Safety early warning system

Abstract

The present invention discloses a safety early warning system, which relates to the field of early warning and control technology, including an acquisition and processing module, an analysis and prediction module and an early warning and control module; the acquisition and processing module measures temperature and pressure in the current cycle, combines the temperature and pressure of the historical cycle, and processes them into a standard data tensor; the analysis and prediction module establishes a matching degree matrix through dynamic time regularization, introduces physical regularity constraints in principal component analysis based on the ideal gas equation, extracts and generates a coupling feature sequence, splices it with the standard data tensor, and then inputs it into a prediction model, outputs the prediction data tensor and fits to generate a trend function of temperature and pressure; the early warning and control module dynamically determines the dangerous moment and dynamically determines the adjustable time, performs model predictive control based on the equipment control model, defines the control target function and solves the optimal control problem within a finite time, obtains the optimal parameter adjustment vector and executes control, and realizes accurate prediction and control considering the temperature, pressure coupling relationship and adjustable time.

Description

Technical Field

[0001] The present invention relates to the technical field of early warning and control, and in particular to a safety early warning system. Background Art

[0002] In industrial production, coupled changes in temperature and pressure in closed reaction vessels, pressure-bearing equipment, and bio-fermentation tanks directly impact production safety, process stability, and product quality. These devices often experience dramatic temperature and pressure fluctuations due to processes such as material reactions, energy conversion, and microbial metabolism. High temperatures and high pressures can cause major safety incidents, such as equipment explosions due to overpressure or uncontrolled material decomposition. Abnormally coupled temperature and pressure fluctuations can disrupt reaction equilibrium, reduce product purity, and even lead to production interruptions.

[0003] Accurately monitoring and predicting temperature and pressure changes in such equipment is key to identifying potential risks in advance and enabling timely intervention to ensure safe and stable production operations. However, existing technologies have significant flaws: First, most prediction techniques fail to fully consider the strong coupling between temperature and pressure in such equipment, ignoring inherent physical connections and leading to significant deviations between predictions and actual results. Second, existing control technologies lack a dynamic definition of the equipment's controllable duration, making it impossible to precisely plan parameter adjustments based on equipment characteristics within a limited timeframe. This can easily lead to control lags or imbalances, making them particularly unsuitable for closed or pressurized environments with fixed reaction cycles and high parameter sensitivity. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention proposes a safety warning system to achieve accurate pre-control by taking into account the coupled changes of temperature and pressure and adaptively determining the adjustable time.

[0005] The technical solutions adopted to achieve the purpose of the present invention are:

[0006] Safety early warning system, including collection and processing module, analysis and prediction module and early warning and control module;

[0007] The acquisition and processing module is in cycle Collection time Measuring temperature and pressure , call cycle Before The temperature and pressure of each cycle are organized into cycles Data tensor And normalized for temperature and pressure respectively, generating the cycle Standard data tensor , is the total number across periods;

[0008] The analysis and prediction module is a cycle set up virtual moments, and calculate the period through dynamic time warping Standard temperature , standard pressure With the Matching degree of virtual moments , establish cycle The matching matrix , based on the ideal gas equation, physical regularity constraints are introduced into principal component analysis to extract the cycle The matching matrix Generation cycle The coupling characteristic sequence and with the cycle Standard data tensor After splicing, input the prediction model, introduce position encoding and output the cycle through attention mechanism and nonlinear transformation The predicted data tensor And fit the generation cycle Temperature trend function and pressure trend function , ;

[0009] Early warning and control module based on cycle Temperature trend function and pressure trend function Determining dangerous moments And dynamically determine the cycle Adjustable duration , perform model predictive control based on the equipment control model to minimize control energy consumption and dangerous moments Temperature exceeds limit and pressure exceeding limit Define the control objective function for the target And solve the optimal control problem within a limited time, and get the cycle The optimal parameter adjustment vector and perform regulation.

[0010] Furthermore, the analysis and prediction module includes a feature construction unit, a feature screening unit, and a prediction unit;

[0011] The feature building block is a cycle set up virtual moments and calculate the standard temperature and standard pressure at each virtual moment based on linear interpolation, using the cycle Standard temperature , standard pressure Perform dynamic time warping with the Euclidean distance of standard temperature and standard pressure at each virtual moment, determine the dynamic weight of each virtual moment and calculate the matching degree of each virtual moment, and establish a cycle The matching matrix ;

[0012] The feature screening unit establishes physical regularization constraints based on the ideal gas equation and regularization expression considering the coupling relationship between pressure and temperature and introduces principal component analysis to extract the periodicity. The coupling characteristic sequence ;

[0013] The prediction unit will cycle Standard data tensor and coupled characteristic sequences Splicing into cycles The combined input tensor of And input the prediction model, use trigonometric function encoding to convert each cycle into a coding vector and add it to the transpose of the joint input vector corresponding to the cycle, use attention mechanism and nonlinear transformation mapping to output the predicted data tensor , using a third-order polynomial model Fitting generation cycle Temperature trend function and pressure trend function .

[0014] Going further, establish a cycle The matching matrix , including the following steps:

[0015] Based on the standard temperature and standard pressure of adjacent periods, the temperature line segment expression and pressure line segment expression between adjacent periods are constructed;

[0016] At the time of collection and collection time Set evenly between virtual moments and substitute each virtual moment into the cycle To cycle The temperature segment expression and pressure segment expression of the period are obtained Inside The standard temperature and standard pressure at a virtual moment, ;

[0017] Based on the previous The mean slope of the temperature line segment expression and the mean slope of the pressure line segment expression for each cycle are respectively Standard temperature With standard pressure Perform linear extrapolation to obtain the period of Standard temperature and standard pressure at a virtual moment;

[0018] Calculate the acquisition time and The Euclidean distance of each virtual moment is summed up to get the Euclidean distance of each cycle, and the first The standard temperature and standard pressure at each virtual moment are regarded as virtual coordinates, and the standard temperature and standard pressure at the collection moment of each cycle are regarded as collection coordinates. The collection coordinates in each cycle are calculated with the standard temperature and standard pressure at the virtual moment. The Euclidean distance of the virtual coordinates of the virtual moment is divided by the sum of the Euclidean distances of each cycle to generate the first The dynamic weight of each virtual moment;

[0019] The acquisition time in each cycle is The Euclidean distance of each virtual moment is multiplied by the corresponding dynamic weight to obtain the The matching degree of virtual moments is arranged in each cycle The matching degree of each virtual moment is used to obtain the matching degree vector of each cycle, and the cycle To cycle The matching degree vectors are concatenated to obtain the period The matching matrix .

[0020] Furthermore, the generation cycle The coupling characteristic sequence , including the following specific steps:

[0021] Based on the ideal gas equation, pressure and temperature There is a linear positive correlation between pressure Multiply by volume Equal to the amount of substance , gas constant and temperature The product of

[0022] The cycle The matching matrix Perform Z-Score normalization and calculate cycle The covariance matrix of , based on eigenvalue decomposition, period The covariance matrix of Can be equivalent to a cycle The eigenvector matrix of and the eigenvalue diagonal matrix The goal of principal component analysis is to minimize the period The covariance matrix of With cycle The eigenvector matrix of and the eigenvalue diagonal matrix The norm error of the quadratic form, where the period The eigenvector matrix of and the eigenvalue diagonal matrix The quadratic form of The eigenvector matrix of The transpose of multiplied by the eigenvalue diagonal matrix Multiply by the eigenvector matrix ;

[0023] The cycle To cycle The standard pressure and standard temperature are arranged to generate the cycle The pressure diagonal matrix and the temperature diagonal matrix , set physical constraints to minimize the cycle The eigenvector matrix of and the pressure diagonal matrix The quadratic form minus the period The eigenvector matrix of Diagonal matrix with temperature The norm of the quadratic form of , multiplying the physical constraint by the regularization coefficient and superimposed with the objectives of principal component analysis;

[0024] Using alternating optimization algorithm, initialization cycle The eigenvector matrix of Diagonal matrix with eigenvalues , in each round of iteration, fix the cycle obtained in the previous round The eigenvalue diagonal matrix of , update the cycle of this round according to gradient descent The eigenvector matrix of , fix the update cycle of this round The eigenvector matrix of And update the cycle of this round by gradient descent The eigenvalue diagonal matrix of , repeat the iteration until the alternating optimization algorithm converges and stops, obtaining the final cycle The eigenvector matrix of Diagonal matrix with eigenvalues ;

[0025] From the cycle The eigenvector matrix of Extraction cycle The eigenvalue diagonal matrix of middle The eigenvalues ​​corresponding to the eigenvectors are extracted The first eigenvector Dimension and splice into cycles The coupling characteristic vector of , integrated to get the cycle The coupling characteristic sequence .

[0026] Furthermore, the processing steps of the prediction model include:

[0027] Use triangle encoding to convert the period Convert to cycle The encoding vector , the cycle The encoding vector Middle code elements and the joint input vector In the transpose of Add the elements of the dimensions to generate the cycle The embedded input vector , integrated to get the cycle The embedded input sequence ,in, ;

[0028] Using attention mechanism to capture cycles The combined input sequence The temporal dependency between different cycles and the dimensional dependency between different dimensions of the embedded input vector of the same cycle generate spatiotemporal correlation features. ;

[0029] Temporal and spatial correlation features Input feedforward network for nonlinear transformation and feature enhancement to generate enhanced correlation features , which will enhance the associated features Adjust the dimension through linear weights and linear biases, and map it to the predicted data tensor through the ReLU function .

[0030] Furthermore, the temperature error function is defined as For cycle The predicted temperature vector and cycle Temperature With the third-order polynomial model The total fitting error and pressure error function are defined as For cycle The predicted pressure vector and cycle Pressure With the third-order polynomial model The total error of the fitting is summed, and the nonlinear least squares method is used for fitting to initialize the unknown parameter vector , in each round of iteration, according to the temperature error function and pressure error function The negative gradient of each parameter vector is updated , until the temperature error function and pressure error function Convergence, get the cycle The optimal temperature parameter vector and the optimal pressure parameter vector And substitute into the third-order polynomial model , generation cycle Temperature trend function and pressure trend function .

[0031] Furthermore, determine the cycle Adjustable duration , including the following steps:

[0032] Cycle-based Temperature trend function and pressure trend function , judging from the cycle To cycle Is there a temperature between the upper and lower thresholds? and / or upper pressure threshold moment;

[0033] If it does not exist, no processing will be done;

[0034] If it exists, it will first be equal to the upper temperature threshold and / or upper pressure threshold Moments of danger , judge the dangerous moment Is it in cycle Inside;

[0035] If dangerous moments In the cycle Internal cycle Adjustable duration Equal to dangerous moments Reduce cycle Collection time , if in the cycle Afterwards, the cycle Adjustable duration Equal to the period Collection time Reduce cycle Collection time .

[0036] Specifically, the equipment control model includes the temperature control equipment control model and the pressure control equipment control model. The temperature control equipment control model is specifically the temperature change With adjustable duration The ratio is equal to the temperature control parameter adjustment The linear transformation of the pressure control equipment control model is specifically the pressure change With adjustable duration The ratio is equal to the pressure control parameter adjustment The unknown parameters of the linear transformation are determined by least squares fitting. The control objective function Defined as a dangerous moment Temperature exceeds limit , Pressure exceeds limit , temperature control energy consumption Energy consumption with pressure regulation The weighted sum of Temperature exceeds limit Equal to 0 and dangerous moments Corrected temperature and upper temperature threshold The larger the difference, the more dangerous it is. Pressure exceeds limit Equal to 0 and dangerous moments Corrected pressure and pressure upper threshold The larger the difference, the greater the energy consumption of temperature control. Energy consumption with pressure regulation Equal to the corresponding parameter adjustment amount multiplied by the adjustable duration .

[0037] Specifically, based on the equipment control model, the cycle Temperature trend function and pressure trend function Make adjustments and corrections;

[0038] When danger strikes Located in the cycle Inside, dangerous moment Corrected temperature With corrected pressure Separation is a dangerous moment Temperature trend function value Add the cycle determined by the equipment control model Temperature control parameter adjustment and dangerous moments Pressure trend function value Add the cycle determined by the equipment control model Pressure control parameter adjustment ;

[0039] When danger strikes Located in the cycle Afterwards, according to the cycle Collection time Corrected temperature and temperature trend function value The difference in dangerous moments Temperature trend function value Linear extrapolation to obtain the critical moment Corrected temperature , according to the cycle Collection time Corrected pressure and pressure trend function value The difference in dangerous moments Pressure trend function value Linear extrapolation to obtain the critical moment Corrected pressure ;

[0040] To minimize dangerous moments The control objective function For the optimization goal, set the optimization constraint to the period Collection time To the dangerous moment The corrected temperature and corrected pressure at any moment in the cycle are within the corresponding temperature range and pressure range. Temperature control parameter adjustment Adjustment of pressure control parameters Within the corresponding temperature control range and pressure control range, the optimization constraints are introduced into the optimization objective through the Lagrange multiplier method to construct the Lagrange augmented function and the KKT condition is used to solve the period Optimal temperature control parameter adjustment Adjustment of pressure control parameters , spliced ​​into cycles The optimal parameter adjustment vector .

[0041] Compared with the prior art, the present invention sets multiple virtual moments for each cycle to expand the amount of time series data, calculates the matching degree between the standard temperature and standard pressure of each cycle and each virtual moment in the cycle through dynamic time warping, establishes a matching degree matrix for the current cycle, introduces physical regularity constraints in principal component analysis based on the ideal gas equation, refines the matching degree matrix of the current cycle to generate a coupling feature sequence of the current cycle, and splices it with the standard data tensor of the current cycle before inputting it into the prediction model. Position encoding is introduced and the predicted data tensor of the current cycle is output through an attention mechanism and nonlinear transformation, and the temperature trend function and pressure trend function of the current cycle are fitted to achieve accurate prediction considering the temperature and pressure coupling relationship. The early warning and control module determines the dangerous moment based on the temperature trend function and pressure trend function of the current cycle and dynamically determines the adjustable duration of the current cycle to avoid the collection, analysis and prediction of the cycle under the influence of subsequent control. Model predictive control is performed based on the equipment control model. The control objective function is defined with the goal of minimizing the control energy consumption and the temperature and pressure limit values ​​at the dangerous moment, and the optimal control problem within a limited time is solved to obtain the optimal parameter adjustment vector for the current cycle and execute control. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a schematic diagram of the safety early warning system of the present invention;

[0043] Figure 2 This is a flow chart of the alternating optimization algorithm of the present invention;

[0044] Figure 3 is a processing flow chart of the prediction model of the present invention;

[0045] Figure 4 A flow chart for determining an adjustable duration for the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0047] Example 1

[0048] like Figure 1 As shown, a specific embodiment of the present invention discloses a safety early warning system, including a collection and processing module, an analysis and prediction module, and an early warning and control module;

[0049] The acquisition and processing module is in cycle Collection time Measuring the temperature inside the device and pressure , select the cycle and before The temperature and pressure of each cycle are connected in rows and arranged in order as cycles. Data tensor X ( i ) = [ T i − K + k ; P i − K + k ] k = 1 K , normalize the minimum and maximum values ​​for temperature and pressure respectively, and generate the cycle Standard data tensor X ˜ ( i ) = [ T ˜ i − K + k ; P ˜ i − K + k ] k = 1 K ,in, and The cycles Data tensor Middle Temperature and pressure of the cycle, and The cycles Standard data tensor Middle Standard temperature and standard pressure of the cycle, normalized by minimum and maximum The temperature and pressure are obtained, is the total number across periods;

[0050] The analysis and prediction module is a cycle set up virtual moments, and calculate the period through dynamic time warping Standard temperature , standard pressure With the Matching degree of virtual moments , establish cycle The matching matrix Based on the ideal gas equation, physical regularity constraints are introduced into principal component analysis to analyze the periodic The matching matrix Perform screening and extraction to generate cycles The coupling characteristic sequence and with the cycle Standard data tensor After splicing by row, input the prediction model, perform position encoding based on the period order and use the attention mechanism to capture the dependency between the period dimension and the feature dimension, and output the period through nonlinear transformation. The predicted data tensor X ^ ( i ) = [ T ^ i + k ; P ^ i + k ] k = 1 K And fit the generation cycle Temperature trend function and pressure trend function ,in, and The cycles The predicted data tensor Middle The predicted temperature and pressure of the cycle, i.e. the cycle The predicted temperature and predicted pressure, For cycle Start to cycle At any moment in the end, ;

[0051] Early warning and control module based on cycle Temperature trend function and pressure trend function Determining dangerous moments , based on dangerous moments The cycle is dynamically determined Adjustable duration , perform model predictive control based on the equipment control model to minimize control energy consumption and dangerous moments Temperature exceeds limit and pressure exceeding limit Define the control objective function for the target , combined with the cycle Temperature trend function and pressure trend function Solve the optimal control problem within a finite time period and obtain the period The optimal parameter adjustment vector And perform control, where the equipment control model is used to describe the temperature control parameter adjustment , pressure control parameter adjustment and temperature change , pressure change relationship.

[0052] Furthermore, the analysis and prediction module includes a feature construction unit, a feature screening unit, and a prediction unit;

[0053] The feature building block is a cycle set up virtual moments and calculate the standard temperature and standard pressure at each virtual moment based on linear interpolation, using the cycle Standard temperature , standard pressure Perform dynamic time warping with the Euclidean distance of standard temperature and standard pressure at each virtual moment, determine the dynamic weight of each virtual moment and calculate the matching degree of each virtual moment, and establish a cycle The matching matrix ;

[0054] The feature screening unit considers the period based on the ideal gas equation Standard pressure With cycle Standard temperature The coupling relationship is established by regularizing the expression and establishing physical regular constraints. The traditional principal component analysis is introduced to make the feature extraction consistent with the coupling change correlation between pressure and temperature, and the period is extracted. The coupling characteristic sequence γ ( i ) = [ γ i − K + k ] k = 1 K ,in, For cycle The coupling characteristic vector of Middle The coupling characteristic vector of the period, that is, the period The coupling characteristic vector of

[0055] The prediction unit will cycle Standard data tensor and coupled characteristic sequences Splice rows into cycles The combined input tensor of And input the prediction model, use trigonometric function encoding to convert each cycle into a coding vector and add it to the transpose of the joint input vector corresponding to the cycle, use the attention mechanism to capture the dependency between the cycle dimension and the feature dimension, and output the predicted data tensor after nonlinear transformation. , using a third-order polynomial model Fitting generation cycle Temperature trend function and pressure trend function .

[0056] Going further, establish a cycle The matching matrix , including the following steps:

[0057] For the cycle To cycle Every two adjacent cycles in the period Collection time With cycle Collection time As the domain, with period Standard temperature With cycle Standard temperature is the temperature range, with period Standard pressure With cycle Standard pressure For the pressure range, construct the cycle To cycle The temperature segment expression and pressure segment expression of ;

[0058] At the time of collection and collection time Set evenly between virtual moments and substitute each virtual moment into the cycle To cycle The temperature segment expression and pressure segment expression of the period are obtained Inside Standard temperature and standard pressure at a virtual moment;

[0059] Based on the previous The mean slope of the temperature line segment expression and the mean slope of the pressure line segment expression for each cycle are respectively Standard temperature With standard pressure Perform linear extrapolation to obtain the period of Standard temperature and standard pressure at a virtual moment;

[0060] Calculate the acquisition time and The Euclidean distance of each virtual moment is summed up to get the Euclidean distance of each cycle, and the first The standard temperature and standard pressure at each virtual moment are regarded as virtual coordinates, and the standard temperature and standard pressure at the collection moment of each cycle are regarded as collection coordinates. The collection coordinates in each cycle are calculated with the standard temperature and standard pressure at the virtual moment. The Euclidean distance of the virtual coordinates of the virtual moment is divided by the sum of the Euclidean distances of each cycle to generate the first The dynamic weight of each virtual moment;

[0061] The acquisition time in each cycle is The Euclidean distance of each virtual moment is multiplied by the corresponding dynamic weight to obtain the The matching degree of each virtual moment is The matching degrees of virtual moments are spliced ​​in columns to obtain the matching degree vector of each cycle. To cycle The matching degree vectors are concatenated row by row to obtain Dimensional Cycle The matching matrix , where the period The matching matrix The Rank Column matching That is the cycle Neidi The matching degree of a virtual moment, Through dynamic time warping, the standard temperature and standard pressure of the original collection time in each cycle are expanded in time series. Without adding sensors, it is beneficial for the prediction unit to capture the correlation of the period dimension more fully.

[0062] Furthermore, the generation cycle The coupling characteristic sequence , including the following specific steps:

[0063] Based on the ideal gas equation, pressure and temperature There is a linear positive correlation between them, as follows:

[0064] ;

[0065] in, 、 and They are volume, amount of substance and gas constant respectively, which are all fixed values ​​in a certain equipment of a certain process;

[0066] For traditional principal component analysis, the period The matching matrix Perform Z-Score standardization to obtain the cycle The standard matching matrix And calculate the cycle The covariance matrix of , as follows:

[0067] Σ ( i ) = 1 K [ D ˜ ( i ) ] T D ˜ ( i ) ;

[0068] in, [ ] T Represents the transpose of a matrix or vector;

[0069] Based on eigenvalue decomposition, the period The covariance matrix of Can be equivalent to a cycle The eigenvector matrix of and the eigenvalue diagonal matrix Therefore, the goal of traditional principal component analysis is to find a set of cycles The eigenvector matrix of and the eigenvalue diagonal matrix , so that the cycle The covariance matrix of With cycle The eigenvector matrix of and the eigenvalue diagonal matrix The norm error of the quadratic form is the smallest, as follows:

[0070] min U ( i ) , Λ ( i ) ‖ Σ ( i ) − [ U ( i ) ] T Λ ( i ) U ( i ) ‖ F 2 ;

[0071] in, represents the F norm of a matrix or vector, [ U ( i ) ] T Λ ( i ) U ( i ) For cycle The eigenvector matrix of and the eigenvalue diagonal matrix The quadratic form of

[0072] Refer to the ideal gas equation, pressure and temperature There is a linear correlation between them. When the minimum and maximum normalization are unified to [ 0 , 1 ] Within the range, standard pressure and standard temperature are approximately equal, and the period To cycle The standard pressure and standard temperature are arranged in a periodic order in the dimension The diagonal of the diagonal matrix generates the cycle The pressure diagonal matrix and the temperature diagonal matrix , the period obtained by principal component analysis The eigenvector matrix of In the processing cycle The pressure diagonal matrix Diagonal matrix with temperature Should meet the standard pressure and standard temperature The relationship of approximate equality, that is, the period The eigenvector matrix of and the pressure diagonal matrix The quadratic form minus the period The eigenvector matrix of Diagonal matrix with temperature The norm of the quadratic form should be as small as possible, and the physical constraints are as follows:

[0073] min U ( i ) ‖ [ U ( i ) ] T diag P ( i ) U ( i ) − [ U ( i ) ] T diag T ( i ) U ( i ) ‖ F 2 ;

[0074] Multiply the physical constraint by the regularization coefficient And superimposed with the objectives of traditional principal component analysis, we get the objectives of principal component analysis with physical regularity constraints.

[0075] like Figure 2 As shown, the alternating optimization algorithm is used to solve the problem, and the initialization cycle The eigenvector matrix of Diagonal matrix with eigenvalues , in each round of iteration, fix the cycle obtained in the previous round The eigenvalue diagonal matrix of , update the cycle of this round according to gradient descent The eigenvector matrix of , fix the update cycle of this round The eigenvector matrix of And update the cycle of this round again through gradient descent The eigenvalue diagonal matrix of , repeat the iteration until the alternating optimization algorithm converges and stops, obtaining the final cycle The eigenvector matrix of Diagonal matrix with eigenvalues ;

[0076] Based on cycle The eigenvalue diagonal matrix of middle The larger eigenvalues ​​from the period The eigenvector matrix of Corresponding extraction The dimensions are Extract the feature vector of The first eigenvector Dimension and concatenate row by row into dimensions equal to Cycle The coupling characteristic vector of , according to the order of the period, we can get the period The coupling characteristic sequence .

[0077] like Figure 3 As shown, further, the processing steps of the prediction model include:

[0078] For the cycle The joint input vector y i − K + k = [ T ˜ i − K + k ; P ˜ i − K + k ; γ i − K + k ] , using triangle coding to convert the period Converted to the combined input vector The transpose of the period of the same dimension The encoding vector z i − K + k = [ z i − K + k , 1 , ⋯ , z i − K + k , 2 + M 0 ] ,in, For cycle The encoding vector Middle Code elements, , the specific calculation formula is as follows:

[0079] ;

[0080] in, and Respectively represent Take odd and even numbers, ;

[0081] The cycle The encoding vector Middle code elements and the joint input vector In the transpose of Add the elements of the dimensions to generate the cycle The embedded input vector ,in, , the cycle To cycle The embedded input vector is concatenated into periods according to the periodic order. The embedded input sequence ;

[0082] Using attention mechanism to capture cycles The combined input sequence The temporal dependency between different cycles and the dimensional dependency between different dimensions of the embedded input vector of the same cycle generate spatiotemporal correlation features. , as follows:

[0083] ;

[0084] in, 、 and The cycles The embedded input sequence Cycles generated by different linear transformations The query matrix, key matrix and value matrix of and are the matrix dimension and Softmax function respectively;

[0085] Temporal and spatial correlation features Input the feedforward network in the existing Transformer architecture for nonlinear transformation and feature enhancement to generate enhanced correlation features , as follows:

[0086] ;

[0087] in, To take the largest function, used to convert the matrix All elements less than or equal to 0 are replaced with 0, and elements greater than 0 are retained. and are the first feedforward weight and the second feedforward weight, and are the first feedforward bias and the second feedforward bias respectively;

[0088] Will enhance the associated features Again, the dimension is adjusted by linear weight and linear bias to , mapped to a cycle through the ReLU function The predicted data tensor .

[0089] Furthermore, the third-order polynomial model The basic expressions are as follows:

[0090] ;

[0091] in, 、 、 and They are the constant term, first-order coefficient, second-order coefficient and third-order coefficient, based on the period The predicted temperature vector T ^ ( i ) = [ T ^ i + k ] k = 1 K With cycle Measured temperature ,cycle The predicted pressure vector P ^ ( i ) = [ P ^ i + k ] k = 1 K With cycle Measured pressure Define the temperature error function separately and pressure error function , as follows:

[0092] e T ( a ) = ∑ k = 0 K [ T ^ i + k − f ( t i + k ) ] 2 ;

[0093] e P ( a ) = ∑ k = 0 K [ P ^ i + k − f ( t i + k ) ] 2 ;

[0094] in, a = [ a 0 , a 1 , a 2 , a 3 ] is the unknown parameter vector, For cycle The acquisition time is to minimize the temperature error function and pressure error function As the target, nonlinear least squares method is used as the fitting method and the unknown parameter vectors are initialized respectively. , in each round of iteration, according to the temperature error function and pressure error function About the Undetermined Parameter Vector Update the undetermined parameter vector in the negative gradient direction , until the temperature error function and pressure error function When converged, we get the period The optimal temperature parameter vector and the optimal pressure parameter vector And reverse substitution into the third-order polynomial model , fitting generation cycle Temperature trend function and pressure trend function .

[0095] like Figure 4 As shown, further, determine the cycle Adjustable duration , including the following steps:

[0096] Cycle-based Temperature trend function and pressure trend function , judging from the cycle To cycle Is there a temperature between the upper and lower thresholds? and / or upper pressure threshold moment;

[0097] If it does not exist, no processing is done and the waiting period is The arrival of

[0098] If it exists, it will first be equal to the upper temperature threshold and / or upper pressure threshold Moments of danger , judge the dangerous moment Is it in cycle Inside;

[0099] If dangerous moments In the cycle Within, the cycle Adjustable duration Equal to the period Collection time To the dangerous moment The duration of the experience, if the dangerous moment In the cycle After that, the cycle Adjustable duration Equal to the period Collection time To cycle Collection time Duration of experience, period Adjustable duration The determination can always ensure that the equipment control is in the cycle within the cycle Affects the cycle when it arrives measurement, analysis and prediction of

[0100] Specifically, the equipment control model includes a method for describing the temperature control parameter adjustment and temperature change The temperature control equipment control model based on the relationship between the two and the pressure control parameter adjustment and pressure change The pressure control equipment control model of the relationship between them is as follows:

[0101] ;

[0102] ;

[0103] in, To adjust the duration, and are the temperature control calibration coefficient and temperature control correction value respectively, and They are respectively the pressure control calibration coefficient and the pressure control correction value, which are determined by fitting the historical operation data and operation effect of the temperature control equipment and the pressure control equipment through the least square method. The control objective function Defined as a dangerous moment Temperature exceeds limit , Pressure exceeds limit , temperature control energy consumption Energy consumption with pressure regulation The weighted sum of is as follows:

[0104] ;

[0105] in, and The dangerous moments after regulation The corrected temperature and corrected pressure are only valid if the dangerous moment Corrected temperature Greater than the upper temperature threshold When the temperature exceeds the limit Equal to the temperature excess, if and only if the dangerous moment Corrected pressure Greater than the upper pressure threshold When the pressure exceeds the limit Equal to the excess pressure, 、 、 and They are respectively the preset first balance coefficient, second balance coefficient, third balance coefficient and fourth balance coefficient.

[0106] Specifically, the model predictive control is performed based on the equipment control model to Temperature trend function and pressure trend function Make adjustments and corrections;

[0107] When danger strikes Located in the cycle In time, after a period of Adjustable duration It's a dangerous moment , then dangerous moments Corrected temperature With corrected pressure The specific formula is as follows:

[0108] ;

[0109] ;

[0110] in, and The cycles The temperature control parameter adjustment amount and pressure control parameter adjustment amount;

[0111] When danger strikes Located in the cycle Afterwards, after a period of Adjustable duration That is the cycle Collection time , then dangerous moments Corrected temperature With corrected pressure The specific formula is as follows:

[0112] ;

[0113] ;

[0114] in, and They are the cycles adjusted by the equipment control model Collection time The corrected temperature and corrected pressure, while in the cycle Collection time To the dangerous moment Since the temperature control equipment and pressure control equipment are not adjusted at this time, the temperature and pressure are based on the cycle Adjustable duration The original change trend within the system continues to change;

[0115] To minimize dangerous moments The control objective function To optimize the target, set the optimization constraint to Collection time To the dangerous moment The corrected temperature and corrected pressure at any moment in the range are both at the lower temperature threshold. and upper temperature threshold In-range and lower pressure thresholds and pressure upper threshold Within the range and period Temperature control parameter adjustment Adjustment of pressure control parameters Located within the corresponding temperature control range and pressure control range, the optimization constraints are introduced into the optimization objective through the Lagrange multiplier method to construct an unconstrained Lagrange augmented function. Since the Lagrange augmented function is a convex optimization problem, the period can be obtained by directly solving it using the KKT condition. Optimal temperature control parameter adjustment Adjustment of pressure control parameters And splice by column into a cycle The optimal parameter adjustment vector .

[0116] Example 2

[0117] As an example, a specific embodiment of the present invention discloses the working steps of applying the safety warning system to a fluorination reactor, including:

[0118] In the cycle Collection time The temperature sensor and pressure sensor deployed in the reactor are controlled by the Modbus RTU half-duplex communication protocol to measure, and the temperature is obtained through the CANopen bus transmission and pressure ;

[0119] Combination cycle Previous continuous The temperature and pressure of the cycle are sorted and normalized to the cycle Standard data tensor , is the period To cycle Each cycle setting between Virtual moments, creating cycles through dynamic time warping The matching matrix ;

[0120] Based on the ideal gas equation, physical regularity constraints are introduced into principal component analysis to screen the generation cycle. The coupling characteristic sequence and with the cycle Standard data tensor After splicing, input the prediction model and output the cycle The predicted data tensor And fit the generation cycle Temperature trend function and pressure trend function ;

[0121] Based on cycle Temperature trend function and pressure trend function Determining the dangerous moment of the reactor And dynamically determine the cycle Adjustable duration ;

[0122] Model predictive control based on the reactor control model to minimize control energy consumption and dangerous moments The temperature and pressure limit values ​​are set as the target to define the control objective function, combined with the acquisition time Measured temperature and pressure and cycle Temperature trend function and pressure trend function Solve the optimal control problem within a finite time period and obtain the period The power adjustment of the temperature control equipment and the opening adjustment of the pressure relief valve;

[0123] The period is converted into The power adjustment and opening adjustment are converted into control instructions recognized by the temperature control equipment and the pressure relief valve, and are transmitted by the PLC to the temperature control equipment and the pressure relief valve through the PROFINET industrial Ethernet protocol to perform corresponding regulation.

[0124] Example 3

[0125] As an example, a specific embodiment of the present invention discloses the working steps of applying the safety warning system to a steam boiler, including:

[0126] In the cycle Collection time The temperature sensor and pressure sensor deployed in the boiler gas drum are controlled by the Modbus RTU half-duplex communication protocol to measure the internal steam temperature. and pressure ;

[0127] Combination cycle Previous continuous The temperature and pressure of the steam of the cycle are sorted and normalized to the cycle Standard data tensor , is the period To cycle Each cycle setting between Virtual moments, creating cycles through dynamic time warping The matching matrix ;

[0128] Based on the ideal gas equation, physical regularity constraints are introduced into principal component analysis to screen the generation cycle. The coupling characteristic sequence and with the cycle Standard data tensor After splicing, input the prediction model and output the cycle The predicted data tensor And fit the generation cycle The temperature trend function of steam and pressure trend function ;

[0129] Based on cycle Temperature trend function and pressure trend function Determining the dangerous moment of the reactor And dynamically determine the cycle Adjustable duration ;

[0130] Model predictive control based on the steam boiler control model to minimize control energy consumption (fuel consumption) and dangerous moments The temperature and pressure limit values ​​are set as the target to define the control objective function, combined with the acquisition time Measured temperature and pressure and cycle Temperature trend function and pressure trend function Solve the optimal control problem within a finite time period and obtain the period The power adjustment of the coal feeder and the opening adjustment of the steam control valve;

[0131] The period is converted into The power adjustment amount and opening adjustment amount are converted into control instructions recognized by the coal feeder and steam regulating valve, and transmitted to the coal feeder and steam regulating valve by the PLC through the PROFINET industrial Ethernet protocol to perform corresponding regulation.

[0132] Example 4

[0133] As an example, a specific embodiment of the present invention discloses the working steps of applying the safety warning system to a beer fermentation tank, including:

[0134] In the cycle Collection time The temperature sensor and pressure sensor deployed in the gas phase space of the fermenter are controlled by the Modbus RTU half-duplex communication protocol to measure the internal steam temperature. (gas phase temperature) and pressure (carbon dioxide pressure);

[0135] Combination cycle Previous continuous The temperature and pressure of the cycle are sorted and normalized to the cycle Standard data tensor , is the period To cycle Each cycle setting between Virtual moments, creating cycles through dynamic time warping The matching matrix ;

[0136] Based on the ideal gas equation, physical regularity constraints are introduced into principal component analysis to screen the generation cycle. The coupling characteristic sequence and with the cycle Standard data tensor After splicing, input the prediction model and output the cycle The predicted data tensor And fit the generation cycle Temperature trend function and pressure trend function ;

[0137] Based on cycle Temperature trend function and pressure trend function Determining the dangerous moment of the reactor And dynamically determine the cycle Adjustable duration ;

[0138] Model predictive control based on the fermentation tank control model to minimize control energy consumption (fuel consumption) and dangerous moments The temperature and pressure limit values ​​are set as the target to define the control objective function, combined with the acquisition time Measured temperature and pressure and cycle Temperature trend function and pressure trend function Solve the optimal control problem within a finite time period and obtain the period The power adjustment of the cooling equipment and the opening adjustment of the exhaust valve;

[0139] The period is converted into The power adjustment and opening adjustment are converted into control instructions recognized by the cooling equipment and exhaust valves, and transmitted by the PLC to the cooling equipment and exhaust valves through the PROFINET industrial Ethernet protocol to perform corresponding regulation.

[0140] The present invention discloses a safety early warning system, including an acquisition and processing module, an analysis and prediction module, and an early warning and control module; the acquisition and processing module measures temperature and pressure in the current cycle, calls the temperature and pressure of multiple cycles before the current cycle, and organizes and normalizes them into the current standard data tensor; the analysis and prediction module sets multiple virtual moments for each cycle to expand the amount of time series data, calculates the matching degree of the standard temperature and standard pressure of each cycle with each virtual moment in the cycle through dynamic time regularization, establishes the matching degree matrix of the current cycle, introduces physical regularity constraints in principal component analysis based on the ideal gas equation, refines the matching degree matrix of the current cycle, generates the coupling feature sequence of the current cycle, and splices it with the standard data tensor of the current cycle before inputting it The prediction model introduces position encoding and outputs the predicted data tensor of the current cycle through attention mechanism and nonlinear transformation, and fits to generate the temperature trend function and pressure trend function of the current cycle, so as to achieve accurate prediction considering the coupling relationship between temperature and pressure; the early warning and control module determines the dangerous moment based on the temperature trend function and pressure trend function of the current cycle and dynamically determines the controllable duration of the current cycle to avoid the collection, analysis and prediction of the cycle affected by subsequent control, and performs model predictive control based on the equipment control model, with the goal of minimizing the control energy consumption and the temperature and pressure limit values ​​at the dangerous moment to define the control objective function and solve the optimal control problem within a limited time, obtain the optimal parameter adjustment vector of the current cycle and execute control.

[0141] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. Safety early warning system, applicable to industrial production scenarios including closed reaction vessels, pressure equipment and biological fermentation tanks, characterized by: It includes collection and processing module, analysis and prediction module and early warning and control module; The acquisition and processing module is in a periodic Measure temperature and pressure, call cycle Before The temperature and pressure of each cycle are sorted and normalized into cycles The standard data tensor of is the total number across periods; The analysis and prediction module sets The standard temperature, standard pressure and the first virtual moment in each cycle are calculated by dynamic time warping. The matching degree of virtual moments, the establishment cycle The matching matrix of the ideal gas equation is used to introduce physical regularity constraints in principal component analysis to extract the cycle The matching matrix generation cycle The coupling characteristic sequence and the period After the standard data tensor is spliced ​​and input into the prediction model, position encoding is introduced and the cycle is output through attention mechanism and nonlinear transformation. The predicted data tensor and fitting generation cycle Temperature trend function and pressure trend function; The early warning control module is based on the cycle The temperature trend function and pressure trend function are used to determine the dangerous moment and dynamically determine the cycle The controllable time is determined by model predictive control based on the equipment control model. The control objective function is defined to minimize the control energy consumption and the temperature and pressure limit values ​​at the dangerous moment. The optimal control problem within a limited time is solved to obtain the cycle. The optimal parameter adjustment vector is obtained and control is performed.

2. The safety warning system according to claim 1, characterized in that: The analysis and prediction module includes a feature construction unit; The feature construction unit sets Virtual moments are calculated based on linear interpolation to calculate the standard temperature and standard pressure of each virtual moment. The Euclidean distance between the standard temperature and standard pressure of each cycle and the standard temperature and standard pressure of each virtual moment is used for dynamic time warping. The dynamic weight of each virtual moment is determined and the matching degree of each virtual moment is calculated to establish a cycle. The matching matrix.

3. The safety warning system according to claim 2, characterized in that: Establishment cycle The matching matrix includes the following steps: Based on the standard temperature and standard pressure of adjacent periods, the temperature line segment expression and pressure line segment expression between adjacent periods are constructed; At the time of collection and collection time Set evenly between virtual moments and substitute each virtual moment into the cycle To cycle The temperature segment expression and pressure segment expression of the period are obtained Inside The standard temperature and standard pressure at a virtual moment, ; Based on the previous The mean slope of the temperature line segment expression and the mean slope of the pressure line segment expression for each cycle are respectively The standard temperature and standard pressure are linearly extrapolated to obtain the period of Standard temperature and standard pressure at a virtual moment; Calculate the acquisition time and The Euclidean distance of each virtual moment is summed up to get the Euclidean distance of each cycle, and the first The standard temperature and standard pressure at each virtual moment are regarded as virtual coordinates, and the standard temperature and standard pressure at the collection moment of each cycle are regarded as collection coordinates. The collection coordinates in each cycle are calculated with the standard temperature and standard pressure at the virtual moment. The Euclidean distance of the virtual coordinates of the virtual moment is divided by the sum of the Euclidean distances of each cycle to generate the first The dynamic weight of each virtual moment; The acquisition time in each cycle is The Euclidean distance of each virtual moment is multiplied by the corresponding dynamic weight to obtain the The matching degree of virtual moments is arranged in each cycle The matching degree of each virtual moment is used to obtain the matching degree vector of each cycle, and the cycle To cycle The matching degree vectors are concatenated to obtain the period The matching matrix.

4. The safety warning system according to claim 1, characterized in that: The analysis and prediction module includes a feature screening unit and a prediction unit; The feature screening unit establishes physical regularization constraints based on the ideal gas equation and regularization expression considering the coupling relationship between pressure and temperature and introduces principal component analysis to extract the period The coupling characteristic sequence of The prediction unit will periodically The standard data tensor and the coupled feature sequence are spliced ​​into a period The joint input tensor is input into the prediction model, and trigonometric function encoding is used to convert each cycle into a coding vector and add the transpose of the joint input vector corresponding to the cycle. The attention mechanism and nonlinear transformation mapping are used to output the predicted data tensor, and the third-order polynomial model is used to fit the generated cycle. Temperature trend function and pressure trend function.

5. The safety warning system according to claim 4, characterized in that: Generation cycle The coupling characteristic sequence includes the following specific steps: Based on the ideal gas equation, pressure times volume equals the product of the amount of substance, the gas constant, and temperature; The cycle The matching matrix is ​​Z-Score standardized and the cycle is calculated The covariance matrix of , based on eigenvalue decomposition, period The covariance matrix of can be equivalent to the period The quadratic form of the eigenvector matrix and the eigenvalue diagonal matrix of The covariance matrix and period of The norm error of the quadratic form of the eigenvector matrix and the eigenvalue diagonal matrix of The quadratic form of the eigenvector matrix and the eigenvalue diagonal matrix is ​​equal to the period The transpose of the eigenvector matrix multiplied by the eigenvalue diagonal matrix multiplied by the eigenvector matrix; The cycle To cycle The standard pressure and standard temperature are arranged to generate the cycle The pressure diagonal matrix and temperature diagonal matrix of , set the physical constraint to minimize the cycle The eigenvector matrix of the pressure diagonal matrix is ​​the quadratic form minus the period The norm of the quadratic form of the eigenvector matrix and the temperature diagonal matrix, multiplying the physical constraints by the regularization coefficient and superimposed with the objectives of principal component analysis; Use the alternating optimization algorithm and initialize the variables. In each iteration, fix the first variable and update the second variable according to the gradient descent. Fix the updated second variable and update the first variable according to the gradient descent. Repeat the iteration until the alternating optimization algorithm converges to obtain the final variable. The variable includes the period The eigenvector matrix and eigenvalue diagonal matrix of ; From the cycle Extract the period from the eigenvector matrix The eigenvalue diagonal matrix of The eigenvalues ​​corresponding to the eigenvectors are extracted The first eigenvector Dimension and splice into cycles The coupling characteristic vector of The coupled characteristic sequence.

6. The safety warning system according to claim 4, characterized in that: The processing steps of the prediction model include: Use triangle encoding to convert the period Convert to cycle The encoding vector will be periodic The encoding vector of The transpose of the code element and the joint input vector Add the elements of the dimensions to generate the cycle The embedded input vector is integrated to obtain the period The embedded input sequence is ; Using attention mechanism to capture cycles The temporal dependency of the joint input sequence between different periods and the dimensional dependency between different dimensions of the embedded input vector of the same period are used to generate spatiotemporal correlation features; The spatiotemporal correlation features are input into the feedforward network for nonlinear transformation and feature enhancement, and the dimensions are adjusted again through linear weights and linear biases, and mapped to the predicted data tensor through the ReLU function.

7. The safety warning system according to claim 1, characterized in that: Determine the cycle The adjustable duration includes the following steps: Cycle-based The temperature trend function and pressure trend function are used to determine the period To cycle Whether there is a moment equal to an upper temperature threshold and / or an upper pressure threshold during the period; If it does not exist, no processing will be done; If it exists, the moment that first reaches the upper temperature threshold and / or the upper pressure threshold is regarded as the dangerous moment, and whether the dangerous moment is within the cycle Inside; If the danger is in the cycle Internal cycle The adjustable time is equal to the dangerous moment minus the period The collection time, if in the cycle Afterwards, the cycle The adjustable time is equal to the cycle The collection time minus the period The collection time.

8. The safety warning system according to claim 1, characterized in that: The equipment control model includes a temperature control equipment control model and a pressure control equipment control model. The temperature control equipment control model is specifically a linear transformation in which the ratio of the temperature change to the adjustable time is equal to the temperature control parameter adjustment amount. The pressure control model is specifically a linear transformation in which the ratio of the pressure change to the adjustable time is equal to the pressure control parameter adjustment amount. The unknown parameters of the linear transformation are determined by the least squares fitting method. The control objective function at the dangerous moment is defined as the weighted sum of the temperature limit value, pressure limit value, temperature control energy consumption and pressure control energy consumption at the dangerous moment. Among them, the temperature limit value at the dangerous moment is equal to 0 and the dangerous moment is equal to 0. The larger value of the difference between the corrected temperature and the temperature upper limit threshold, the pressure overlimit value at the dangerous moment is equal to 0 and the larger value of the difference between the corrected pressure at the dangerous moment and the pressure upper limit threshold, the temperature control energy consumption and the pressure control energy consumption are equal to the corresponding parameter adjustment amount multiplied by the controllable time.

9. The safety warning system according to claim 1, characterized in that: Cycle based on equipment control model Adjust and correct the temperature trend function and pressure trend function; When danger lies in the cycle The corrected temperature and corrected pressure at the dangerous moment are equal to the temperature trend function value at the dangerous moment plus the period determined by the equipment control model. The temperature control parameter adjustment amount and the pressure trend function value at the dangerous moment plus the cycle determined by the equipment control model The pressure control parameter adjustment amount; When danger lies in the cycle Afterwards, according to the cycle The difference between the corrected temperature at the time of collection and the temperature trend function value is used to linearly extrapolate the temperature trend function value at the dangerous moment to obtain the corrected temperature at the dangerous moment. The difference between the corrected pressure at the acquisition moment and the pressure trend function value is used to linearly extrapolate the pressure trend function value at the dangerous moment to obtain the corrected pressure at the dangerous moment.

10. The safety warning system according to claim 1, characterized in that: To minimize dangerous moments The control objective function is the optimization objective, and the optimization constraint is set as the cycle The corrected temperature and corrected pressure at any moment from the collection time to the dangerous moment are within the corresponding temperature range and pressure range and the cycle The temperature control parameter adjustment amount and the pressure control parameter adjustment amount are within the corresponding temperature control range and pressure control range. The optimization constraint is introduced into the optimization objective through the Lagrange multiplier method to construct the Lagrange augmented function and the KKT condition is used to solve the period The optimal temperature control parameter adjustment amount and pressure control parameter adjustment amount are spliced ​​into a cycle The optimal parameter adjustment vector.

Images