control instruction to be forwarded carries its associated status data number, device identifier, domain identifier, domain parameter version, instruction generation time, and instruction validity period.

[0098] Optionally, the system verifies whether the status data number, device identifier, domain identifier, and domain parameter version associated with the control instruction to be forwarded are consistent with the current status data, and verifies whether the current time is within the validity period of the status data and the control instruction to be forwarded. When all verifications pass, the status validity flag is valid, and the risk index is lower than the stop control threshold, the control instruction to be forwarded is forwarded to the target device; when any verification fails, the status validity flag is invalid, or the risk index reaches the stop control threshold, the control instruction to be forwarded is rejected, and downgraded control is implemented according to the safety action type corresponding to the target device.

[0099] In this embodiment, the anomaly indication is combined into an anomaly cause code, and the anomaly cause code, the number of anomaly conditions, the consecutive anomaly cycle count, the distribution offset distance, and the failure flag are written into the status data. When the failure flag is 1, the pending control command generated based on the current reconfiguration state is rejected, and degraded control is implemented according to the safety action type corresponding to the target device. After the failure flag is set to 1, only... The validity status flag is valid and there is no hard threshold exceeding the limit and the number of valid measurements is 0, and the validity period is continuously reached. The failure flag will be reset to 0 only after one control cycle.

[0100] Furthermore, set the continuous recovery cycle count value ( After the failure flag is set to 1, only when When the status validity flag is valid and there is no hard threshold exceeding the limit, and the number of valid measurements is 0, let... Otherwise Set to 0. When When the failure flag is reset to 0, the consecutive abnormal cycle count and consecutive recovery cycle count are cleared. Different target industrial processes have their own corresponding operating condition statistics and thresholds, and do not share unverified domain mismatch parameters.

[0101] For example, in the temperature control scenario of a chemical reactor, the controller generates a control command to be forwarded, "increase heating power by 10%", based on reconstructed industrial state data (such as the temperature and pressure fields inside the reactor). This command carries a validity period of 5 minutes (calculated from the time of generation). After receiving the command, the system first determines whether the current time of generating the industrial state data (e.g., 3 minutes after generation) is within the validity period of the command. At this time, 3 minutes < 5 minutes, satisfying the time condition. Next, the risk index is calculated to be 0.3 and compared with the stop control threshold of 0.5. At this time, 0.3 < 0.5, satisfying the risk condition. Therefore, the system forwards the "increase heating power by 10%" command to the reactor, maintaining the temperature inside the reactor within the normal operating range of 80 degrees Celsius (°C), avoiding product quality problems or equipment safety hazards caused by temperature deviations. If the current time exceeds 5 minutes (e.g., 6 minutes after generation) or the risk index reaches 0.6 (exceeding the threshold), the system refuses to forward the command and triggers degraded control (e.g., maintaining the current power or implementing a safety setpoint transition) to ensure equipment safety.

[0102] Optionally, before executing step S101, a set of first domain parameters corresponding to the first physical process is pre-stored. and the second domain parameter set corresponding to the second physical process. The first and second domain parameter sets use the same field names and field order, but the specific parameter values ​​of each field are determined according to the corresponding physical process. Each domain parameter set includes at least the domain identifier, domain parameter version, applicable equipment identifier, physical process type, applicable operating condition range, state variable definition and order, state variable unit, spatial coordinate definition, measurement point mapping relationship, observation operator, physical residual operator, normalization parameter, inversion operator model parameter, reduced-order basis matrix, error upper bound parameter, operating boundary set, risk threshold, and safety action parameter.

[0103] Optionally, the first physical process and the second physical process have different physical residual operators, and they differ in at least one of the following: state variable definition, spatial coordinate dimension, measurement point mapping relationship, observation operator, reduced basis matrix, and running boundary set. Observation operators, physical residual operators, normalization parameters, inversion operator model parameters, reduced basis matrix, error upper bound parameters, and running boundary sets belonging to the same domain parameter set are all bound to the same domain identifier and domain parameter version. It is prohibited to combine parameter items with different domain identifiers or different domain parameter versions.

[0104] Optionally, when the target industrial process changes between the first physical process and the second physical process, the domain parameter set corresponding to the changed target industrial process is selected, and the observation operator, physical residual operator, reduced-order basis matrix, error upper bound parameter and operating boundary set in the selected domain parameter set are used to execute steps S101 to S105.

[0105] Optionally, when the target device, physical process type, or domain parameter version changes, the device identifier, current physical process type, device configuration identifier, and domain parameter version to be enabled of the target device are read, and the corresponding domain parameter set is searched based on the above information. If no domain parameter set is found, two or more domain parameter sets that simultaneously meet the conditions are found, or the applicable operating conditions of the domain parameter set do not include the current operating conditions, the domain parameter selection is deemed to have failed, and the domain parameter switching is not performed. When a unique corresponding target domain parameter set is found, the original sensor identifier output by the target device is mapped to the measurement arrangement order required by the target domain observation operator according to the measurement point mapping relationship in the target domain parameter set; unit conversion is performed according to the state variable unit and measurement unit in the target domain parameter set, and coordinate origin, coordinate axis direction, or coordinate scale transformation is performed according to the spatial coordinate definition in the target domain parameter set. For measurement data for which a unique sensor mapping relationship cannot be established or where unit and coordinate transformation cannot be completed, the corresponding measurement data is marked as invalid.

[0106] Optionally, after completing the measurement mapping, verify whether the dimension of the mapped measurement vector is equal to the input dimension of the target domain observation operator; verify whether the number, order, and output dimension of the inversion operator's state variables are consistent with the definition of the target domain state variables; verify whether the output dimension of the inversion operator is equal to the number of rows in the reduced-order basis matrix; verify whether the number of columns in the reduced-order basis matrix is ​​equal to the reduced-order state dimension required by the running boundary function; verify whether the state variables, spatial coordinate dimension, physical parameter units, and initial boundary conditions used by the physical residual operator are consistent with the corresponding definitions in the target domain parameter set; verify whether the upper bound parameter of the error, the distribution offset statistic, and the risk threshold are consistent with the target domain parameter version. If any verification fails, the target domain parameter set is prohibited from being enabled, and the target device is kept in a preset safety control state.

[0107] Furthermore, when all verifications pass, and the target device is in a preset security control state, the target domain parameter set is used for continuous execution. Each verification cycle consists of steps S101 to S105, and within each switching verification cycle, the observation residual, physical residual, upper bound of state error, state validity indicator, distribution offset distance, risk index, and main risk boundary are recorded; when All status validity flags generated within each verification cycle are valid, and the risk index for each verification cycle is less than the domain switching risk threshold. If the target domain parameter set is enabled, proceed with steps S104 and S105; otherwise, disable the target domain parameter set and maintain the preset security control state.

[0108] Optionally, if any switching verification cycle fails to meet the above conditions, the target domain parameter set is disabled and the preset security control state is maintained. During the domain parameter switching process, the state data and control commands generated but not yet executed by the previous domain parameter version are invalidated, and the continuous abnormal cycle count and continuous recovery cycle count corresponding to the previous domain parameters are cleared. The domain parameter switching event is recorded, including the target device identifier, the previous and target domain identifiers and versions, the switching start time, the switching verification result, and the reason for the switching failure.

[0109] Optionally, a monotonically increasing and non-repeating control cycle number is set for each target device. The control cycle number is incremented each time new status data is generated, and the control cycle number most recently used by the target device is persistently saved; when the device restarts or the control process is reinitialized, the numbering continues from the already saved control cycle number.

[0110] Optionally, the first physical process is a heat conduction process, and the second physical process is an incompressible fluid transport process. The first domain parameter set corresponding to the heat conduction process is the temperature field (…). As a state variable, construct the residuals of the heat conduction equation according to the following formula: , in, Used to represent the density of a medium Used to express specific heat capacity Used to represent thermal conductivity Used to represent control quantities in thermal processes. The volumetric heat source term is used to represent the influence of control variables on the heating process. The initial condition residuals, isothermal boundary residuals, and heat flux boundary residuals of the heat conduction process are expressed by the following formulas: , , , in, Used to represent the initial temperature residual; Used to represent the temperature field at the initial moment, at position Location, initial moment Temperature distribution; Used to represent the initial reference temperature field; Used to indicate the temperature specified at the boundary; Used to represent boundary temperature residuals; Used to represent thermal conductivity; Used to represent temperature gradient; Represents the outer normal vector of the boundary; Used to represent the heat flux density specified at the boundary; Used to represent boundary heat flux residuals; Used to indicate the boundary region where the first type of boundary condition is located; This represents the boundary region where the second type of boundary condition exists. The residuals of the internal governing equations, initial conditions, and boundary conditions are combined according to preset weights to form the physical residual operator corresponding to the heat conduction process. Sparse measurement data of the heat conduction process are obtained from temperature sensors placed at different spatial locations. For the first... A temperature sensor, based on its spatial coordinates ( ) and sampling time ( The predicted temperature can be obtained by selecting the corresponding node temperature from the reconstructed temperature field or by interpolating from adjacent nodes. , The predicted temperatures from each effective temperature sensor are combined according to a preset sensor order to form the output of the heat conduction process observation operator. The set of operating boundaries for the heat conduction process includes at least the upper limit boundary of the highest temperature and the upper limit boundary of the maximum temperature gradient. The discrete temperature field is recovered based on the reduced-order state. After that, construct , The above content is compared with the maximum allowable temperature ( ) and maximum allowable temperature gradient ( The thermal process control variables include at least one of heating power, heater duty cycle, cooling medium flow rate, or cooling valve opening. Heating power affects the temperature field through the heat source term, while cooling medium flow rate affects the temperature field through the boundary heat flux or convective heat transfer coefficient. When the heat conduction process risk index reaches the corresponding risk threshold, the above control variables are limited, their rate of change is reduced, they transition to a safe setpoint, or a safe shutdown is executed.

[0111] Optionally, the physical residual operator in the first domain parameter set also includes the initial condition residual of the temperature field and the heat flux boundary condition residual. The observation operator is used to select or interpolate the predicted temperature from the temperature field based on the spatial coordinates of the temperature sensor. The running boundary set includes at least the upper temperature boundary and the upper temperature gradient boundary. The second domain parameter set corresponding to the incompressible fluid transport process is based on the velocity field ( ) and pressure field ( As state variables, construct the mass conservation residuals and momentum conservation residuals according to the following formulas: , , in, Used to represent the residuals of continuity equations; Used to represent velocity vector fields; Used to represent spatial differential operators; Used to represent the divergence of a velocity vector field; For fluid density; Used to represent the spatial gradient of the pressure field; For dynamic viscosity, For fluid process control variables. The residuals are the momentum conservation equation residuals; the physical residual operators in the second domain parameter set also include the initial condition residuals and boundary condition residuals of the velocity and pressure fields. The observation operators are used to obtain the predicted flow rate, predicted velocity, or predicted pressure from the velocity and pressure fields. The operating boundary set includes at least the upper pressure boundary, the lower flow rate boundary, and the upper velocity boundary. When the target industrial process is a heat conduction process, the first domain parameter set is loaded, and thermal process control commands are generated according to steps S101 to S105, and applied to the heating power regulating device, the cooling medium regulating device, or a combination thereof. When the target industrial process is an incompressible fluid transport process, the second domain parameter set is loaded, and fluid process control commands are generated according to steps S101 to S105, and applied to the pump speed regulating device, the valve opening regulating device, or a combination thereof.

[0112] Optionally, the boundary conditions of the fluid transport process include at least the inlet flow rate or inlet velocity boundary, the solid wall no-slip boundary, and the outlet pressure boundary. The corresponding boundary residuals include the difference between the predicted inlet value and the set value, the difference between the predicted solid wall velocity and zero velocity, and the difference between the predicted outlet pressure and the set outlet pressure. The mass conservation residual, momentum conservation residual, initial condition residual, and boundary condition residual are combined into a fluid physics residual operator.

[0113] Optionally, the sparse measurement data for the fluid transport process includes at least one of the measurements from pressure sensors, velocity sensors, and flow sensors. For measurements set at a cross-section ( The flow sensor at point () calculates the predicted flow rate according to the following formula: , in, Used to represent predicted flow; Used to represent the estimated velocity vector; Used to represent cross-section The unit normal vector.

[0114] Optionally, the set of operating boundaries for the fluid transport process includes at least an upper pressure boundary, a lower flow rate boundary, and an upper velocity boundary. After recovering the discrete velocity and pressure fields based on the reduced-order state, a... , , , in, Used to represent the first objective function; Used to represent estimated probabilities; Used to represent the second objective function; Used to represent the output area; Used to represent the estimated velocity vector field; Used to indicate the output area The unit outward normal vector; Used to represent the third objective function; The vector used to represent the l-th estimated velocity vector, and the safety conditions corresponding to the above objective functions are respectively... , and ,in, Used to represent the maximum allowed probability; Used to represent the minimum permissible flux; Used to indicate the maximum permissible speed.

[0115] Optionally, the fluid process control variables include at least one of pump speed, inlet pressure setpoint, inlet flow rate setpoint, and valve opening. Pump speed or inlet pressure setpoint affects the velocity and pressure fields through equivalent driving terms or inlet boundary conditions, while valve opening affects the velocity and pressure fields through local resistance or outlet pressure boundary conditions. When the fluid transport process risk index reaches the corresponding risk threshold, the above control variables are limited, their rate of change is reduced, they transition to safe setpoints, or a safe shutdown is performed according to the shutdown sequence of the target fluid equipment.

[0116] Optionally, the heat conduction process and the fluid transport process use the corresponding inversion operator, physical residual operator, observation operator, reduced-order basis matrix, error upper bound parameter, distribution offset statistic and running boundary set, respectively, but both execute the methods proposed in steps S101 to S105.

[0117] Optionally, when the target industrial process changes between heat conduction and fluid transport processes, the model parameters, reduced-order basis matrix, error upper bound calibration results, residual threshold, and operating boundary threshold of the physical process before the change are not directly shared. Instead, the target domain parameter set is selected, its consistency is verified, and the switching is verified before it is enabled.

[0118] In this embodiment of the application, when the controller generates a control command to be forwarded based on the status data, it numbers the status data. Periodic number ( Device identifier, domain identifier, domain parameter version, instruction generation time and the duration of the instruction Write the control command to be forwarded. Verify that the status data number in the control command to be forwarded corresponds to the currently stored status data, verify that the cycle number, device identifier, domain identifier, and domain parameter version are consistent, and obtain the verification time. ,check , The first item verifies whether the status data is still valid; the second item verifies whether the control command to be forwarded is still valid; and the third item prevents the generation time of the control command to be forwarded from being earlier than the generation time of its associated status data. Simultaneously, the cycle number in the control command to be forwarded... The cycle number corresponding to the most recently forwarded control command of the target device ( Compare and verify The control command to be forwarded will only be forwarded if all verifications pass, the status validity flag is valid, and the risk index is below the stop control threshold. Records are updated If any verification fails, the control command to be forwarded will be refused and downgraded control will be implemented; when the domain parameter version changes, all state data and control commands generated in the old version but not yet executed will be invalidated.

[0119] Optionally, if any verification fails, the forwarding of the control command to be forwarded is rejected, the corresponding reason for the verification failure is recorded, and a preset degradation control is implemented. When the domain identifier or domain parameter version of the target device changes, all corresponding and unused status data records and control commands to be forwarded before the change are marked as invalid. Those already marked as used, invalid, or with a control cycle number not greater than [a certain value] are excluded. The state data cannot be used again to generate forwardable control instructions.

[0120] In this embodiment of the application, for each target device, the domain parameter set further includes a recovery threshold ( ), first risk threshold ( ), second risk threshold ( Stop control threshold () ), lower bound vector of execution variables ( ), Execution variable upper limit vector ( ), the vector of maximum change in a single period ( Speed ​​limit factor () ), Safety setting value vector ( ), Safety Transition Change Vector ( ), type of safety action, maximum holding time ( ), stop command ( ) and number of recovery confirmation cycles ( ),in, , The set values ​​of each execution variable in the control command to be forwarded generated by the controller are arranged in a preset order to form a vector of control quantities to be forwarded, denoted as [vector]. The set values ​​of each execution variable that were actually forwarded and confirmed by execution in the previous control cycle are combined to form a control vector, denoted as . .when When the status validity flag is valid and the failure flag is 0, the following formula is used to determine the status validity. By applying upper and lower limit constraints and single-cycle variation constraints, the actual control commands to be executed are determined. , , in, A numerical vector representing the setpoints of control variables in a control command after clipping; clip[ ] is used to represent the amplitude limiting function; A numerical vector used to represent the set values ​​of control variables in the actual executed control commands; when When the status validity flag is valid and the failure flag is 0, a speed limiting coefficient greater than 0 and less than 1 is adopted according to the following formula. Reducing the maximum change per cycle can decrease the allowable change in control per cycle. .

[0121] when When the status validity flag is valid and the failure flag is 0, calculate the degradation transition control vector: , in, It is used to represent the safety setting intervention weight that increases with the increase of the risk index, and can also be called the downgrade transition coefficient; This is used to represent the degraded transition control vector obtained by weighting the control vector to be forwarded with the safety setting value vector.

[0122] The above formula enables the achievement of... Execute upper and lower limit constraints and single-cycle change constraints; the closer the risk index is to the stop control threshold, the closer the actual executed control command is to the safety setpoint. When the risk index reaches the stop control threshold, the status validity flag is invalid, the failure flag is 1, or any of the control command association consistency check, time validity check, and anti-repeated execution check fails, the control command to be executed is rejected, and one of the following processing options is selected based on the safety action type: safety setpoint transition, maintain control, or safety shutdown. When the safety action type is safety setpoint transition, a degradation control command is generated according to the following formula: , At the same time, it is repeated in subsequent control cycles until the safety set value is reached.

[0123] Optionally, when the safety action type is hold control, obtain the numerical vector composed of the control variable setpoints from the most recently executed control command that passed all checks. ) and its execution time ( Only at the current moment and The difference is no greater than ,and If the execution variable is within the allowed range and the target device operating procedure allows for retaining the most recent control command, then retain. If any condition is not met or the duration reaches [a certain threshold], [the condition will be considered as follows]. When the time comes, end the hold control and transition to the safety setpoint.

[0124] Optionally, when the safety action type is safety shutdown, a shutdown command corresponding to the target device is generated. A shutdown command must include at least the order in which execution variables are shut down, the safe rate of change for each execution variable, and the shutdown completion conditions. Once in a safe shutdown state, the target device will not be automatically restarted due to a decrease in the risk index; it will only be allowed to exit the safe shutdown state after receiving a valid restart permission and completing the verification of domain parameters, state validity, and operational boundaries.

[0125] Optionally, for the degraded control state formed by the transition and maintenance of the safety setpoint, only when the risk index is continuous Each control cycle satisfies The device exits the degraded control state only when the status validity flag remains valid, the failure flag remains at 0, and the control command verification continues to pass. Each degraded control event is recorded, including device identifier, domain identifier, domain parameter version, trigger time, trigger reason, risk index at trigger time, main risk boundaries, type of safety action used, and exit time or shutdown completion status.

[0126] Optionally, when If the status validity flag is invalid, the failure flag is 1, or the verification of the control command to be forwarded fails, forwarding is rejected. It will then execute safety setpoint transition, limited-duration hold control, or safety shutdown based on the type of safety action; after entering degraded control, it will only proceed if the risk index is continuously... Each control cycle satisfies Degradation control will only exit when the status validity flag remains valid continuously and the failure flag remains at 0.

[0127] By applying the technical solution of this embodiment, the domain parameter set, reconstructed state field coordinates, target sparse measurement data, executed control variables, and target physical condition data of the target equipment in the target industrial process are first obtained. Then, the above data are input into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on the state field, target sparse measurement data, and physical residual operator, the observation residual and physical residual are determined. Subsequently, the target reconstructed state field is dimensionally reduced using a reduced-order basis matrix to obtain a low-dimensional representation vector to determine the reduced-order projection residual and sparse interpolation residual. Combining the domain parameter set, observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual, the upper bound of the reduced-order state error is generated through error calculation. Then, for each initial operating boundary margin of the operating boundary set in the domain parameter set, it is corrected based on the sensitivity upper bound and the reduced-order state error upper bound of the corresponding operating boundary to obtain the operating boundary risk component and aggregate it into a risk index. Finally, the industrial state data of the current control cycle is determined based on the domain parameter set, target sparse measurement data, and risk index, thereby controlling the target equipment to operate normally in the target industrial process. In other words, in this embodiment, by combining physical constraint inversion, observation residuals, physical residuals, reduced-order projection residuals, and sparse interpolation residuals, a quantifiable upper bound for the reduced-order state error is generated. Furthermore, the operating boundary margin is corrected based on the sensitivity upper bound, and risk components are aggregated to obtain a risk index. This overcomes the problems of unquantifiable state reconstruction errors and lack of dynamic assessment of operating boundary risks in existing technologies. Simultaneously, based on the domain parameter set, sparse measurement data, and risk index, industrial state data is determined, compensating for the issues of misforwarding old or mismatched instructions due to the lack of binding between the control instructions to be forwarded and the industrial state data, as well as the lack of validity period binding. This solves the technical problem of difficulty in accurately controlling the operating state of equipment and achieves the technical effect of accurately controlling the operating state of equipment.

[0128] Furthermore, as Figure 1 In the specific implementation of the method, in the embodiments of this application, Figure 2 The following is a schematic flowchart of a cross-physical domain industrial control method provided in an embodiment of this application, as shown in the figure. Figure 2 As shown, the method includes the following steps: Step S201: Determine the target equipment and industrial process, and load the corresponding domain parameter set.

[0129] Step S202: Obtain sparse measurement data and control data, and perform time alignment and normalization on the above data.

[0130] Step S203: Input the physical constraint inversion operator, reconstruct the state field and calculate the residual.

[0131] Step S204: Reduce the order of projection and calculate the upper bound of the state error and the state validity.

[0132] Step S205: Propagate the error to the operating boundary, reduce the margin, and calculate the risk index.

[0133] Step S206: Generate status data for the binding identifier and receive control commands to be forwarded.

[0134] Step S207: Verify the consistency between the data bound to the control command and the status data, and execute forwarding or degradation control.

[0135] In this embodiment, an industrial heating plate with multiple independent heating zones and at least one cooling medium regulating channel is used as the target device. The control objective is to regulate the power and cooling medium flow rate of each heating zone while meeting the operating boundaries of the highest temperature and maximum temperature gradient. Target values ​​are set at preset measurement locations inside or on the surface of the industrial heating plate. There are 10 temperature sensors, each with a unique identifier and fixed spatial coordinates. Each measurement record includes at least the sensor identifier, temperature measurement value, sampling time, spatial coordinates, and quality flag. A control cycle is set, and... This means that at the beginning of each control cycle, the most recent valid quality flag measurement value of each temperature sensor within a preset time window is read, and calculations are performed. , in, Used to represent measured values; Used to indicate the current time; Used to indicate a reference time. When there is no valid measurement within the time window, the corresponding missing flag is set to 1, and a placeholder value that is not involved in the calculation of the observation residual is used to fill the gap.

[0136] Optionally, the target spatial region ( Temperature field within) The state field to be reconstructed is identified and discretized into preset values ​​( There are 10 temperature nodes. The control vector is composed of the power of each heating zone and the flow rate of the cooling medium, using the following formula: , in, Used to represent control vectors For the first Heating power of each heating zone; This refers to the flow rate of the cooling medium or its equivalent control quantity.

[0137] Optionally, the medium density, specific heat capacity, thermal conductivity, initial temperature distribution, external convective heat transfer conditions, and spatial distribution of heat sources corresponding to each heating zone are determined based on the material and structure of the industrial heating plate. The volumetric heat source term is constructed according to the following formula: , in, Used to represent volumetric heat source terms; For the first Heating power of each heating zone; For the first Normalized heat source distribution function for each heating zone.

[0138] Optionally, a thermal process domain parameter set is established for the industrial heating plate. This thermal process domain parameter set includes at least: thermal process domain identifier and version, equipment and configuration identifier, applicable operating condition range, temperature state variable definition, node order, coordinate definition, sensor mapping, temperature observation operator, heat conduction physical residual operator, normalization parameter, inversion operator parameter, order reduction parameter, error calibration parameter, maximum temperature and maximum temperature gradient boundaries, risk threshold, and safety action parameters.

[0139] Optionally, when the temperature sensor outputs a temperature in Celsius, it is first converted to the temperature unit used in physical calculations, and then normalized. The above process can be expressed by the following formula: , in, Used to indicate the lower limit of temperature; Used to indicate the upper temperature limit; Used to indicate the output temperature of a temperature sensor; This is used to represent the normalization problem. Furthermore, heating power, cooling medium flow rate, spatial coordinates, and relative sampling time are all processed using the same calculation method.

[0140] Optionally, when the thermal process domain parameter set is first enabled, the following parameters are verified: equipment identifier, number of heating zones, temperature sensor configuration, cooling channel configuration, temperature measurement vector dimension, observation operator input dimension, and number of rows in the reduced-order basis matrix. If any verification fails, the thermal process domain parameter set is not enabled, and the industrial heating plate is maintained in the safe state specified in the equipment operating procedures.

[0141] Optionally, before the industrial heating plate is put into online control, a thermal process reference sample is generated within the applicable ambient temperature, initial temperature, heating power, cooling medium flow rate, and workpiece thermal load range. The reference sample is derived from high-fidelity numerical calculation results verified by actual temperature measurements, encrypted temperature measurement results during the equipment commissioning phase, or a combination of both.

[0142] Optionally, sparse measurements are extracted from the reference temperature field based on the actual temperature sensor location, and corresponding measurement noise, relative sampling time, and missing data are set in the training samples according to the actual sensor accuracy, sampling period, and historical missing data. Measurements artificially set to be missing are not included in the calculation of observation loss.

[0143] Optionally, the thermal process reference samples are divided into an inversion model training sample set, an inversion model validation sample set, an error estimation training sample set, an independent calibration sample set, and an independent test sample set. Each sample set is unique; adjacent time samples from the same continuous heating experiment or the same numerical trajectory are grouped into the same set.

[0144] Optionally, the normalized sparse temperature measurements, missing markers, quality indicators, relative sampling time, and the heating power and cooling medium flow rate already executed within the corresponding time window are combined into a thermal process measurement feature vector. ).Will The input branch network is used to input the location to be reconstructed and the target time into the backbone network to obtain the predicted temperature field. The training loss for thermal process inversion is... , in, , , and These correspond to the effective measurement observation loss, thermal conduction physical residual loss, initial boundary condition loss, and reference temperature field difference loss, respectively. After training, the inversion operator parameters are fixed, and independent test samples do not participate in parameter or calibration quantile adjustments.

[0145] Optionally, the reference temperature field is arranged into a temperature state snapshot matrix according to a fixed node order and temperature normalization method. ), and perform singular value decomposition on it, based on the cumulative energy ratio ( Choose the reduced dimension ( ), to obtain the reduced-order basis matrix of the thermal process ( ), and calculate The heat conduction equations, heat source terms, and boundary heat flux terms corresponding to each reference temperature field are combined to form a thermal process residual snapshot matrix ( ), to obtain the residual basis matrix ( ) and selection matrix ( And bind it to the same reduced-order parameter version as the state error weighting matrix and the residual error weighting matrix.

[0146] Optionally, calculate the error estimation training samples separately. ,composition , And according to , Training the base temperature error estimation function with non-negative output Calculate calibration scores using independent calibration samples. For significance level ( ),Sure , and the corresponding calibration quantiles ( The upper bound of the online statistical temperature field error is expressed by the following formula: , Optionally, the effective calibration range is determined based on the ambient temperature, initial temperature, heating power, cooling medium flow rate, workpiece thermal load, number of effective sensors, and missing sensor type covered by the independent calibration samples; the mean, regularized covariance matrix, and distribution offset threshold are calculated based on the characteristics of normal training conditions. If the effective calibration range or offset threshold is exceeded, the upper bound of the statistical error with confidence level implications is not output.

[0147] Optionally, the parameters of the thermal process inversion operator, normalization parameters, order reduction parameters, error estimation model parameters, calibration quantiles, significance levels, calibration effective range, distribution offset statistics, maximum permissible temperature field error, and versions of each parameter are written into the thermal process domain parameter set. After any parameter is updated, a new version of the domain parameters is generated, and the affected training and calibration are re-executed.

[0148] Optionally, before the industrial heating plate starts operating, the thermal process domain parameter set is selected and loaded according to the equipment identification, number of heating zones, temperature sensor configuration, and current domain parameter version, and the equipment, sensor, state dimension, model, reduced order basis, error calibration, and operating boundary version verification are completed.

[0149] Optionally, in each control cycle, the temperature measurement record, the heating power record of each heating zone, and the cooling medium flow record are read, time-aligned with the reference time, and a temperature measurement vector is generated. ), valid measurement mark ( The data includes the relative sampling time vector and the executed thermal process control vector. Inputting these data into the thermal process physical constraint inversion operator yields the reconstructed field (…). ), and calculate , , Calculated based on the reduced basis matrix of the thermal process , And based on the thermal process residual basis and the selection matrix, the interpolation residual is calculated to form the current error eigenvector. ).

[0150] Optionally, the upper bound of the state error is calculated based on the fundamental error estimation function and the calibration quantile. Verify that the ambient temperature, initial temperature, applied heating power, cooling medium flow rate, number of valid sensors, and missing sensor type are within the calibration validity range, and calculate the distributed offset distance. If the error boundary is valid, proceed according to... The upper bound of the complete temperature field error is propagated into the upper bound of the reduced-order state error to obtain the upper bound of the reduced-order state error.

[0151] Optionally, the temperature field is recovered from the reduced-order state of the thermal process, and the following calculations are performed. , At the same time, the corresponding uncorrected margin is calculated.

[0152] Optionally, according to , Shrinkage boundary margin, calculation The thermal process risk index is obtained using the following formula: =clip(1- ,0,1), =clip(1- ,0,1), , in, Used to represent the risk index of thermal processes.

[0153] Optionally, the anomaly indication and continuous anomaly period are calculated based on the temperature observation residual, thermal conduction physical residual, distribution offset distance, and the number of effective temperature sensors. When the continuous mismatch condition, hard threshold condition, or no effective temperature measurement is met, the thermal process failure flag is set to 1.

[0154] Optionally, thermal process status data for the current control cycle is generated. Based on the status data and the preset temperature target, the controller uses existing control rules of the industrial heating plate to generate control commands to be executed for the power and cooling medium flow rate of each heating zone, and writes the associated status data number, device identifier, domain identifier, domain parameter version, and validity period.

[0155] Optionally, the thermal process control command to be executed is validated for status data number, control cycle number, equipment identifier, domain identifier, domain parameter version, control variable name, control variable unit, and validity period. If all validations pass and the risk is below the first risk threshold, the command is forwarded after executing the upper and lower limits and single-cycle change constraints.

[0156] Optionally, when the thermal process risk index is between the first risk threshold and the second risk threshold, the allowable change in the control quantity is reduced; when it is between the second risk threshold and the stop control threshold, it is gradually adjusted to the safe setpoint determined by the safety test; when the stop control threshold is reached, the state is invalid, the failure flag is 1, or the command verification fails, the original command is rejected and the heating power and cooling medium flow rate are reduced in a safe sequence.

[0157] Optionally, after entering degraded control, temperature measurements continue to be acquired and state reconstruction and risk calculations are performed. Exiting degraded control is only permitted if the risk index continuously meets the recovery threshold, the state remains valid, and the failure flag remains at 0. If a safe shutdown has already been performed, the sensor configuration, domain parameter version, current temperature state, and operating boundaries are re-verified before restarting.

[0158] Optionally, when the estimated maximum temperature is still below the maximum allowable temperature, but the error correction margin obtained after considering the upper bound of the reduced-order state error is less than or equal to 0, the maximum temperature risk component is set to 1. This embodiment thus refuses to continue executing normal control commands that might increase heating power and enters a safe setpoint transition or safe shutdown process.

[0159] In this embodiment, the target device is an industrial fluid transport apparatus including a variable frequency pump, a transport pipeline, and at least one regulating valve. The transport medium is a fluid that can be processed as a single-phase incompressible Newtonian fluid within the corresponding applicable operating conditions. The control objective is to ensure that the outlet flow rate meets production requirements and that the pipeline pressure and flow rate are maintained within permissible ranges.

[0160] Optionally, preset values ​​can be set at different locations in the pipeline. ) pressure sensors, with preset values ​​set at one or more cross sections ( ) flow sensors, and set preset values ​​as needed ( ( ) velocity sensors. Each fluid measurement record must include at least the sensor identifier, sensor type, measured value, sampling time, coordinates of the measurement point or measurement section identifier, measurement direction, quality mark, and unit of measurement.

[0161] Optionally, at the beginning of each control cycle, the most recent valid measurement value of each sensor within a preset time window is read based on a reference time, and the relative sampling time is calculated. When no valid measurement value is found, the corresponding missing flag is set to 1. Pressure, flow, and velocity measurements are arranged into a fixed-field fluid measurement vector according to the order of their respective sensors.

[0162] Optionally, the velocity field within the target pipeline space region and pressure field The state field to be reconstructed is identified, and the fluid state vector is formed according to a preset order, so as to combine the pump speed and the control valve opening into the fluid process control vector: , , in, Used to represent the fluid state vector; F is used to represent the fluid physical domain; These represent the components of velocity in the x, y, and z directions in space, respectively; p represents the pressure data. A vector used to represent control quantities in a fluid process; This indicates the pump speed or pump angular velocity setpoint. Used to indicate the set value of the control valve opening.

[0163] Optionally, the fluid density, dynamic viscosity, geometry, inlet cross-section, outlet cross-section, solid wall location, and local resistance location are determined based on the transport medium and target pipeline. The physical residual operator includes at least: , , in, Used to represent physical residual operators.

[0164] Optionally, pump speed affects the velocity and pressure fields through inlet pressure, inlet flow rate, or equivalent driving terms; valve opening affects the velocity and pressure fields through local resistance relationships or outlet pressure boundaries. Solid pipe walls employ no-slip boundaries. New versions of fluid domain parameters are generated when pump characteristic curves, valve flow characteristics, or pipeline structure change.

[0165] Optionally, the fluid domain parameter set ( It should include at least the fluid domain identifier and version, equipment and configuration identifier, applicable fluid parameters and operating condition range, velocity and pressure state definition, pipeline coordinates and node sequence, sensor mapping, fluid observation operator, fluid physics residual operator, normalization parameter, inversion operator parameter, order reduction parameter, error calibration parameter, pressure, flow rate and velocity boundaries, risk threshold, and pump and valve safety action parameters.

[0166] Optionally, pressure, velocity, and flow rate are each normalized using their respective parameters. The fluid state error weighting matrix is ​​determined based on the allowable scales of the pressure error and the errors of each velocity component. The flow sensor prediction is calculated according to the following formula: , Furthermore, the discrete implementation uses the area weight of the measurement section and the normal velocity component for calculation.

[0167] Optionally, when the fluid domain parameter set is first enabled, the pump and valve identifiers, sensor mappings, measurement unit conversions, observation operator input dimension, inversion operator output dimension, number of rows in the reduced-order basis matrix, and running boundary function input dimension are verified. If any verification fails, the fluid domain parameter set is not enabled, and a safe pump speed and safe valve opening are maintained, or a safe shutdown is performed.

[0168] Optionally, before the fluid transport equipment is put into online control, a fluid process reference sample is generated within the range of fluid density, dynamic viscosity, pump speed, valve opening, inlet conditions, and outlet conditions specified in the fluid domain parameter set. The reference sample is derived from high-fidelity fluid numerical calculation results verified by actual pressure and flow measurements, encrypted measurement results obtained during the equipment commissioning phase, or a combination of both.

[0169] Optionally, since the pressure field of incompressible fluids requires a pressure reference, this embodiment determines the preset outlet section pressure or the pressure at a specified pressure measurement point as the reference pressure, and writes the reference pressure definition into the fluid domain parameter set. The reference pressure field, pressure measurement values, and pressure boundary thresholds all use the same absolute pressure or gauge pressure definition. When the reference pressure definition changes, the pressure state sample, normalized parameters, reduced-order basis matrix, and pressure boundary parameters are regenerated.

[0170] Optionally, point measurements are obtained by selecting or interpolating from the reference state field based on the actual pressure and velocity sensor locations; reference flow rates are obtained by integrating the reference velocity field based on the measurement cross-sections of each flow sensor. Appropriate measurement noise, relative sampling time, and missing data are set in the training samples according to sensor accuracy, sampling period, and historical data loss.

[0171] Optionally, the fluid process reference samples can be divided into an inversion model training sample set, an inversion model validation sample set, an error estimation training sample set, an independent calibration sample set, and an independent test sample set. Adjacent time samples from the same fluid dynamics experiment, the same pump / valve adjustment trajectory, or the same numerical calculation trajectory are grouped into the same set.

[0172] Optionally, the normalized pressure, flow, and velocity measurements, missing data markers, quality indicators, relative sampling time, pump speed and valve opening already performed within the corresponding time window, fluid density, and dynamic viscosity are combined into a fluid process measurement feature vector. .Will Input the branch network, input the location to be reconstructed and the target time into the backbone network, and obtain the predicted velocity field and predicted pressure field through the output heads corresponding to each velocity component and pressure variable.

[0173] Optionally, the training loss for fluid process inversion is , in, , , , and These correspond to observation consistency loss, mass conservation residual loss, momentum conservation residual loss, initial boundary condition residual loss, and reference state field difference loss, respectively. and This refers to the residual loss of the governing equations in this scenario, also known as the fluid physics residual loss. Each loss term is calculated at its corresponding normalized scale or weighted to a comparable scale. The fluid physics residual loss includes the mass conservation residual loss. and momentum conservation residual loss And weights are set for each of them.

[0174] Optionally, the model parameters of the fluid inversion operator are updated using training samples, and the loss weights, number of feature terms, and training stopping time are selected using validation samples. After training, the model parameters are fixed, and the observation residuals of pressure, flow rate, velocity, mass conservation, momentum conservation, and fluid state field reconstruction error are calculated on both validation and independent test samples. Independent test samples do not participate in any parameter adjustment.

[0175] Optionally, after normalizing the reference velocity field and reference pressure field, a fluid state snapshot matrix is ​​formed according to the variable order and node order specified in the fluid domain parameter set. ).right Perform singular value decomposition based on a preset cumulative energy ratio ( Choose the fluid reduction dimension ( ), to obtain the fluid reduced-order basis matrix ( ), and calculate The fluid residual snapshot matrix is ​​composed of the mass conservation term, convection term, viscosity term, pressure gradient term, driving term, and boundary residual term. ), to obtain the fluid residual basis matrix ( ) and selection matrix ( This is bound to the same fluid state error weighting matrix and residual error weighting matrix as a version of the fluid reduction parameter.

[0176] Optionally, calculate the error estimation training samples separately. ,composition , Training the basic fluid error estimation function The calibration quantiles were determined using independent calibration samples. The upper bound of the online statistical fluid state error is... , Optionally, the effective calibration range and distribution offset threshold are determined based on the fluid density, dynamic viscosity, Reynolds number, pump speed, valve opening, inlet conditions, outlet conditions, number of effective sensors, and measurement missing methods covered by the independent calibration samples. When pump characteristics, valve relationships, pressure reference, or any model and calibration parameters change, a new version of the fluid domain parameters is generated, and the affected training and calibration are re-executed.

[0177] Optionally, before the fluid transport equipment is put into operation, the fluid domain parameter set is selected and loaded according to the equipment identification, pump and valve configuration, pipeline configuration, sensor configuration and current domain parameter version, and the equipment, pressure reference, sensor mapping, state dimension, model, reduced order basis, error calibration and operating boundary version verification are completed.

[0178] Optionally, pressure, flow, and velocity measurement records are read for each control cycle, along with the pump speed and valve opening already executed within the corresponding time window. Time alignment is performed using a reference time to generate a fluid measurement vector. ), valid measurement mark ( ), relative sampling time vector and executed fluid control vector.

[0179] Optionally, the above data, fluid density, dynamic viscosity, and coordinates of the node to be reconstructed are input into the fluid physics-constrained inversion operator to obtain the reconstructed velocity and pressure fields, and then the calculations are performed. , The observation operator uses velocity field cross-sectional integration for flow measurement. The reconstructed velocity field, pressure field, fluid parameters, executed pump and valve control variables, inlet conditions, outlet conditions, and wall conditions are input into the fluid physics residual operator to calculate the mass conservation residuals. ), momentum conservation residual ( ) and boundary condition residuals ( ), and form , Calculated from the fluid reduced basis matrix , And based on the residual basis and the selection matrix, the interpolation residual is calculated to form the current fluid error eigenvector. ).

[0180] Optionally, the upper bound of the state error is calculated based on the fundamental error estimation function and the calibration quantile. Verify that the current fluid density, dynamic viscosity, Reynolds number, pump speed, valve opening, inlet conditions, outlet conditions, number of valid sensors, and missing sensor type are within the calibration validity range, and calculate the fluid distribution offset distance. When the operating conditions exceed the calibration validity range, the offset distance exceeds the threshold, or the pressure reference is inconsistent, set the statistical error limit and status validity flag to invalid.

[0181] Optionally, when the statistical fluid state error upper bound is valid, the fluid reduced-order state error upper bound is calculated according to the following formula: , in, Used to represent the upper bound of the fluid reduction state error. When When the maximum permissible fluid state error is exceeded, the state validity flag will be set to invalid.

[0182] Optionally, the velocity and pressure fields are recovered from the reduced-order fluid state, and the following calculations are performed. , , , Calculate the uncorrected margins for the upper pressure limit, lower flow limit, and upper velocity limit separately. Calculate the contraction boundary margin using the following formula: , , , Based on the upper bound of the sensitivity contraction margin corresponding to the three operating boundaries, the risk components are calculated respectively, and the fluid risk index is obtained using the following formula: =clip(1- ,0,1), =clip(1- ,0,1), =clip(1- ,0,1), , in, Used to represent the fluid risk index.

[0183] Optionally, the anomaly indication, the number of anomaly conditions, and the count of consecutive anomaly cycles are determined based on the fluid observation residual, fluid physics residual, fluid distribution offset distance, and the number of effective fluid measurements. When the continuous mismatch condition, hard threshold condition, or no effective fluid measurement is met, the fluid failure flag is set to 1.

[0184] Optionally, fluid state data for the current control cycle is generated. Based on the fluid state data and the target outlet flow rate or target pressure, the controller uses existing control laws to generate control commands to be executed for pump speed and valve opening, and writes the associated state data number, device identifier, domain identifier, domain parameter version, and validity period.

[0185] Optionally, the control command to be executed is verified for status data number, control cycle number, device identifier, domain identifier, domain parameter version, pump / valve identifier, control variable unit, and validity period. If all verifications pass and the risk is below the first risk threshold, upper and lower limits and single-cycle change constraints are applied to the pump speed and valve opening before forwarding; if the risk increases, speed limiting and transition to safe setpoints are executed sequentially.

[0186] Optionally, when the fluid risk index reaches the stop control threshold, the status is invalid, the failure flag is 1, or the control command verification fails, the original control command is rejected, and the pump speed and valve opening are adjusted according to a preset safety sequence. For pipelines that may experience sudden pressure changes, instead of simultaneously executing instantaneous pump shutdown and instantaneous valve closure, adjustments are made gradually according to the preset maximum rate of change.

[0187] Optionally, when the estimated maximum pressure point is lower than the maximum allowable pressure, but the pressure error correction margin obtained after considering the upper bound of the fluid reduced-order state error is less than or equal to 0, the pressure risk component is set to 1, and normal control commands that may increase pump speed or pipeline pressure are rejected. The same error contraction judgment is used for the lower limit boundary of the outlet flow rate.

[0188] In this embodiment, the control computing device pre-stores a set of thermal process domain parameters and a set of fluid domain parameters. The current control task is the heat conduction process of an industrial heating plate; when the control task changes to a fluid transport device, a domain parameter switching method is executed. The change in control task includes changes in the target device identifier and the physical process type.

[0189] Optionally, upon receiving a control task change instruction, the generation of new normal control instructions for the industrial heating plate is stopped, and the thermal process status data and control instructions that have already been generated but not yet forwarded are marked as invalid. The industrial heating plate is kept in a safe state according to the equipment operating procedures, while the fluid delivery equipment is kept at a preset safe pump speed and safe valve opening.

[0190] Optionally, the equipment identifier, pump configuration, valve configuration, pipeline configuration, sensor configuration, and the version of the domain parameters to be enabled for the fluid transport equipment are read. When a unique set of fluid domain parameters that meets the conditions and whose applicable operating conditions include the current operating conditions is found, it is identified as the set of domain parameters to be verified; otherwise, the switching result is determined to be a failure.

[0191] Optionally, pressure, flow, and velocity sensor fields are mapped according to the sensor mapping relationships in the parameter set of the domain to be verified; pressure reference, flow, and velocity units are unified according to unit definition; and measurement points, measurement sections, and normal vectors are transformed according to pipeline coordinates. Measurement data that cannot complete a unique mapping or transformation are marked as invalid.

[0192] Optionally, the following parameters are verified sequentially: the input dimension of the fluid measurement vector and observation operator, the output variables and dimension of the inversion operator, the number of rows and columns of the reduced basis matrix, the input dimension of the running boundary function, the physical parameters and pressure reference, the error calibration version, the distribution offset statistic, the risk threshold, and the pump and valve execution variable definitions. If any of these fail, the parameter set of the domain to be verified is not enabled.

[0193] Optionally, if all consistency checks pass, and the fluid transport equipment maintains a safe pump speed and safety valve opening, the parameter set of the domain to be verified is used for continuous execution. The switching verification cycle is performed in steps S101 to S105, but normal control commands generated based on the reconfiguration state are not forwarded to the pumps and valves.

[0194] Optionally, when the status of each verification cycle is valid, the failure flag is 0, the observation residual and physical residual do not exceed the threshold, the distribution offset distance does not exceed the threshold, and the risk index is lower than the domain switching risk threshold, the fluid domain parameter set is enabled, and the status data number continues to be generated according to the control cycle number persistently saved by the device.

[0195] Optionally, if the above conditions are not met in any verification cycle, the fluid domain parameter set will not be enabled, the temporary state data generated in this verification will be cleared, and the fluid transport equipment will remain in a safe control state. The equipment identifier, domain parameter version, failure time, failure steps, and specific failure reasons will be recorded.

[0196] Optionally, when the controller incorrectly uses thermal process state data to generate fluid control commands, the device identifier, domain identifier, domain parameter version, and control variable name will be inconsistent with the current fluid state data. When the thermal process domain parameter set is incorrectly selected as the fluid verification domain parameter set, the variable dimension and unit consistency checks will fail.

[0197] Optionally, when switching from a fluid transport control task to a heat conduction control task, the same domain parameter selection, consistency check, and continuous verification steps are performed in reverse order. The two physical domains only reuse the processing order of steps S101 to S105, and do not share inversion operator parameters, physical residual operators, reduced-order basis matrices, error calibration results, or operating boundary thresholds.

[0198] Optionally, each operating boundary threshold is determined based on the target equipment's rated parameters, process allowable range, material allowable range, and equipment operating procedures. For the upper limit boundary, a threshold value of [missing value] can be used. For the lower boundary, we can take... This results in an uncorrected normalization margin of 1 when the process variable is at the warning value, and 0 when it reaches the boundary threshold. Wherein, Used to indicate the first j Warning values ​​for the operational boundary and boundary function values and boundary threshold Using the same physical units or normalized scale, i.e., located at the actual safety boundary. The warning threshold that was previously within the safe zone.

[0199] Optionally, let the first The minimum normalization margin required to be retained at the operating boundary is Then the maximum permissible reduced-order state error satisfies: , And according to Determine the maximum permissible state field error.

[0200] Optionally, for the maximum risk aggregation, let the error correction margins corresponding to normal control, speed limiting control, and stopping normal control be respectively... ,and Then it is acceptable. , When using other aggregation methods, the threshold is re-determined using the corresponding device verification data.

[0201] Optionally, the observation residual threshold, physical residual threshold, and distribution offset threshold can be determined based on the preset upper quantile and conservative margin of the normal verification sample statistics; the hard threshold can be determined based on the maximum allowable deviation of the sensor, the maximum allowable mismatch of the physical equation, and the results of abnormal equipment tests. The number of failure confirmation cycles should filter out normal transients, and the number of recovery confirmation cycles should be determined based on the time required for the equipment to return to stability.

[0202] Optionally, the longest valid duration of the state data is determined based on the rate of change of the process state and the allowable additional time error; that is, the above process can be represented as: , in, Used to represent the upper bound of the maximum rate of change of the reduced-order state; This is used to indicate that additional reduced-order state errors are allowed, while letting This ensures that the effective cutoff time of control commands is no later than the effective cutoff time of the status data upon which they are based.

[0203] Optionally, let the first The maximum allowed rate of change for each execution variable is: The control cycle is Therefore, the safe transition change should also limit the temperature gradient of the heating equipment and the pressure change of the fluid equipment, which can be expressed by the following formula: , in, It is used to represent the maximum absolute change that the i-th execution variable is allowed to occur within a control cycle.

[0204] Optionally, safety setpoints and shutdown commands are determined through safety tests and operating procedures for the target equipment. For equipment with multiple execution variables, the safety setpoints need not be equal to the minimum or shut-off values ​​of each execution variable, but rather a combination of control variables that enables the equipment to stably enter a verified safe state.

[0205] Optionally, the number of independent calibration samples and the significance level of the upper bound of the statistical error satisfy... , in, This indicates the number of valid calibration samples in an independent calibration sample set. Independent calibration samples should cover the main operating conditions, valid sensor quantity range, and permissible missing data range specified in the domain parameter set; they only increase the number of consecutive adjacent samples under the same stable operating condition and cannot replace coverage of different operating conditions.

[0206] Optionally, before the domain parameter set is officially enabled, at least the following should be verified: state reconstruction, error upper bound, and continuous output of risk index under normal operating conditions; the error correction margin decreases as the error upper bound increases when approaching the operating boundary; sensor missing, measurement delay, measurement deviation, expired state data number, faulty device identifier, faulty domain parameter version, repetitive control cycle number, and overdue control command should be verified; the amplitude limit, speed limit, safety setpoint transition, and safe shutdown for each risk level should be verified; and the old data and old commands should be verified to be invalid when switching domain parameters.

[0207] Optionally, record the equipment configuration, domain parameter version, input conditions, expected results, actual results, and judgment conclusions used for each verification. If the verification results do not meet the expected conditions, the corresponding domain parameter set is not enabled. If the model, normalization, order reduction, error calibration, operating boundary, risk threshold, effective duration, or safety action parameters are changed, the affected verification items are re-executed according to the changes, and a new version is generated.

[0208] In the embodiments of this application, Figure 3 This illustration shows an offline training diagram of a physical constraint inversion operator provided in an embodiment of this application, such as... Figure 3 As shown in the figure, this diagram illustrates the architecture of a state field prediction model based on deep learning. The input on the left includes sparse measurements of the reference state field, control variables, physical conditions, and spatiotemporal coordinates. r , t The network is divided into a branch network and a trunk network, which process the reference state field data and the spatiotemporal coordinates, respectively. The outputs of the two networks are fused through an output header structure to generate the predicted state field. The model is optimized using multiple loss functions: observation loss. (Matching actual observations), physical residual loss (Satisfies physical laws), initial boundary loss (Meets initial / boundary conditions), reference difference loss (Narrowing the difference from the reference field), weighted average of each loss ( The total loss was then obtained. L By updating network parameters through backpropagation, accurate prediction of the state field can be achieved.

[0209] In the embodiments of this application, Figure 4 This illustration shows a schematic diagram of the construction of a reduced-order basis matrix and sparse interpolation points according to an embodiment of this application, as shown below. Figure 4 As shown, the left side represents state reduction: first, a state snapshot matrix is ​​constructed. After singular value decomposition (SVD), we obtain Cut off the front Column (satisfying) η After following the criteria, the reduced-order basis matrix is ​​obtained. The right side shows residual order reduction and interpolation: first, construct the residual snapshot matrix G, and G = [ f ( x 1), f ( x 2), ..., f ( x Nt )], obtained through SVD decomposition Extracting the residual basis matrix The selection matrix is ​​then determined using Discrete Empirical Interpolation (DEIM / QR). P And sparse interpolation points, run the preset algorithm (DEIM algorithm) to generate the selection matrix. P This is used for residual approximation. Overall, model simplification and efficient residual processing are achieved through SVD order reduction and DEIM interpolation.

[0210] In the embodiments of this application, Figure 5 The following is a flowchart illustrating a method for calculating the upper bound of a state error according to an embodiment of this application. Figure 5 As shown, the method includes at least the following steps: Step S501: Obtain the observation residual, physical residual, projection residual, and interpolation residual.

[0211] Step S502: Determine whether the current operating condition of the target equipment belongs to the set of deterministic applicable operating conditions and whether the lower bound of the stability constant is valid.

[0212] In this embodiment, it is necessary to determine whether the current operating condition of the target device belongs to the set of deterministic applicable operating conditions and whether the lower bound of the stability constant is valid. If not, step S503 is executed; if yes, step S505 is executed.

[0213] Step S503: Determine whether the current operating condition is within the effective range of independent calibration and whether no distribution shift has occurred.

[0214] In this embodiment, it is necessary to determine whether the current operating condition is within the effective range of independent calibration and whether no distribution shift has occurred. If not, proceed to step S504; if yes, proceed to step S506.

[0215] Step S504: Set the status validity flag to invalid.

[0216] Step S505: Calculate the upper bound of the deterministic state error and output the deterministic error bound type and applicable operating condition identifier.

[0217] Step S506: Output the basic error estimation function, the upper bound of the statistical error, and the confidence level.

[0218] In the embodiments of this application, Figure 6 This illustration shows a structural diagram of the propagation of state error to operational boundary risk according to an embodiment of this application. Figure 6 The diagram illustrates the margin adjustment and risk calculation process in risk analysis. The upper part of the diagram shows the state point estimation. With boundary threshold Error margin between After sensitivity correction (considering) After that, a correction margin is formed. This reflects the consumption of the safety margin and the remaining safety distance. The second half of the data flow is as follows: first calculate the uncorrected margin. Introduce a lower-order state error upper limit. Contraction margin, then calculate boundary risk components The final aggregation yields the risk index. R It is used to quantify the risks of system state and safety boundaries, and improves the accuracy of risk assessment through margin correction and error constraints.

[0219] In the embodiments of this application, Figure 7 This illustration shows a conversion diagram of domain mismatch detection and failure flag conversion provided in an embodiment of this application, as shown below. Figure 7 The diagram illustrates the system state determination and switching process, divided into normal and failure states. In the normal state: data input includes observation residuals, physical residuals, distribution offset distances, and the number of valid measurements; the judgment logic box calculates the abnormal indicators (…). , , and ), and sum them up to get C ( t If any failure trigger condition is met (if...); C (t )≥2 and continuous K f If the number of cycles reaches a certain threshold (or the number of effective measurements is 0), the system enters a failure state, i.e., the failure flag is set to 1, triggering degradation control. In the failure state: the recovery conditions are met ( C ( t ) = 0 and continues K r If the system returns to normal if the status is valid for a given period, it will continue to degrade control. In other words, this process achieves system state switching and fault recovery management through anomaly detection and condition triggering.

[0220] In the embodiments of this application, Figure 8 This application provides a flowchart illustrating a domain parameter switching method according to an embodiment of the present application. Figure 8 As shown, the method includes the following steps: Step S801: Receive control task change instruction.

[0221] Step S802: Keep the target device in a preset security control state and set the status data and control commands that have not yet been executed in the old domain to invalid.

[0222] Step S803: Read the device identifier, physical process type, device configuration identifier, and parameter version to be enabled, and search for the target domain parameter set.

[0223] Step S804: Has a unique target domain parameter set that is applicable to the current working condition been found?

[0224] In this embodiment, it is necessary to determine whether a unique target domain parameter set applicable to the current working condition has been found. If so, step S805 is executed; otherwise, step S806 is executed.

[0225] Step S805: Measurement data mapping and transformation.

[0226] Step S806: Prohibit the startup of the target domain parameter set, maintain the preset security control state, discard temporary state data, and record the reason for the switch failure.

[0227] Step S807, target domain consistency verification.

[0228] Step S808: Determine whether the dimension, variable definition, and parameter version of the measurement data are all correct.

[0229] In this embodiment, it is necessary to determine whether the dimension, variable definition, and parameter version of the measurement data are all correct. If yes, then step S809 is executed; otherwise, then step S806 is executed.

[0230] Step S809: Execute the target value verification cycle under the preset safety control state, without forwarding normal control commands, and record the residual, upper limit of error, state validity, failure flag, distribution offset distance and risk index.

[0231] Step S810: Within each verification period of the target value, whether the status validity flag and failure flag are both 0, the residual and distribution offset do not exceed the threshold, and the risk index is less than the preset threshold.

[0232] In this embodiment, it is determined whether the status validity flag and failure flag are both 0, the residual and distribution offset do not exceed the threshold, and the risk index is less than the preset threshold within the target value verification period. If yes, step S811 is executed; if no, step S806 is executed.

[0233] Step S811: Enable the new domain parameter set.

[0234] In the embodiments of this application, Figure 9 This illustration shows a flowchart of a method for binding and verifying status data and control commands according to an embodiment of this application. Figure 9 As shown in the flowchart, this process diagram illustrates the verification process and data structure before status data forwarding. The "Status Data Structure" on the left includes fields such as status data number, cycle number, status generation time, effective expiration time, device identifier, domain identifier, domain parameter version, status validity / expiration flag, and risk index R. The "Control Instruction to be Forwarded" on the right includes fields such as associated status data number, cycle number, executing device identifier, domain identifier, domain parameter version, instruction generation time, instruction validity duration, control variable name and unit, and execution variable set value. The intermediate process involves four verification steps: 1. Consistency verification of associated fields, ensuring that the status data number, cycle number, etc., are consistent with the current domain parameter set definition; 2. Time validity verification, determining whether the current time is between the generation time and the effective expiration time, and within the control cycle; 3. Anti-duplicate execution and status recording verification, checking whether the record exists, is unused, and is not expired, and the cycle number... 4. Status and risk verification: confirm that the status validity flag is valid, the failure flag is 0, and the risk index is [missing information]. R < If the verification passes, the command is atomically updated and forwarded (marking the status data as used, updating the cycle number, and forwarding to the device); if it fails, forwarding is rejected, the reason for the failure is recorded, and degradation control is implemented. Overall, multi-dimensional verification ensures the reliability of control command forwarding.

[0235] In the embodiments of this application, Figure 10 This illustration shows a control diagram of a risk classification and degradation control method provided in an embodiment of this application, such as... Figure 10As shown, this hierarchical control process begins with state generation and risk assessment, and then proceeds to obtain the risk index. R Afterwards, according to R Different control strategies are adopted for different intervals: when When, refuse to forward instructions, execute either maintain control or safe shutdown; when During this period, a safe transition will be implemented proportionally. λ ( t Approaching the safety set value ;when At that time, speed limit control, proportional γ Reduce the amount of change in a single period; when During normal operation, upper and lower limits and variation constraints are executed. However, if an invalid state, a failure flag of 1, or an abnormal trigger condition occurs, a super jump to interval 4 will be triggered. Overall, safety control decisions are achieved through risk index grading, where interval 4 represents the interval where the risk index meets the requirements. The interval, i.e., the stop / safety stop interval.

[0236] In the embodiments of this application, Figure 11 This illustration shows a schematic diagram illustrating the application relationship between heat conduction and fluid transport processes according to an embodiment of this application. Figure 11 The diagram illustrates the domain parameter switching consistency verification process, divided into three parts: The heat conduction domain includes domain parameters (temperature field, heat conduction residual operators, etc.), operating boundaries (maximum temperature, maximum temperature gradient), and execution objects (heaters, cooling valves); the fluid transport domain includes domain parameters (velocity / pressure field, mass / momentum conservation residual operators, etc.), operating boundaries (pressure upper limit, flow velocity upper limit, etc.), and execution objects (variable frequency pumps, regulating valves); the core logic of cross-physical domain industrial control involves verifying the consistency of domain parameter switching between the heat conduction and fluid transport domains through the steps of "state reconstruction → error propagation → boundary risk → command verification," ensuring the safety and accuracy of cross-domain control. The overall process revolves around cross-domain parameter consistency verification, guaranteeing the stable operation of the industrial control system.

[0237] Furthermore, as Figure 1 In the specific implementation of the method, in the embodiments of this application, Figure 12 This illustration shows a schematic diagram of a device operation state control device based on a physical constraint inversion operator according to an embodiment of this application. Figure 12 As shown, the device operation status control device 1200 based on physical constraint inversion operator includes: acquisition unit 1201, determination unit 1202, generation unit 1203, correction unit 1204 and control unit 1205.

[0238] The acquisition unit 1201 is used to acquire the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control quantities and target physical condition data of the target device within the current measurement time window.

[0239] The determination unit 1202 is used to input the target sparse measurement data, the target executed control variables, the target physical condition data, and the reconstructed state field coordinates into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined.

[0240] The generation unit 1203 is used to perform dimensionality reduction and compression processing on the target reconstructed state field using the reduced-order basis matrix to obtain a low-dimensional representation vector. Based on the domain parameter set, observation residual, physical residual, reduced-order projection residual and sparse interpolation residual, the upper bound of the reduced-order state error is generated by using an error calculation method.

[0241] The correction unit 1204 is used to correct the initial operating boundary margin of each operating boundary in the set of operating boundaries in the domain parameter set based on the upper bound of sensitivity, the upper bound of reduced state error and the low-dimensional representation vector of each operating boundary, to obtain multiple operating boundary risk components, and to aggregate the multiple operating boundary risk components to obtain a risk index.

[0242] The control unit 1205 is used to determine the industrial status data of the target equipment in the current control cycle based on the domain parameter set, target sparse measurement data and risk index, and bind the industrial status data with the control command to be forwarded, so as to control the target equipment to be in normal operation in the target industrial process through the control command to be forwarded.

[0243] In this embodiment, the acquisition unit acquires the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control variables, and target physical condition data of the target device within the current measurement time window. The determination unit inputs the target sparse measurement data, target executed control variables, target physical condition data, and reconstructed state field coordinates into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on the target reconstructed state field, target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined. The generation unit uses the reduced-order basis matrix to perform dimensionality reduction and compression processing on the target reconstructed state field to obtain a low-dimensional representation vector. Based on the domain parameter set, observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual, the generation unit uses... The error calculation method generates a reduced-order state error upper bound. The correction unit, based on the sensitivity upper bound, reduced-order state error upper bound, and low-dimensional representation vector of each operating boundary in the domain parameter set's operating boundary set, corrects each initial operating boundary margin to obtain multiple operating boundary risk components. These risk components are then aggregated to obtain a risk index. The control unit, based on the domain parameter set, target sparse measurement data, and risk index, determines the target equipment's industrial state data in the current control cycle and binds this data to the control commands to be forwarded. This allows the target equipment to be controlled in normal operation during the target industrial process, thus solving the technical problem of accurately controlling equipment operating status and achieving the technical effect of accurate equipment operating status control.

[0244] It should be noted that other corresponding descriptions of the functional units involved in the device operation state control device based on physical constraint inversion operator provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in [the document] will not be repeated here.

[0245] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 13 This application provides a schematic diagram of the device structure of a computer device according to an embodiment of the present application. Figure 13As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0246] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0247] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0248] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0249] It should be noted that the user personal information involved in the embodiments of this application is all authorized (with the knowledge and consent) by the relevant parties or fully authorized by all parties, and the executing entity can obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with the relevant laws and regulations of the relevant countries and regions, and do not violate public order and good morals. It should be noted that if any software tools or components other than those of this company appear in the embodiments of this application, they are merely illustrative examples and do not represent actual use.

[0250] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0251] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0252] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A method for controlling the operating state of equipment based on a physical constraint inversion operator, characterized in that, The method includes: Acquire the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control quantities, and target physical condition data of the target device within the current measurement time window; The target sparse measurement data, the target executed control variables, the target physical condition data, and the reconstructed state field coordinates are input into the target physical constraint inversion operator to obtain the target reconstructed state field. Based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined. Using the reduced-order basis matrix, the target reconstructed state field is subjected to dimensionality reduction and compression to obtain a low-dimensional representation vector. Based on the domain parameter set, the observation residual, the physical residual, the reduced-order projection residual, and the sparse interpolation residual, an upper bound for the reduced-order state error is generated using an error calculation method. For the initial operating boundary margin of each operating boundary in the set of operating boundaries in the domain parameter set, based on the upper bound of sensitivity of each operating boundary, the upper bound of reduced state error, and the low-dimensional representation vector, each initial operating boundary margin is corrected to obtain multiple operating boundary risk components, and the multiple operating boundary risk components are aggregated to obtain a risk index. Based on the domain parameter set, the target sparse measurement data, and the risk index, the industrial status data of the target device in the current control cycle is determined, and the industrial status data is bound to the control command to be forwarded, so as to control the target device to be in normal operation in the target industrial process through the control command to be forwarded.

2. The method according to claim 1, characterized in that, The target physical constraint inversion operator includes: Obtain an initial physical constraint inversion operator and inversion training samples, and construct a training measurement feature vector based on the inversion training samples, wherein the initial physical constraint inversion operator includes at least a backbone network and a branch network; Obtain the observation consistency loss, the control equation residual loss, the initial boundary condition loss, and the state field difference loss, and construct a training loss function based on the observation consistency loss, the control equation residual loss, the initial boundary condition loss, and the state field difference loss; The training measurement feature vector is input into the branch network to obtain the first feature vector, and the spatial coordinates of the target location in the target device and the target time are input into the backbone network to obtain the second feature vector. Based on the first feature vector and the second feature vector, the predicted reconstructed state field is determined. The initial physical constraint inversion operator is trained using the training loss function and the predicted reconstructed state field to obtain the target physical constraint inversion operator.

3. The method according to claim 1, characterized in that, Based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set, the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual are determined, including: The observation residual is determined based on the predicted value at the target measurement location in the target reconstructed state field and the measured value at the target measurement location in the target sparse measurement data. Obtain the physical parameters, and input the target reconstructed state field, the physical parameters, the target executed control variables, and the target physical condition data into the physical residual operator to obtain the physical residual; Based on the target reconstructed state field and the reduced basis matrix, the reduced projection residual and the sparse interpolation residual are determined.

4. The method according to claim 3, characterized in that, Based on the target reconstructed state field and the reduced-order basis matrix, the reduced-order projection residual and the sparse interpolation residual are determined, including: Obtain the state error weighting matrix, and determine the reduced-order projection residual based on the reduced-order basis matrix, the transpose matrix corresponding to the reduced-order basis matrix, the target reconstructed state field, and the state error weighting matrix; Obtain the residual snapshot matrix, perform singular value decomposition on the residual snapshot matrix to obtain the residual basis matrix, and generate the selection matrix based on the residual basis matrix using a matrix column selection algorithm; Obtain the current residual vector, and determine the sparse residual vector based on the residual basis matrix, the transpose matrix corresponding to the selection matrix, and the current residual vector; Obtain the residual error weighting matrix, and determine the sparse interpolation residual based on the current residual vector, the sparse residual vector, and the residual error weighting matrix.

5. The method according to claim 1, characterized in that, The domain parameter set includes at least: a deterministic set of applicable operating conditions, a lower bound for the combined residual stability constant, observation residual error conversion coefficients, physical residual error conversion coefficients, projection residual error conversion coefficients, interpolation residual error conversion coefficients, an upper bound for physical condition disturbance error, and an upper bound for numerical calculation error. Based on the domain parameter set, the observation residuals, the physical residuals, the reduced-order projection residuals, and the sparse interpolation residuals, an error calculation method is used to generate a reduced-order state error upper bound, including: In response to the fact that the error calculation method is a deterministic error calculation method, the current operating condition of the target device is obtained, and it is determined whether the current operating condition belongs to the set of deterministic applicable operating conditions; In response to the fact that the current operating condition belongs to the set of deterministic applicable operating conditions, the upper bound of the deterministic state error is calculated using the following formula: ， in, Used to represent the upper bound of the deterministic state error; Used to represent the observation residual error conversion coefficient; Used to represent the observation residual; Used to represent the physical residual error conversion coefficient; Used to represent the physical residual; Used to represent the lower bound of the combined residual stability constant; Used to represent the conversion coefficient of the projection residual error; Used to represent the reduced-order projection residual; Used to represent the interpolation residual error conversion coefficient; Used to represent the sparse interpolation residual; Used for the upper bound of the physical condition disturbance error; Used to indicate the upper bound of the numerical calculation error; Since the error calculation method is a statistical error calculation method, the upper bound of the statistical state error is calculated using the following formula: ， in, Used to represent the upper bound of the statistical state error; Used to represent the basic error estimation function; Used to represent quantile thresholds; The deterministic state error upper bound or the statistical state error upper bound is converted into a reduced-order state error upper bound.

6. The method according to claim 1, characterized in that, Using the following formula, for each operational boundary in the operational boundary set of the domain parameter set, the initial operational boundary margin is modified based on the upper bound of the sensitivity of each operational boundary, the upper bound of the reduced-order state error, and the low-dimensional representation vector, to obtain multiple operational boundary risk components, including: ， ， in, Used to represent the first in the set of operating boundaries j The initial running boundary margin corresponding to each running boundary; Used to represent the operating boundary margin after the initial operating boundary margin has been corrected. Used to represent the first j The upper bound of sensitivity corresponding to each operational boundary; Used to represent the upper bound of the reduced-order state error; Used to represent the first j Margin scale corresponding to each operating boundary; Used to represent the first j Each operational boundary corresponds to an operational boundary risk component; Used to represent a limiting function; when the first j When the running boundary is the upper limit boundary This can be expressed by the following formula: ， In the above formula, Used to represent boundary thresholds; Used to represent boundary functions, ,and , Used to represent the transpose of the weight vector. Used to represent bias terms; when the first j When each operating boundary is a lower limit boundary This can be expressed by the following formula: ， In the above formula, .

7. The method according to claim 1, characterized in that, The domain parameter set further includes at least: an observation residual hard threshold and a physical residual hard threshold. Based on the domain parameter set, the target sparse measurement data, and the risk index, the industrial status data of the target equipment in the current control cycle is determined, including: The training condition feature vectors of the target device under multiple normal training conditions are obtained, and the current condition feature vectors of the target device under the current condition are obtained under multiple control cycles. The mean value of the multiple training condition feature vectors is calculated to obtain the feature mean vector. Based on the multiple training condition feature vectors and the feature mean vector, a regularized covariance matrix is ​​determined, and based on the multiple current condition feature vectors, the feature mean vector, and the regularized covariance matrix, multiple distribution offset distances are determined. The observation residuals, physical residuals, and effective measurement counts in the target sparse measurement data are obtained under the multiple control cycles. Based on the multiple observation residuals, the multiple physical residuals, the multiple effective measurement counts, and the multiple distribution offset distances, the target anomaly indication is determined under the multiple control cycles. In response to the target anomaly indication being greater than or equal to the anomaly indication threshold, or if any of the following conditions are met: the observation residual is greater than the observation residual hard threshold, the physical residual is greater than the physical residual hard threshold, and the number of valid measurements is 0, failure flag information is determined; The industrial status data is determined based on the target anomaly indication, the distribution offset distance, the failure flag information, and the risk index.

8. The method according to claim 7, characterized in that, The regularized covariance matrix is ​​determined based on the multiple training condition feature vectors and the feature mean vector using the following formula, including: ， in, Used to represent the regularized covariance matrix; Used to indicate the number of training samples; Used to indicate that the target device is in the... i A training condition feature vector under normal training conditions; Used to represent the feature mean vector; Used to represent regularization weights; Used to represent the identity matrix; Multiple distribution offset distances are determined using the following formula, based on the multiple current operating condition feature vectors, the feature mean vector, and the regularized covariance matrix: ， in, Used to indicate the current control cycle t The following distribution offset distance; Used to indicate in the current control cycle t The current operating condition feature vector is described below; The matrix used to represent the inverse of the regularized covariance matrix; The domain parameter set further includes at least: an observation residual threshold, a physical residual threshold, a distribution offset threshold, and an effective measurement quantity threshold. Based on the multiple observation residuals, the multiple physical residuals, the multiple effective measurement quantities, and the multiple distribution offset distances, the target anomaly indication quantity under the multiple control periods is determined using the following formula: ， ， ， in, Used to indicate the current control cycle t The observation residuals described below; Used to represent the observation residual threshold, and the observation residual threshold is less than the observation residual hard threshold; Used to indicate in the current control cycle t The observed residual abnormality indicator quantity; This is used to represent a conditional judgment function. When the condition inside the parentheses is true, it outputs 1; otherwise, it outputs 0. Used to indicate in the current control cycle t The physical residuals described below; Used to represent the physical residual threshold, and the physical residual threshold is less than the physical residual hard threshold; Used to indicate in the current control cycle t The physical residual abnormality indicator; Used to represent the distribution offset threshold; Used to indicate in the current control cycle t The offset distance is an abnormal indicator; Used to indicate in the current control cycle t The number of effective measurements below; Used to indicate the threshold of valid measurement quantity; Used to indicate in the current control cycle t The number of valid measurements is abnormal. Used to indicate in the current control cycle t The target anomaly indicator quantity is as follows.

9. The method according to claim 1, characterized in that, Binding the industrial status data with control commands to be forwarded, so as to control the target equipment to be in normal operation in the target industrial process through the control commands to be forwarded, including: Receive a control instruction to be forwarded generated by the controller based on the industrial status data, wherein the control instruction to be forwarded carries the instruction validity period; Determine whether the current time of generating the industrial status data is within the validity period of the instruction. If the current time is within the validity period of the instruction and the risk index is less than the stop control threshold, forward the control instruction to be forwarded to the target device to control the target device to be in the normal operating state in the target industrial process.

10. A device for controlling the operating state of an equipment based on a physical constraint inversion operator, characterized in that, The device includes: The acquisition unit is used to acquire the domain parameter set and reconstructed state field coordinates of the target device in the target industrial process, as well as the target sparse measurement data, target executed control quantities and target physical condition data of the target device within the current measurement time window; The determining unit is used to input the target sparse measurement data, the target executed control quantity, the target physical condition data, and the reconstructed state field coordinates into the target physical constraint inversion operator to obtain the target reconstructed state field, and to determine the observation residual, physical residual, reduced-order projection residual, and sparse interpolation residual based on the target reconstructed state field, the target sparse measurement data, and the physical residual operator and reduced-order basis matrix in the domain parameter set. The generation unit is used to perform dimensionality reduction and compression processing on the target reconstructed state field using the reduced-order basis matrix to obtain a low-dimensional representation vector, and to generate an upper bound of the reduced-order state error based on the domain parameter set, the observation residual, the physical residual, the reduced-order projection residual, and the sparse interpolation residual using an error calculation method. The correction unit is used to correct the initial running boundary margin of each running boundary in the set of running boundaries in the domain parameter set based on the upper bound of sensitivity of each running boundary, the upper bound of reduced state error, and the low-dimensional representation vector, to obtain multiple running boundary risk components, and to aggregate the multiple running boundary risk components to obtain a risk index. The control unit is used to determine the industrial status data of the target device in the current control cycle based on the domain parameter set, the target sparse measurement data and the risk index, and bind the industrial status data with the control command to be forwarded, so as to control the target device to be in normal operation in the target industrial process through the control command to be forwarded.

Images