Temperature and pressure coordinated regulation and control method and system for calcium stearate reaction process

Abstract

The invention discloses a temperature and pressure coordinated regulation and control method and system for a calcium stearate reaction process, and relates to the field of industrial control, the system comprises an acquisition module, a processing decision module and a regulation and control module, and the acquisition module measures temperature and pressure at high frequency in each period and generates a parameter sequence; the processing decision module performs Kalman filtering on the parameter sequence to obtain a de-noising parameter sequence, performs subtraction on the de-noising parameter sequence and a target parameter to generate a deviation value sequence, and decides whether to regulate and control by counting an over-limit time ratio; when regulation and control are needed, the regulation and control module obtains a reaction fixed parameter set, establishes a correlation formula, calculates a coupling coefficient vector sequence, splices the coupling coefficient vector sequence and a denoising parameter sequence into a trend characteristic matrix, inputs the trend characteristic matrix and a control parameter vector into a regulation and control network, and carries out dynamic attention convolution, equipment aging deduction, graph convolution and time sequence prediction. And iteratively updating the parameter control vector to obtain an optimal parameter control vector, thereby realizing precise linkage control of temperature and pressure.

Description

Technical Field

[0001] This invention relates to the field of industrial control, and in particular to a method and system for coordinated temperature and pressure control of calcium stearate reaction process. Background Technology

[0002] Calcium stearate, a key additive in plastics, rubber, and coatings, relies heavily on the reaction stability of its preparation process for core properties such as purity and particle size uniformity. Temperature and pressure are crucial process parameters affecting this exothermic chemical reaction. During the synthesis of calcium stearate, the system temperature continuously changes with the exothermic reaction, causing internal gas volume expansion or contraction, leading to significant pressure fluctuations. Simultaneously, pressure changes affect the contact frequency and reactivity of material molecules, inversely regulating the exothermic reaction rate, forming a tight temperature-pressure coupling cycle. If the temperature and pressure parameters deviate from the preset process range, not only will product performance fail to meet standards, but abnormal pressure increases may also pose safety risks to the reactor. Therefore, precise and coordinated control of temperature and pressure parameters is a core prerequisite for ensuring production safety and product quality.

[0003] A Chinese patent application with publication number CN121300560A discloses a method and system for temperature and pressure control of a reactor. This scheme collects temperature and pressure data in different areas of the reactor through distributed sensors, filters and reduces noise, calculates the average value, and then performs dynamic compensation in combination with environmental parameters. The coupling coefficient is then calculated to determine the temperature and pressure coupling strength, and the coordinated or independent adjustment strategy is switched. The adjustment amount is generated based on the predictive control model, and closed-loop adjustment is executed using segmented control logic. This solves the problems of low accuracy and weak anti-interference of traditional control, and has the advantages of high precision, fast response and high safety, making it suitable for high-requirement reaction processes in chemical, pharmaceutical and other fields.

[0004] However, existing technologies have not been optimized for the specific reaction characteristics of calcium stearate. The coupling analysis is based only on general partial derivative calculations and medium corrections, without combining the reaction mechanism of calcium stearate, the gas state equation, and the heat conduction law to build a customized correlation model. It is difficult to accurately match the characteristics of temperature and pressure interaction in this reaction system. Predictive control relies on rolling optimization based on limited historical data, lacks in-depth feature extraction of temperature and pressure coupling relationship, is not adaptable to the dynamic changes in coupling strength during the reaction process, and does not incorporate the aging and degradation of the control equipment after long-term operation into the control logic. It is impossible to maintain a long-term stable control effect through dynamic efficiency compensation, and it is difficult to meet the stringent requirements of continuous industrial production of calcium stearate. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by proposing a temperature and pressure coordinated control method and system for the calcium stearate reaction process. Through Kalman filtering, temperature-pressure correlation, attention mechanism convolution, equipment aging extrapolation, and iterative optimization, precise linkage control of temperature and pressure is achieved, ensuring the reaction stability of calcium stearate.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A temperature-pressure synergistic control method for the calcium stearate reaction process includes the following steps:

[0008] Measured in the current period The temperature and pressure were then analyzed, and Kalman filtering was performed to obtain a denoised parameter sequence. and with target parameter By subtracting the values, we obtain the sequence of deviations. The percentage of times exceeding the limit is statistically analyzed to determine whether to wait until the next cycle or if adjustments are needed. For the number of measurements;

[0009] If regulation is required, a fixed set of reaction parameters is obtained, and correlations are established based on the reaction mechanism, the gas law, and the heat conduction formula, according to the denoised parameter sequence. middle The coupling coefficient vector sequence is obtained by calculating the parametric vectors of the second measurement and the fixed parameter set of the response. , and the denoised parameter sequence Concatenate into a trend feature matrix The total number of recursive times is determined based on the remaining time. ;

[0010] Trend feature matrix With preset control parameter vector Collaborative input regulation networks, utilizing attention-tuned... Each convolutional kernel pairs the trend feature matrix Perform 1D convolution and map to a sequence of coupled feature vectors Considering the aging of the control equipment, and combining the control parameter vector... By performing a moment-by-moment deduction, the efficiency vector sequence is obtained. and coupled with the feature vector sequence After concatenation, graph convolution and temporal prediction are performed to obtain the first... Recursive parameters at each recursive time step and with target parameter Comparison to iteratively update control parameter vectors The optimal control parameter vector is obtained. And send control equipment.

[0011] Furthermore, Kalman filtering includes the following steps:

[0012] Definition of the first The parameters of this measurement Filtering yields denoising parameters And initialize the denoising parameters for the 0th measurement. The covariance matrix of the 0th measurement ;

[0013] Constructing the transfer equation recurrence relation The main diagonal element is 1, and the secondary diagonal elements in the first and second rows are the thermal conductivity rates, respectively. With interval duration Product of gas expansion coefficient With interval duration The accumulation of;

[0014] Constructing the noise matrix noise matrix The elements on the main diagonal are, in order, the temperature variance. and pressure variance ;

[0015] For parameter sequences Perform successive filtering on the th The parameters of this measurement , will the Denoising parameters for the second measurement With the transfer equation Multiplying gives the first Predicted parameters for this measurement The recurrence relation , No. The covariance matrix of the second measurement Recurrence relation Multiply the transposes and add the noise matrix , obtained the The prediction covariance matrix of the second measurement ;

[0016] The first The parameters of this measurement With prediction parameters Divide and diagonalize to the first Error matrix of the second measurement , will the The prediction covariance matrix of the second measurement With the Error matrix of the second measurement The inverse matrix of the sum of the first and second matrices multiplied by the first... The prediction covariance matrix of the second measurement Get the first Gain matrix of the second measurement ;

[0017] The first Predicted parameters for this measurement Add gain matrix Multiply by the first The parameters of this measurement With prediction parameters The difference is to obtain the first Denoising parameters for the second measurement Combine the identity matrix and the gain matrix The difference multiplied by the first The prediction covariance matrix of the second measurement , obtained the The covariance matrix of the second measurement ;

[0018] Filtering in sequence The parameters measured in this step are used to obtain a denoised parameter sequence. .

[0019] Furthermore, the deviation sequence is calculated. The process involves calculating the percentage of times exceeding limits to determine whether to implement control measures, including the following steps:

[0020] Extracting denoised parameter sequences The Middle Denoising parameters for the second measurement And subtract the target parameter Get the first Deviation of the measurement ;

[0021] Compare the first Deviation of the measurement The absolute value and the deviation threshold vector Set the number of times the limit is exceeded to 0;

[0022] If the first Deviation of the measurement Medium temperature deviation The absolute value is greater than the temperature deviation threshold. and / or pressure deviation The absolute value is greater than the pressure deviation threshold. When the time limit is exceeded, the number of times exceeds the limit is incremented by 1; otherwise, the number of times exceeds the limit remains unchanged.

[0023] Iterate through and confirm the number of times the limit was exceeded, then divide by the number of measurements. The out-of-limit time ratio is obtained. If the out-of-limit time ratio is greater than the time ratio threshold, it is determined that control is needed. Otherwise, it is determined that no control is needed.

[0024] Furthermore, determine the coupling coefficient vector sequence. This includes the following steps:

[0025] Define the gas expansion coefficient and thermal conductivity Obtain the fixed set of reaction parameters;

[0026] A pressure-temperature correlation is constructed based on the gas law, which describes the pressure variation. Proportional to temperature change Based on the reaction mechanism and heat conduction formula of calcium stearate, relying on the exothermic change A temperature-pressure correlation is constructed for the transition quantity, and the temperature-pressure correlation describes the temperature change. Proportional to transformer capacity ;

[0027] The end parameter of the previous cycle, measured at the end time of the previous cycle, is used as the denoising parameter for the 0th measurement. denoising parameter sequence middle Subtracting the denoising parameter from the previous measurement from the denoising parameter of each subsequent measurement yields... The parametric vectors measured in this study are used to calculate the reaction fixed parameter set. The pressure-temperature coupling coefficient and temperature-pressure coupling coefficient measured in this study are respectively concatenated as follows: The coupling coefficient vectors from each measurement are integrated to obtain a coupling coefficient vector sequence. .

[0028] Furthermore, the optimal control parameter vector output by the control network is determined. Includes the following steps:

[0029] Based on the attention mechanism, and according to the coupling coefficient vector sequence Dynamic adjustment of coupling strength The receptive field range and weight distribution of each convolutional kernel are analyzed, along with the trend feature matrix. Perform 1D convolution and mapping to output a coupled feature vector sequence. ;

[0030] Obtain the cumulative temperature control duration of the temperature control device Cumulative pressure control time of pressure control equipment Divide each factor by the corresponding equipment lifespan and combine them to obtain the aging rate vector. Combined with preset control parameter vectors With efficiency function Performing a time-by-time simulation yields an efficiency vector sequence. Among them, the efficiency function At a given aging rate and recursion time Multiply by the control parameter vector The corresponding control parameters can be used to obtain the first... Control efficiency and aging rate at each recursive time step Including temperature control aging rate and pressure aging rate Control efficiency includes temperature control efficiency and pressure control efficiency;

[0031] Coupled feature vector sequences and efficiency vector sequence Concatenate them into a joint feature vector sequence, and then concatenate the joint feature vector sequence into a joint feature vector sequence. The joint feature vectors at each recursive time step are used as nodes and graph convolution is performed to obtain the reconstructed feature vector sequence, which is then input into a bidirectional gated recurrent layer to obtain the th ... Recursive parameters at each recursive time step and with target parameter Difference to obtain the first Deviation at each recursive time step To perform the verification, compare the first Deviation at each recursive time step The absolute value and the deviation threshold vector To decide whether to update the control parameter vector And re-execute hourly extrapolation, aggregated prediction, and verification, or change the control parameter vector. As the optimal control parameter vector Output.

[0032] Furthermore, from the trend feature matrix Extracting the coupling coefficient vector sequence Global average pooling is performed to obtain the global coupling coefficient, and the global coupling coefficient is mapped using a fully connected layer. The dilation parameter vector of dimension, the dilation parameter vector includes The dilation adjustment parameters of each convolution kernel will affect the trend feature matrix. As a query matrix, the coupling coefficient vector sequence As the key and value matrices, scaled dot product attention is performed to obtain weighted coupled features, which are then mapped sequentially through global pooling and fully connected layers to obtain a sequence of coupled attention weight matrices. Initialization is then performed. The convolution weight matrix of each convolution kernel will couple the attention weight matrix sequence. The coupling attention weight matrix and Each convolution weight matrix is ​​superimposed to obtain The dynamic weight matrix of each convolutional kernel, using Each convolutional kernel pairs the trend feature matrix Perform 1D temporal convolution, adjust the dimensions through a fully connected layer, and map the output coupled feature vector sequence. .

[0033] Furthermore, the first Recursive parameters at each recursive time step With target parameter Difference to obtain the first Deviation at each recursive time step And after taking the absolute value, compare it with the deviation threshold vector If the first Deviation at each recursive time step Medium temperature deviation The absolute value is greater than the temperature deviation threshold. and / or pressure deviation The absolute value is greater than the pressure deviation threshold. At that time, the control parameter vector is updated based on the gradient descent method. And re-execute the hourly extrapolation, aggregated prediction, and verification until the [number]th [time]. Deviation at each recursive time step Medium temperature deviation The absolute value is less than or equal to the temperature deviation threshold. And pressure deviation The absolute value is less than or equal to the pressure deviation threshold. At that time, the control parameter vector will be... As the optimal control parameter vector Output.

[0034] A temperature and pressure coordinated control system for the calcium stearate reaction process includes a data acquisition module, a processing and decision-making module, and a control module.

[0035] The acquisition module measures each cycle separately. Secondary temperature and pressure, generating parameter sequences And send it to the processing decision module. For the number of measurements;

[0036] The decision processing module processes the parameter sequence for each period. Perform Kalman filtering to obtain the denoised parameter sequence. and with target parameter By subtracting the values, we obtain the sequence of deviations. The system calculates the ratio of times exceeding the limit to determine whether to implement control measures. If no control is needed, no action is taken; if control is required, a denoised parameter sequence is sent. To the control module;

[0037] The control module receives the denoised parameter sequence. It also obtains a fixed set of reaction parameters, establishes correlations based on the reaction mechanism, gas law, and heat conduction formula, and uses a denoised parameter sequence. middle The coupling coefficient vector sequence is obtained by calculating the parametric vectors of the second measurement and the fixed parameter set of the response. , and the denoised parameter sequence Concatenate into a trend feature matrix and with the preset control parameter vector Collaborative input regulation networks, utilizing attention-tuned... Each convolutional kernel pairs the trend feature matrix Perform 1D convolution and map to a sequence of coupled feature vectors Considering the aging of the control equipment, and combining the control parameter vector... By performing a moment-by-moment deduction, the efficiency vector sequence is obtained. and coupled with the feature vector sequence After concatenation, graph convolution and temporal prediction are performed to obtain the first... Recursive parameters at each recursive time step and with target parameter Comparison to iteratively update control parameter vectors The optimal control parameter vector is obtained. And send it to the control equipment, whereby the total number of recursive time intervals is... Determined based on the remaining time.

[0038] Compared with the prior art, the significant advantages of this invention are:

[0039] 1. High-frequency temperature and pressure parameters are denoised using Kalman filtering to obtain a denoised parameter sequence. Then, a temperature-pressure correlation is established based on the calcium stearate reaction mechanism, the gas law, and the heat conduction formula. The coupling coefficient vector sequence is calculated by combining the denoised parameter sequence and concatenating it with the denoised parameter sequence to form a trend feature matrix. Finally, the receptive field range and weight distribution of the convolution kernel are dynamically adjusted through an attention mechanism, and convolution is performed on the trend feature matrix to extract accurate coupling features. This achieves deep capture of the nonlinear coupling relationship between temperature and pressure, solving the problem of insufficient accuracy caused by single or simple correlation control.

[0040] 2. Obtain the cumulative running time and corresponding equipment life of the temperature and pressure control equipment, calculate the temperature control aging rate and pressure control aging rate; construct an efficiency function, substitute the aging rate and the recursive time into the function, and combine it with the preset control parameter vector to calculate the effective efficiency of temperature and pressure control at each recursive time, forming an efficiency vector sequence; then integrate this sequence into the moment-by-moment deduction of the control network to compensate for the control efficiency decay caused by equipment aging in real time, effectively solving the problem of control effect deviating from the target during long-term operation and ensuring control stability. Attached Figure Description

[0041] Figure 1 A flowchart of a temperature and pressure synergistic control method for the calcium stearate reaction process;

[0042] Figure 2 A flowchart for decision-making and control processes;

[0043] Figure 3 To regulate the network execution flowchart;

[0044] Figure 4 This is a system architecture diagram for temperature and pressure coordinated control of the calcium stearate reaction process. Detailed Implementation

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0046] Example 1

[0047] like Figure 1 As shown, a specific embodiment of the present invention discloses a method for synergistic temperature and pressure control in the calcium stearate reaction process, comprising the following steps:

[0048] Based on the interval duration in the current cycle High-frequency measurements were performed on the reaction vessel, and the following measurements were taken respectively. Subsequent temperature and pressure, generating dimensional parameter sequence , In this embodiment, for the purpose of measuring the number of times, ;

[0049] right dimensional parameter sequence Perform Kalman filtering to obtain Denoising parameter sequences of dimension And The noise reduction parameters for this measurement are respectively with Dimensional target parameter To do the work, one gets Dimensional deviation sequence The system calculates the percentage of times the limit is exceeded to determine whether to implement control measures. If no control is needed, the system waits until the next cycle.

[0050] If regulation is required, a fixed set of reaction parameters is obtained. Based on the reaction mechanism, and combining the gas law and heat conduction formula, a correlation between temperature and pressure is established. Denoising parameter sequences of dimension Sure Substitute the parametric vectors from the measurement into the correlation equation to calculate... Second measurement The coupling coefficient vectors of dimension are concatenated to obtain dimensional coupling coefficient vector sequence and Denoising parameter sequences of dimension Consistent splicing in time sequence Trend feature matrix of dimension Get the remaining duration of the current period, and divide the remaining duration by the interval duration. Round down and add 1 to get the total number of recursive times. The remaining duration is the time within this period. dimensional coupling coefficient vector sequence The duration from the determined moment to the end moment, and the interval between the end moments of this cycle. This refers to the measurement time of the first measurement in the next cycle;

[0051] Will Trend feature matrix of dimension With preset dimensional control parameter vector Collaborative input regulation network, dynamically adjusting using attention mechanisms Each convolutional kernel and... Trend feature matrix of dimension Perform 1D convolution and mapping to obtain dimensional coupled feature vector sequence Considering the aging of the control equipment, combined with dimensional control parameter vector conduct The step-by-step deduction of each recursive time step yields... dimensional efficiency vector sequence splicing dimensional coupled feature vector sequence and dimensional efficiency vector sequence Then, graph convolution and temporal prediction are performed sequentially to obtain the first... each recursive time step Dimension recursive parameters ,and Dimensional target parameter Comparison with iterative updates dimensional control parameter vector Until the verification is successful Optimal control parameter vector of dimension , To couple the feature dimensions, Optimal control parameter vector of dimension The signals are sent to the control equipment for control, which includes temperature control equipment and pressure control equipment. In this embodiment... .

[0052] Furthermore, a data acquisition card is used to control thermocouple sensors inserted into the material area inside the reactor and diffused silicon pressure transmitters deployed at the pressure interface inside the reactor at preset intervals. High-frequency measurements are performed in each cycle, and measurements are taken separately in the current cycle. The next temperature and pressure will... Temperature measured ,pressure Concatenate the columns into the corresponding parameters. , Represents the transpose of a matrix or vector. Within the current period The parameters obtained from each measurement are sequentially integrated to obtain the current cycle. dimensional parameter sequence .

[0053] Furthermore, Kalman filtering includes the following steps:

[0054] Define denoising parameters For the first The parameters of this measurement The filtering result will be the denoising parameter of the 0th measurement. Initialize to target parameter The covariance matrix of the 0th measurement The matrix is ​​set as a 2×2 diagonal matrix, and the main diagonal elements of the first and second rows are initialized sequentially with the temperature measurement error covariance of the thermocouple sensor. Covariance of pressure measurement error with diffused silicon pressure transmitter , and The first The noise-reduced temperature and noise-reduced pressure measured in this instance, and the temperature measurement error covariance. Covariance of pressure measurement error The temperature measurement error covariance can be determined through prior measurement experiments and statistical analysis. Covariance of pressure measurement error They are 0.01 (°C) respectively. 2 And 0.0025 (MPa) 2 ;

[0055] Based on the heat conduction rate in the calcium stearate reaction system With gas expansion coefficient Construct the transfer equation To describe the correlation of parameter changes between adjacent measurements, the transfer equation It is a 2×2 symmetric matrix with elements of 1 on the main diagonal. The elements on the secondary diagonal of the first and second rows represent the thermal conductivity rates, respectively. With interval duration The product of gas expansion coefficient With interval duration The product of, where, thermal conductivity Interval duration and gas expansion coefficient The values ​​were 0.8, 0.5 s, and 0.036, respectively.

[0056] Constructing the noise matrix noise matrix Set as a 2×2 diagonal matrix, with the main diagonal elements representing the temperature variances. and pressure variance Temperature variance and pressure variance By statistical stabilization at the target parameter The parameter sequence of multiple nearby periods was determined, among which the temperature variance and pressure variance They are 0.005 (°C) respectively. 2 And 0.001 (MPa) 2 ;

[0057] From the parameter sequence The parameters measured in the first measurement Begin sequential filtering, for the first... The parameters of this measurement , will the Denoising parameters for the second measurement Multiply by the recurrence relation , obtained the Predicted parameters for this measurement The recurrence relation , No. The covariance matrix of the second measurement Recurrence relation After multiplying the transposes, the noise matrix is ​​superimposed. , obtained the The prediction covariance matrix of the second measurement ;

[0058] Based on the The parameters of this measurement With prediction parameters Calculate the first one respectively Temperature measurement error of this measurement and pressure measurement error And place them in the first and second rows of a 2×2 diagonal matrix respectively, to obtain the... Error matrix of the second measurement ;

[0059] Due to the The parameters of this measurement With prediction parameters The dimension and physical meaning are the same in Kalman filtering, which is the observation matrix. A 2×2 identity matrix is ​​used to reflect the physical consistency between the observed and measured quantities. Since the identity matrix does not change the result in matrix operations, the calculation of the... Gain matrix of the second measurement , No. Gain matrix of the second measurement equal to the The prediction covariance matrix of the second measurement Multiply by the first The prediction covariance matrix of the second measurement With the Error matrix of the second measurement The inverse matrix of the sum;

[0060] The first Predicted parameters for this measurement Add gain matrix Multiply by the first The parameters of this measurement With prediction parameters The difference is to obtain the first Denoising parameters for the second measurement , will be with the gain matrix Identity matrix and gain matrix of the same dimension The difference multiplied by the first The prediction covariance matrix of the second measurement , obtained the The covariance matrix of the second measurement ;

[0061] Filtering in sequence The parameters of the second measurement were obtained. Denoising parameter sequences of dimension .

[0062] For example, suppose the parameter measured in the first measurement of the current period is... ;

[0063] Initialize the denoising parameters for the 0th time covariance matrix recurrence relation ;

[0064] Calculate the predicted parameters for the first measurement ;

[0065] Calculate the prediction covariance matrix of the first measurement. The details are as follows:

[0066] ;

[0067] Calculate the error matrix of the first measurement. The details are as follows:

[0068] ;

[0069] in, This indicates a diagonalization operation, used to calculate the gain matrix of the first measurement. The details are as follows:

[0070] ;

[0071] Calculate the denoising parameters for the first measurement. Simultaneously, the covariance matrix of the first measurement is updated. Following the above approach, the second to tenth measurements are processed recursively to obtain the results for this period. Denoising parameter sequences of dimension .

[0072] like Figure 2 As shown, further, the deviation sequence is calculated. The process involves calculating the percentage of times exceeding limits to determine whether to implement control measures, including the following steps:

[0073] Get Dimensional target parameter ,in, and These are the target temperature and target pressure preset according to the requirements of the calcium stearate reaction;

[0074] extract Denoising parameter sequences of dimension The Middle Denoising parameters for the second measurement Subtract the target parameter Later obtained the first Second measurement Dimensional deviation ,in, and These are temperature deviation and pressure deviation, respectively. In this embodiment, the target parameter is... ;

[0075] Compare the first Second measurement Dimensional deviation absolute value and dimensional deviation threshold vector Set the number of times the limit is exceeded to 0, where, and These are the temperature deviation threshold and the pressure deviation threshold, respectively. In this embodiment, the temperature deviation threshold... and pressure deviation threshold The pressure was set to 2℃ and 0.001MPa, respectively.

[0076] If the first Second measurement Dimensional deviation Medium temperature deviation The absolute value is greater than the temperature deviation threshold. and / or pressure deviation The absolute value is greater than the pressure deviation threshold. In the case of exceeding the limit, the number of time points will be incremented by 1; otherwise, the number of time points will remain unchanged.

[0077] Traversal The deviation of each measurement is determined by identifying the number of times the measurement exceeds the limit and then dividing by the number of measurements. The out-of-limit time ratio is obtained and compared with the time ratio threshold. If the out-of-limit time ratio is greater than the time ratio threshold, it is determined that control is needed. If the out-of-limit time ratio is less than or equal to the time ratio threshold, it is determined that no control is needed. The time ratio threshold is set to 0.3.

[0078] Further, confirm dimensional coupling coefficient vector sequence This includes the following steps:

[0079] Define the gas expansion coefficient The thermal conductivity rate is defined as the rate of relative change in gas volume for every 1°C change in temperature. To determine the heat transfer rate of the reaction system, obtain a set of fixed reaction parameters, including the reactor volume. Material contact area Gas constant Average molar mass of materials gas average molar mass Specific heat capacity of materials Among them, the gas expansion coefficient By pre-determining the gas composition and reaction conditions under simulated calcium stearate reaction through statistical experiments, the gas volume change at different temperature points was measured, and the relative rate of change of gas volume for every 1°C change in temperature was calculated. The average value of multiple measurements was taken as the gas expansion coefficient. thermal conductivity By pre-determining through statistical experiments, under the steady-state condition of the reaction system, the temperature gradient and the total amount of heat transferred during the heat transfer process are measured. The average heat transfer rate of the reaction system is then calculated using the fundamental laws of heat conduction and taken as the heat conduction rate. The raw materials for the calcium stearate reaction are stearic acid and calcium carbonate. The technical solution of this application is clearly applied to the steady-state process of the calcium stearate reaction, which requires maintaining a stable reaction environment and balancing the raw material consumption and product formation rates. Based on engineering approximation, the average molar mass of the materials is obtained by weighted summation of the molar masses of stearic acid and calcium carbonate according to the feed ratio. Based on the chemical equation, the product ratio of carbon dioxide and water is 1:1. The average molar mass of the gas is obtained by averaging the molar masses of carbon dioxide and water. ;

[0080] Based on the gas law and the pressure-temperature correlation of anchor pressure with respect to temperature, during the reaction, the increase in temperature drives an increase in the kinetic energy of gas molecules, which is related to the gas expansion coefficient. The correlation triggers a pressure change inside the reactor; the pressure-temperature correlation describes the pressure change. Proportional to the gas expansion coefficient gas average molar mass Gas constant Temperature variation and reactor volume The reciprocal of;

[0081] Since the reaction of calcium stearate is exothermic, the amount of heat released is directly proportional to the reaction rate. When the pressure increases, the probability of molecular collisions increases, the reaction rate accelerates, and the amount of heat released increases. Proportional to transformer capacity Combining the heat conduction formula, the change in heat release Proportional to the rate of thermal conduction Material contact area Specific heat capacity of materials With temperature change The temperature-pressure relationship describes the temperature variation. Proportional to transformer capacity And inversely proportional to the rate of thermal conduction Material contact area and specific heat capacity of materials ;

[0082] The end parameter of the previous cycle is used as the value of the 0th measurement. Dimensional denoising parameters ,Will Denoising parameter sequences of dimension middle Subtracting the denoising parameter from the previous measurement from the denoising parameter of each subsequent measurement yields... The parametric vectors measured were obtained and calculated based on the pressure-temperature correlation, the temperature-pressure correlation, and the fixed set of reaction parameters. The pressure-temperature coupling coefficient and the temperature-pressure coupling coefficient measured in this step are then combined to obtain... The coupling coefficient vectors of each measurement are sequentially concatenated to obtain... dimensional coupling coefficient vector sequence The ending parameter of the previous cycle is obtained by measuring at the end time of the previous cycle. The pressure-temperature coupling coefficient of the second measurement and temperature-pressure coupling coefficient Reflecting the first The parametric vector of the second measurement Substitute the temperature-pressure correlation and the pressure-temperature correlation, and combine them with the proportional ratio obtained by the fixed set of reaction parameters.

[0083] like Figure 3 As shown, further, the optimal control parameter vector output by the control network is... Includes the following steps:

[0084] right Trend feature matrix of dimension Perform dynamic attention convolution from the trend feature matrix Extract dimensional coupling coefficient vector sequence Global average pooling is performed to obtain the global coupling coefficient, and the global coupling coefficient is mapped using a fully connected layer. The dilation parameter vector of dimension, the dilation parameter vector includes The dilation adjustment parameters of each convolution kernel will affect the trend feature matrix. As a query matrix, the coupling coefficient vector sequence As the key matrix and value matrix, scaling dot product attention is performed to obtain... The weighted coupling features of the dimension are obtained by sequentially mapping through global pooling and fully connected layers. Dimensional coupling: Pay attention to the weight matrix sequence, initialization. Each dimension is The convolution weight matrix of the convolution kernel will... Dimensional coupling, pay attention to the weight matrix sequence indivual Dimensional coupling: pay attention to the weight matrix and Each convolution weight matrix is ​​superimposed to obtain The dynamic weight matrix of each convolutional kernel, using Each convolutional kernel pair Trend feature matrix of dimension Perform a step of 1 and fill with 1 The 1D temporal convolution is adjusted in dimension and the output is mapped through a fully connected layer. dimensional coupled feature vector sequence ,in, The total number of convolution kernels, For timing window size, This is the [number]th [period] in this cycle. The coupling characteristic vector at the recursive time step, where the recursive time step of the same period is always at the . Following the measurement time of this measurement, in this embodiment... , ;

[0085] Obtain the cumulative temperature control duration of the temperature control device Cumulative pressure control time of pressure control equipment Divide each by the corresponding equipment lifespan to obtain Dimensional aging rate vector And combined with preset dimensional control parameter vector Perform a time-by-time simulation, which is based on an efficiency function. Execution, efficiency function Used to describe aging rate The control equipment in the first The efficiency of each recursive time step is as follows:

[0086] ,

[0087] in, As an effective influencing factor, control the aging rate. Regarding the degree of impact of the regulation's effectiveness, when the aging rate The smaller and / or the number of recursion time The closer to the total number of recursion times At the same time, the higher the efficiency of the control equipment, the lower the temperature control aging rate. Substitute the recursion time into the efficiency function in sequence. And multiply by the temperature control parameter ,get Temperature control efficiency sequence of dimensional , For the first The temperature control efficiency at each recursive time point will affect the pressure aging rate. Substitute the recursion time into the efficiency function in sequence. And multiply by the pressure control parameter ,get Dimensional control efficiency sequence , For the first After the measurement time, the first The effectiveness of pressure control at each recursive time point, in The dimension is obtained by aligning and stitching together the recursive time steps. dimensional efficiency vector sequence Among them, the cumulative temperature control time and cumulative pressure control duration In this embodiment, the effective influencing factor refers to the total cumulative operating time of the temperature control equipment and pressure control equipment after they have been put into use. It is 0.7;

[0088] Will dimensional coupled feature vector sequence and efficiency vector sequence The dimensions obtained by splicing are The joint feature vector sequence of dimension is analyzed and aggregated prediction is performed, which will include the joint feature vector sequence. The joint feature vectors at each recursive time step are used as nodes. A graph convolutional layer is used to capture the interaction of feature information between nodes, and the result is aggregated to obtain... The reconstructed nodes at each recursive time step are sequentially restored to... The reconstructed feature vector sequence of dimension 1 is deduced step-by-step up to the 1st dimension based on the temporal memory capability of the bidirectional gated recurrent layer. The recursive time step is obtained by mapping through a fully connected layer to obtain the th time step in this cycle. each recursive time step Dimension recursive parameters ,in, and The first The recursion temperature and recursion pressure at each recursion time step are discussed. Graph convolutional layers and bidirectional gated recurrent layers are common neural network models and are not the focus of this application's technical solution. The number of layers and the dimension of each layer in the graph convolutional layer are 2 and 12, respectively. The number of layers and the hidden dimension of each layer in the bidirectional gated recurrent layer are 2 and 32, respectively. Therefore, the bidirectional gated recurrent layer is derived to the […]. The hidden state dimension at each recursive time step is 32×2, while the fully connected layer reduces the dimension to 1×2.

[0089] The first each recursive time step Dimension recursive parameters and Dimensional target parameter Difference to obtain the first each recursive time step Dimensional deviation To perform the verification, compare the first each recursive time step Dimensional deviation absolute value and dimensional deviation threshold vector Based on the decision of whether to update dimensional control parameter vector And re-execute the moment-by-moment extrapolation, aggregated prediction, and verification or... dimensional control parameter vector As Optimal control parameter vector of dimension Output.

[0090] Furthermore, the first each recursive time step Dimension recursive parameters and Dimensional target parameter Difference to obtain the first each recursive time step Dimensional deviation Comparison of the first each recursive time step Dimensional deviation absolute value and dimensional deviation threshold vector If the first each recursive time step Dimensional deviation Medium temperature deviation The absolute value is greater than the temperature deviation threshold. and / or pressure deviation The absolute value is greater than the pressure deviation threshold. At that time, the gradient descent method is used, based on the first... each recursive time step Dimensional deviation about dimensional control parameter vector The gradient is updated in conjunction with the learning step size. dimensional control parameter vector And re-execute the hourly extrapolation, aggregated prediction, and verification until the [number]th [time]. each recursive time step Dimensional deviation Medium temperature deviation The absolute value is less than or equal to the temperature deviation threshold. And pressure deviation The absolute value is less than or equal to the pressure deviation threshold. At that time, this time dimensional control parameter vector As Optimal control parameter vector of dimension Output.

[0091] Example 2

[0092] like Figure 4 As shown, a temperature and pressure coordinated control system for the calcium stearate reaction process is disclosed to execute the above-mentioned temperature and pressure coordinated control method for the calcium stearate reaction process, including a data acquisition module, a processing and decision module, and a control module;

[0093] The data acquisition module performs periodic data acquisition, and within each period, data is collected based on the interval duration. High-frequency measurements were performed on the reaction vessel, and the following measurements were taken respectively. Subsequent temperature and pressure, generating dimensional parameter sequence And send it to the processing decision module. For the number of measurements;

[0094] The decision-making module processes each cycle. dimensional parameter sequence Perform Kalman filtering to obtain Denoising parameter sequences of dimension And The noise reduction parameters for this measurement are respectively with Dimensional target parameter To do the work, one gets Dimensional deviation sequence The system calculates the number of times limits are exceeded to determine whether intervention is needed. If no intervention is required, no action is taken; if intervention is required, data is sent. Denoising parameter sequences of dimension To the control module;

[0095] Control module receives Denoising parameter sequences of dimension And obtain a fixed set of reaction parameters, establish correlations based on the reaction mechanism, gas law, and heat conduction formula, and then... Denoising parameter sequences of dimension Sure The parametric vectors from the second measurement are substituted into the correlation equation, and combined with the fixed set of reaction parameters, to obtain... dimensional coupling coefficient vector sequence and Denoising parameter sequences of dimension spliced ​​together Trend feature matrix of dimension , and the preset dimensional control parameter vector Collaborative input regulation network, considering the correlation of coupling coefficients, utilizes attention mechanism for adjustment. The convolution weight matrix of each convolution kernel and the... Trend feature matrix of dimension Perform 1D convolution and mapping to obtain dimensional coupled feature vector sequence Considering the aging of the control equipment, combined with dimensional control parameter vector conduct The step-by-step deduction of each recursive time step yields... dimensional efficiency vector sequence and dimensional coupled feature vector sequence After concatenation, graph convolution and temporal prediction are performed sequentially to obtain the first... each recursive time step Dimension recursive parameters and Dimensional target parameter Comparison with iterative updates dimensional control parameter vector ,get Optimal control parameter vector of dimension And sent to the control equipment, including the total number of recursive times. Determined based on the remaining duration of the current cycle.

[0096] Furthermore, the control network needs to be pre-trained by collecting data from the entire industrial production cycle of calcium stearate, including 10×2-dimensional parameter sequences, denoised parameter sequences, reaction fixed parameter sets, equipment operation data, target parameters, and parameter feedback values ​​after control, totaling 10,000 samples. This covers different raw material ratios, reaction stages, and equipment aging scenarios. The data is stratified and divided into training, validation, and test sets using a 7:2:1 stratified sampling method to ensure that the distribution of equipment aging and reaction stages in the three datasets is consistent with the original dataset. An Adam optimizer combined with mean squared error is then used. Training is performed using a difference loss function, the objective of which is to minimize the sum of squared deviations between the 20th recursive parameter and the actual feedback parameter. The initial learning rate is set to 0.001, batch size to 32, L2 regularization coefficient to 0.0001, gradient clipping threshold to 1.0, and maximum number of iterations to 500 batches. During training, training stops when the validation set loss does not decrease for 10 consecutive batches and the 20th recursive parameter of more than 95% of the samples on the test set satisfies the condition that the temperature deviation is less than or equal to 2 and the pressure deviation is less than or equal to 0.00. The structure and weights of the control network are then saved.

[0097] This invention discloses a method and system for coordinated temperature and pressure control in the calcium stearate reaction process. The system includes an acquisition module, a processing and decision module, and a control module. The acquisition module measures temperature and pressure at high frequency in each cycle to generate a parameter sequence. The processing and decision module performs Kalman filtering on the parameter sequence to obtain a denoised parameter sequence. This operation reduces measurement noise interference and improves parameter accuracy. The deviation sequence is obtained by subtracting the denoised parameter sequence from the target parameter, and the over-limit time ratio is statistically analyzed to determine whether to control. When control is needed, the control module acquires a fixed set of reaction parameters and establishes a correlation. It calculates the coupling coefficient vector sequence and concatenates it with the denoised parameter sequence to form a trend feature matrix, which can effectively capture the dynamic coupling relationship between temperature and pressure and determine the total number of recursive time steps. The control parameter vector and the trend feature matrix are input into the control network. After attention convolution, moment-by-moment extrapolation considering equipment aging, graph convolution, and time-series prediction, the control parameter vector is iteratively updated to obtain the optimal control parameter vector and the control is executed, realizing precise linkage control of temperature and pressure and ensuring the stable progress of the calcium stearate reaction.

[0098] 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 embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for coordinated temperature and pressure control in the calcium stearate reaction process, characterized in that, Includes the following steps: In the current cycle, high-frequency temperature and pressure are measured and Kalman filtering is performed to obtain a denoised parameter sequence. The difference between this sequence and the target parameter is used to obtain a deviation sequence. The ratio of out-of-limit moments is statistically analyzed to determine whether control is needed. If regulation is required, obtain the fixed set of reaction parameters, establish a correlation based on the reaction mechanism, gas state equation and heat conduction formula, calculate the coupling coefficient vector sequence based on the parameter vector of each measurement in the denoised parameter sequence and the fixed set of reaction parameters, and concatenate it with the denoised parameter sequence to form a trend feature matrix, and determine the total number of recursive time based on the remaining time. The preset control parameter vector and trend feature matrix are input into the control network. The trend feature matrix is ​​convolved using a preset number of convolution kernels adjusted by the attention mechanism and mapped into a coupled feature vector sequence. The control parameter vector and the aging status of the control equipment are combined to perform step-by-step deduction to obtain an efficiency vector sequence, which is then concatenated with the coupled feature vector sequence. The recursive parameters of the last recursive time step are obtained through graph convolution and time series prediction and compared with the target parameters to iteratively update the control parameter vector. The optimal control parameter vector is then obtained and sent to the control equipment.

2. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, The control network is based on the attention mechanism. According to the coupling strength in the coupling coefficient vector sequence, the receptive field range and weight distribution of a preset number of convolution kernels are dynamically adjusted. The adjusted convolution kernels are used to perform convolution on the trend feature matrix respectively. The convolution results are accumulated and mapped to obtain the coupling feature vector sequence.

3. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, The control network obtains the cumulative temperature control duration and cumulative pressure control duration and divides them by the corresponding equipment lifespan to obtain the temperature control aging rate and pressure control aging rate. An efficiency function is determined to describe the control efficiency of the control equipment under a given aging rate and recursive time. The aging rate and recursive time are substituted into the efficiency function and multiplied by the corresponding control parameters in the control parameter vector to obtain the temperature control effectiveness efficiency and pressure control effectiveness efficiency at each recursive time and integrate them into an efficiency vector sequence.

4. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, The control network concatenates the coupled feature vector sequence and the efficiency vector sequence to obtain a joint feature vector sequence. The joint feature vector at each recursive time step in the joint feature vector sequence is used as a node, and graph convolution operation is performed to capture the interaction of feature information between nodes to obtain a reconstructed feature vector sequence. The reconstructed feature vector sequence is input into a bidirectional gated recurrent layer for step-by-step deduction. The recursive parameter at the last recursive time step is obtained through temporal prediction and subtracted from the target parameter to obtain the deviation at the last recursive time step.

5. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 4, characterized in that, Compare the absolute value of the deviation at the last recursive time step with the deviation threshold vector. If the absolute value of the temperature deviation is greater than the temperature deviation threshold or the absolute value of the pressure deviation is greater than the pressure deviation threshold, update the control parameter vector based on the gradient descent method and re-input it into the control network until the absolute value of the temperature deviation is less than or equal to the temperature deviation threshold and the absolute value of the pressure deviation is less than or equal to the pressure deviation threshold. Then, determine the control parameter vector as the optimal control parameter vector.

6. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, Determining the coupling coefficient vector sequence includes the following steps: defining the gas expansion coefficient and heat conduction rate, obtaining the fixed reaction parameter set, constructing a pressure-temperature correlation based on the gas state equation, constructing a temperature-pressure correlation based on the calcium stearate reaction mechanism and heat conduction formula, using the end parameter of the previous cycle measured at the end of the previous cycle as the denoised parameter of the 0th measurement, calculating the difference between the denoised parameter of each measurement in the denoised parameter sequence and the denoised parameter of the previous measurement to obtain the parametric vector of each measurement, combining the fixed reaction parameter set, calculating the pressure-temperature coupling coefficient and temperature-pressure coupling coefficient corresponding to the parametric vector of each measurement, concatenating the pressure-temperature coupling coefficient and temperature-pressure coupling coefficient of each measurement to form the coupling coefficient vector of each measurement, and integrating them to form the coupling coefficient vector sequence.

7. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, The calculation of the deviation sequence and the statistical analysis of the over-limit time ratio include the following steps: extracting the denoised parameters for each measurement from the denoised parameter sequence, subtracting the target parameter from the denoised parameters to obtain the deviation for each measurement, comparing the absolute value of the deviation for each measurement with the deviation threshold vector, initializing the number of over-limit times, increasing the number of over-limit times when the absolute value of the temperature deviation is greater than the temperature deviation threshold or the absolute value of the pressure deviation is greater than the pressure deviation threshold, calculating the over-limit time ratio, which is the number of over-limit times divided by the number of measurements, and deciding whether to adjust based on the relationship between the over-limit time ratio and the time ratio threshold.

8. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 1, characterized in that, Kalman filtering includes the following steps: Define the filtered result of each measurement parameter as the denoised parameter. Initialize the denoised parameter and covariance matrix of the 0th measurement. Construct a transition equation with 1 as the main diagonal element. The secondary diagonal elements of the first and second rows are the product of heat conduction rate and time interval, and the product of gas expansion coefficient and time interval, respectively. Construct a noise matrix with the temperature variance and pressure variance as the main diagonal elements. Perform successive filtering on the parameter sequence. Calculate the predicted parameter, prediction covariance matrix, error matrix, and gain matrix for each measurement. Calculate the denoised parameter for each measurement based on the predicted parameter, gain matrix, and measurement parameter. Calculate the covariance matrix for each measurement based on the identity matrix, gain matrix, and prediction covariance matrix. Filter the parameters of each measurement sequentially to obtain the denoised parameter sequence.

9. The temperature and pressure synergistic control method for the calcium stearate reaction process as described in claim 2, characterized in that, The steps for adjusting the convolutional kernels include: extracting the coupling coefficient vector sequence from the trend feature matrix; performing global average pooling and fully connected layer mapping on the coupling coefficient vector sequence to obtain the dilation parameter; using the trend feature matrix as the query matrix and the coupling coefficient vector sequence as the key and value matrices, performing scaling dot product attention, global pooling, and fully connected layer mapping to obtain the weight adjustment matrix sequence; superimposing the initial dilation parameter with each dilation parameter in the dilation parameter; and superimposing the initial weight matrix with each weight adjustment matrix in the weight adjustment matrix sequence to obtain the dilation parameter and weight matrix of each convolutional kernel.

10. A temperature and pressure coordinated control system for the calcium stearate reaction process, characterized in that, It includes a data acquisition module, a processing and decision-making module, and a control module; The acquisition module measures temperature and pressure at high frequency in the current cycle and generates a parameter sequence. The processing and decision-making module performs Kalman filtering on the parameter sequence to obtain a denoised parameter sequence and subtracts it from the target parameter to obtain a deviation sequence. It then calculates the over-limit time ratio and decides whether to wait until the next cycle or send the denoised parameter sequence to the control module. The control module receives the denoised parameter sequence, obtains the fixed parameter set of the reaction, and establishes a correlation based on the reaction mechanism, the gas law, and the heat conduction formula. It calculates the coupling coefficient vector sequence and concatenates it with the denoised parameter sequence to form a trend feature matrix. Based on the remaining time, it determines the total number of recursive moments. The preset control parameter vector and the trend feature matrix are input into the control network. A preset number of convolution kernels adjusted by the attention mechanism are used to perform 1D convolution on the trend feature matrix and map it into a coupling feature vector sequence. Combining the control parameter vector and the aging status of the control equipment, the module performs moment-by-moment deduction to obtain the efficiency vector sequence, which is then concatenated with the coupling feature vector sequence. Through graph convolution and time-series prediction, the recursive parameters of the last recursive moment are obtained and compared with the target parameters to iteratively update the control parameter vector. The optimal control parameter vector is obtained and sent to the control equipment.

Images