A control parameter optimization method and device for multi-stage reactor

Abstract

The present application provides a control parameter optimization method and device for a multi-stage reactor. The method provided by the present application includes: obtaining aerogel production data to generate a time series data matrix; training a benchmark model based on the time series data matrix to obtain a production prediction model; obtaining an objective function, establishing a production optimization model, and calculating a first optimal control parameter based on the production optimization model; inputting the first optimal control parameter into the production prediction model, simulating the aerogel production process of a multi-stage reactor based on the production prediction model and the first optimal control parameter, and generating a prediction result; judging whether each item in the objective function meets the preset requirements based on the prediction result, and if so, taking the first optimal control parameter as the final control parameter; if not, determining the gap between the prediction result and the preset requirement, adjusting the objective function based on the gap, and returning to the step of calculating the first optimal control parameter based on the production optimization model. The method provided by the present application is used to accurately and reliably obtain the control parameters for preparing aerogels.

Description

Technical Field

[0001] The present application relates to the technical field of aerogel preparation, and in particular to a control parameter optimization method and device for a multi-stage reactor. Background Art

[0002] As a high-performance nanomaterial, aerogel has broad application prospects in aerospace, building insulation, electronic devices and other fields due to its unique structural properties, such as low density, high porosity, and low thermal conductivity.

[0003] The drying process is a key step in the preparation of aerogels, which directly determines the final performance of aerogels. At present, the drying process usually uses supercritical carbon dioxide extraction, which uses the special properties of carbon dioxide in a supercritical state to extract the ethanol solvent in the aerogel, thereby achieving the purpose of drying. This technology has the advantages of low operating temperature, no solvent residue, and no environmental pollution. However, the control parameter adjustment of the supercritical multi-stage reactor usually depends on the process formula and the experience of the operator. In this way, it is impossible to achieve precise control of key parameters in the preparation process of aerogels, which can easily lead to problems such as low production efficiency, high energy consumption, and insufficient control accuracy. Summary of the invention

[0004] In view of this, the present application provides a method and device for optimizing control parameters of a multi-stage reactor, so as to accurately and reliably obtain control parameters for preparing aerogels.

[0005] Specifically, the present application is implemented through the following technical solutions:

[0006] The first aspect of the present application provides a method for optimizing control parameters of a multi-stage reactor, the method comprising:

[0007] Acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor;

[0008] Training a benchmark model based on the time series data matrix to obtain a production prediction model;

[0009] Obtaining an objective function, establishing a production optimization model, and calculating a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles;

[0010] Inputting the first optimal control parameter into the production prediction model, simulating the aerogel production process of a multi-stage reactor based on the production prediction model and the first optimal control parameter, and generating a prediction result; wherein the prediction result includes the value of each item of the objective function;

[0011] Based on the prediction result, determine whether each item in the objective function meets the preset requirements. If so, use the first optimal control parameter as the final control parameter. If not, determine the gap between the prediction result and the preset requirements, adjust the objective function based on the gap, and return to the step of calculating the first optimal control parameter based on the production optimization model.

[0012] A second aspect of the present application provides a control parameter optimization device for a multi-stage reactor, the device comprising:

[0013] an acquisition module, a training module, a calculation module, a generation module and a processing module; wherein;

[0014] The acquisition module is used to acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor;

[0015] The training module is used to train a benchmark model based on the time series data matrix to obtain a production prediction model;

[0016] The calculation module is used to obtain an objective function, establish a production optimization model, and calculate a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles;

[0017] The generating module is used to input the first optimal control parameter into the production prediction model, simulate the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generate a prediction result; wherein the prediction result includes the value of each item of the objective function;

[0018] The processing module is used to determine whether each item in the objective function meets the preset requirements based on the prediction results. If so, the first optimal control parameter is used as the final control parameter; if not, the gap between the prediction results and the preset requirements is determined, the objective function is adjusted based on the gap, and the step of calculating the first optimal control parameter based on the production optimization model is returned.

[0019] The third aspect of the present application provides a control parameter optimization device for a multi-stage reactor, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any one of the methods provided in the first aspect of the present application are implemented.

[0020] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods provided in the first aspect of the present application.

[0021] The control parameter optimization method and device of the multi-stage reactor provided in the present application firstly solves multiple objectives by generating an optimization model to obtain the optimal control parameters that meet multiple objectives. Furthermore, the generation process of the optimal control parameters is predicted by relying on the prediction model, and the optimized control parameters under different considerations are determined by observing the changes of each objective function in the production process; the two-level optimization of multi-objective solution and simulation calculation is used to meet the optimal control parameters for both product performance and generation process, and the relationship between variables at different times in the aerogel preparation process is captured by the production prediction model to achieve accurate prediction of the production status. Furthermore, the production optimization model comprehensively considers multiple indicators to make the optimization process more comprehensive and scientific. Furthermore, the first optimal control parameter obtained by the production optimization model is combined with the production prediction model, and the production prediction model can be used to determine whether the result output by the production optimization model meets the preset requirements. In this way, through the continuous joint correction of the two models, it is ensured that accurate and reliable optimal control parameters can be obtained in the end. Controlling the generation of aerogels by optimal control parameters not only improves the safety of production, but also ensures the lowest energy consumption in the aerogel preparation process and the best aerogel preparation effect. In addition, in order to ensure the performance of the model, the method provided by the present invention constructs a time series data matrix according to various types of parameter data at various positions and time points in the multi-stage reactor, so that the generated optimization model and the prediction model can simulate the process that changes with time and position, thereby improving the accuracy of the model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of the aerogel preparation process provided for this application;

[0023] Figure 2 A flow chart of Example 1 of a control parameter optimization method for a multi-stage reactor provided in the present application;

[0024] Figure 3 A hardware structure diagram of a control parameter optimization device for a multi-stage reactor in which the control parameter optimization device for the multi-stage reactor of the present application is located;

[0025] Figure 4This is a schematic structural diagram of Example 1 of a control parameter optimization device for a multi-stage reactor provided in the present application. DETAILED DESCRIPTION

[0026] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0027] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0028] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0029] The present application provides a method and device for optimizing control parameters of a multi-stage reactor, which are used to accurately and reliably obtain control parameters for preparing aerogels.

[0030] Figure 1 Flow chart of the aerogel preparation process provided for this application. Please refer to Figure 1 Carbon dioxide is transported from an outdoor carbon dioxide storage tank to a carbon dioxide circulation storage tank through a low-temperature delivery pump, and reacts with aerogel in multiple reactors after being pressurized. The ethanol in the aerogel is continuously replaced by carbon dioxide to obtain dry aerogel.

[0031] For further information, please refer to Figure 1 The carbon dioxide overflowing from the reactor is recovered through the carbon dioxide recovery tank and then re-enters the carbon dioxide circulation tank through the recovery compressor and condenser to continue reacting with the aerogel.

[0032] For further information, please refer to Figure 1The mixture of carbon dioxide and ethanol discharged from the reactor is filtered and separated through multiple separation kettles. The separated ethanol is stored in a collecting tank, and the separated carbon dioxide is condensed by a condenser and then sent back to the carbon dioxide circulation storage tank to continue reacting with the aerogel.

[0033] In this embodiment, the ethanol in the aerogel is continuously replaced by carbon dioxide, so that the aerogel can be dried and the aerogel production can be completed.

[0034] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0035] Figure 2 This is a flow chart of Example 1 of the control parameter optimization method for a multi-stage reactor provided in this application. Figure 2 , the method provided in this embodiment may include:

[0036] S201. Acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor.

[0037] Specifically, the aerogel production data is the data generated during the preparation process of the aerogel. It should be noted that the specific data types included in the aerogel production data are set according to actual needs, and are not limited in this embodiment. In specific implementation, the temperature data, pressure data, composition data, concentration data, etc. at different positions of the aerogel during the preparation process can be determined as the aerogel production data.

[0038] Furthermore, aerogels can be prepared in a multi-stage reactor. It is understandable that since aerogels flow through various positions of the multi-stage reactor during the preparation process, the aerogel production data should include data of various positions of the multi-stage reactor.

[0039] Optionally, the production data of each position of the multi-stage reactor includes: drying preheater data, drying kettle data, drying kettle outlet data, separation preheater data, first separator data, second separator data, circulation storage tank data and booster pump data.

[0040] Specifically, the drying preheater, separation preheater, first separator, second separator and circulation storage tank are all locations passed through during the aerogel preparation process. It should be noted that when obtaining production data, since parameters such as ethanol content are also important indicators, relevant data of ethanol can also be obtained.

[0041] Furthermore, the production data of each position can be obtained through sensors at each position of the multi-stage reactor, and combined to obtain the aerogel production data.

[0042] Furthermore, the time series data matrix is ​​composed of part of the aerogel production data. It should be noted that the time series data matrix only contains part of the aerogel production data, in other words, the amount of data in the time series data matrix is ​​less than that in the aerogel production data.

[0043] Furthermore, the time series data matrix is ​​stored based on multiple data vectors, each of which is the production data corresponding to each position of the multi-stage reactor at a time point.

[0044] It should be noted that the data vectors in the time series data matrix are arranged in chronological order from earliest to latest. Since each data vector is the data at a time point, after different data vectors are arranged in chronological order to form a time series data matrix, the changes of aerogel during the preparation process can be judged through the time series data matrix.

[0045] A specific embodiment is given below to introduce the acquisition process of time series matrix data in detail:

[0046] (1) Acquiring the aerogel production data based on multiple preset time step data and collecting data at multiple time points.

[0047] Specifically, the specific duration of the preset time step is determined according to actual needs, and is not limited in this embodiment. For example, in one embodiment, the preset time step is 30 minutes.

[0048] In specific implementation, in this embodiment, production data of each position of the multi-stage reactor can be collected every preset time step to obtain a data vector corresponding to a time point. Production data of the entire aerogel production process is collected to obtain data at all time points.

[0049] It should be noted that the specific number of preset time step data is determined according to actual needs, and is not limited in this embodiment. For example, in one embodiment, there may be 6 preset time step data.

[0050] (2) Generating a plurality of reconstruction rules and corresponding weight matrices based on the preset time step data; wherein each of the reconstruction rules corresponds to a weight vector of one dimension in the weight matrix.

[0051] Specifically, the time point data changes periodically and has a periodic change trend. In specific implementation, the periodic change trend of the time point data can be a decrease of the same value every two hours, or an increase of the same value every three hours.

[0052] Furthermore, corresponding reconstruction rules may be generated according to the specific duration of the preset time step data. It should be noted that, since each preset time step data corresponds to a different duration, the number of generated reconstruction rules is the same as the number of preset time step data.

[0053] Furthermore, the number of weight vectors included in the weight matrix corresponds to the number of reconstruction rules, and each dimension of the weight vector corresponds to one reconstruction rule.

[0054] It should be noted that each weight vector in the weight matrix has the same value initially.

[0055] (3) The weight matrix is ​​transformed based on the trend law corresponding to the data at the time point to obtain the target weight, the target reconstruction rule is determined based on the target weight, and the time series data matrix is ​​generated based on the target reconstruction rule.

[0056] Specifically, after the weight matrix undergoes linear transformation, the values ​​of the weight vectors in the weight matrix change.

[0057] Furthermore, the target weight can be determined by the value of the weight vector. In specific implementation, the weight vector with the largest value in the weight matrix can be directly determined as the target weight.

[0058] Furthermore, the reconstruction rule corresponding to the target weight can be determined as the target reconstruction rule, and the time step for obtaining the data vector is determined based on the target reconstruction rule. The data vector is extracted once every the time interval to obtain a time series data matrix.

[0059] The control parameter optimization method of the multi-stage reactor provided in this embodiment first obtains aerogel production data based on multiple preset time step data, collects multiple time point data, and then generates multiple reconstruction rules based on the preset time step data, and finally transforms the weight matrix based on the trend law corresponding to the time point data to obtain the target weight, determines the target reconstruction rule based on the target weight, and generates a time series data matrix based on the target reconstruction rule. In this way, the weight vector with the highest value is obtained by screening the weight matrix, and it is determined as the target weight. Through the target reconstruction rule corresponding to the target, the acquisition of the time series data matrix can be completed. Since the target weight has the highest value, the time series data matrix can be closest to the parameter changes in the aerogel preparation process, which can ensure the accurate acquisition of the aerogel control parameters.

[0060] A specific embodiment is given below to explain in detail the method for obtaining trend rules:

[0061] Step 1: transform the aerogel production data to obtain frequency data.

[0062] Specifically, the frequency data can be generated based on a traditional data transformation method, or based on a neural network data transformation method. In this embodiment, this is not limited. In specific implementation, the aerogel production data can be converted into frequency data through Fourier transformation, and then the frequency data with a frequency higher than a preset threshold can be selected, and these data are inversely Fourier transformed, and the obtained data is high-frequency data.

[0063] Step 2: remove the high-frequency data in the frequency data to obtain low-frequency data, wherein the high-frequency data is data with a frequency higher than a preset value, and the low-frequency data is data with a frequency less than or equal to a preset value.

[0064] Specifically, the specific numerical value of the preset value is determined according to actual needs, and is not limited in this embodiment.

[0065] In specific implementation, high-frequency data in the aerogel production data is deleted to obtain low-frequency data. For example, in one embodiment, low-frequency data can be obtained by the following formula:

[0066] XRemaining = X-XFrequency;

[0067] Among them, XRemaining is the low-frequency data, X is the aerogel production data, and XFrequency is the high-frequency data.

[0068] Step three: extract the trend characteristics of the low-frequency data to generate trend data.

[0069] Specifically, the trend characteristics characterize the long-term periodic law of data changes during the production and preparation of aerogels, and the trend characteristics can indicate the changes of aerogels during the preparation process. It should be noted that obtaining trend characteristics through low-frequency data can prevent the influence of short-term sudden changes on long-term laws.

[0070] In specific implementation, trend data can be obtained through the following formula:

[0071] ;

[0072] Among them, XTendency is the trend data, and XRemaining is the low-frequency data. In this formula, the trend features are extracted from the low-frequency data through the Softmax function and the average pooling (i.e., the Average Pooling in the formula) operation to construct the trend data.

[0073] Step 4: Determine the value of the target weight based on the high-frequency data and the trend data, and calculate the target reconstruction rule based on the value of the target weight.

[0074] In specific implementation, according to the correspondence between the data vector and the high-frequency data and trend data, the data vector corresponding to the high-frequency data can be set to a smaller target weight, and the data vector corresponding to the trend data can be set to a larger target weight.

[0075] For example, in one embodiment, the target weight may be calculated based on the following formula:

[0076] ;

[0077] Among them, W r is a predefined weight matrix, R (Xtrans) is the matrix obtained by transforming the predefined weight matrix, Xtrans is the trend data, Wnoise is a trainable parameter, is the noise that obeys the normal distribution. In this formula, the nonlinear matrix transformation is realized by the Softplus function.

[0078] The control parameter optimization method of the multi-stage reactor provided in this embodiment can obtain accurate high-frequency data and low-frequency data through calculations such as Fourier transform and inverse Fourier transform. Further, by obtaining high-frequency data and low-frequency data, obtaining short-term changes through high-frequency data, and obtaining long-term trend changes through low-frequency data, the influence of noise that affects the aerogel preparation rules in a short period of time is effectively eliminated, ensuring that the obtained time series data matrix can effectively reflect the regular changes in the aerogel preparation process. Further, by supervising through high-frequency data and low-frequency data, and determining the final target weight through target input parameters and target output parameters, the target reconstruction rules can be obtained, and the time series data matrix can be effectively and reliably generated.

[0079] Another specific embodiment is given below to introduce the process of obtaining the target reconstruction rule in detail:

[0080] Step 1: Obtain the trend law of the data at the time point, and generate a transformation sample based on the trend law; wherein the trend law includes the frequency distribution of data of the same dimension at the same position; and the transformation sample includes target input parameters and target output parameters.

[0081] Specifically, the trend law of the time point data is the change law of the data of one dimension at one location. It should be noted that the trend law can be a continuous increase or a continuous decrease.

[0082] In specific implementation, for example, in one embodiment, the trend rule can be determined based on the overall change of the pressure value at the dry preheater of the circulation tank during the aerogel preparation process.

[0083] Furthermore, when obtaining trend rules, multiple data type transformations are required. The weight matrix composed of the initial weights can be transformed together with the aerogel production data to obtain the transformed weight matrix.

[0084] Furthermore, the transformation sample includes a target input parameter and a target output parameter. It should be noted that by inputting the transformation sample into the weight matrix, the value of the weight vector is changed through the weight matrix, so that the input target input parameter can be converted into the target output parameter under the influence of the weight matrix. The specific values ​​of the target input parameter and the target output parameter can be pre-set by the staff, and this is not limited in this embodiment.

[0085] In specific implementation, transformation samples can be generated based on target input parameters, target output parameters and trend rules.

[0086] Step 2: transform the weight matrix based on the transformed samples, determine the target weight based on the value of the weight vector in the transformed weight matrix, and determine the target reconstruction rule based on the target weight; wherein the target reconstruction rule limits the time interval between two adjacent data vectors.

[0087] Specifically, the transformed samples are input into the weight matrix, so that the weight matrix learns the connection between the target input parameters and the target output parameters under the constraint of the trend law, and adjusts the value of the weight vector.

[0088] Furthermore, in the transformed weight matrix, the weight value of the weight vector changes, and the weight with the largest weight value is determined as the target weight. It should be noted that the larger the target weight value, the more the corresponding target reconstruction rule can reflect the trend change and periodic law during the preparation of aerogels.

[0089] In specific implementation, the corresponding target reconstruction rule is determined by the target weight, and the constructed target reconstruction rule can obtain the data vector through the corresponding preset time step data. For example, in one embodiment, according to the target weight, it is determined that the target weight of the time point data collected every 3 hours is the largest, so the time point data collected every 3 hours are combined in the order of the collection time to obtain a time series data matrix.

[0090] In specific implementation, a time step can be set, and different target reconstruction rules are based on the time step. For example, in one embodiment, the time step is defined as 1 hour, and the reconstruction rule can be defined as: each element of the time series is 2 hours apart (i.e., 2 time steps); each element of the time series is 3 hours apart (i.e., 3 time steps); each element of the time series is 6 hours apart (i.e., 6 time steps); each element of the time series is 12 hours apart (i.e., 12 time steps); the above rules are modeled, and a rule matrix can be defined, and each dimension of the rule matrix represents a reconstruction rule. At this time, the rule matrix is ​​defined as 2, 3, 6, 12, and the unit is the time step.

[0091] The control parameter optimization method for a multi-stage reactor provided in this embodiment determines a target reconstruction rule through a target weight, and obtains a time series data matrix according to a certain time step through the target reconstruction rule. In this way, not only can valid data be obtained according to the trend characteristics of aerogel production data, but also irrelevant noise data can be eliminated, thereby ensuring the reliability and effectiveness of the time series data matrix.

[0092] S202: Train a benchmark model based on the time series data matrix to obtain a production prediction model.

[0093] Specifically, the production prediction model can predict the equipment energy consumption and drying effect and other preparation results in the aerogel preparation process by inputting control parameters and environmental data.

[0094] Furthermore, the benchmark model has a designed neural network architecture and preliminary parameters. In specific implementation, the time series data can be divided into a training set and a validation set, the training set can be input into the benchmark model, and the prediction results output by the benchmark model can be corrected based on the validation set. It should be noted that as the training process proceeds, the benchmark model will automatically adjust its internal parameters to ensure that the prediction error is minimized, and finally a production prediction model that has completed training is obtained.

[0095] Optionally, the predictive correlation relationship is obtained based on the trend law, and the benchmark model is trained based on the predictive correlation relationship to obtain the production prediction model; wherein the predictive correlation relationship represents the association between the control parameters at different time points; and the maximum weight of the predictive correlation relationship corresponds to the energy consumption data in the objective function.

[0096] Specifically, the predicted association relationship is used to associate the control parameters corresponding to different time points. In other words, the future control parameters can be determined according to the current control parameters through the predicted association relationship.

[0097] Furthermore, the production prediction model is used to determine the preparation effect of aerogel after a period of time based on the currently input control parameters. Therefore, it is necessary to prioritize the preparation of aerogel. In other words, the product quality data is determined as the maximum weight in the prediction association, and the weight corresponding to other data such as equipment energy consumption is set to be smaller than the weight of product quality.

[0098] In this step, the production prediction model can effectively capture the prediction correlation within the time series data matrix and realize the prediction of aerogel production results. Furthermore, the production prediction model predicts the production result data at future moments based on the input control parameters and environmental data through the encoding and decoding mechanism. In this way, it can provide accurate and reliable support for the acquisition of aerogel preparation control parameters.

[0099] S203, obtaining an objective function, establishing a production optimization model, and calculating a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles.

[0100] Specifically, the data in the objective function are the result data of the aerogel production process. In specific implementation, the corresponding equipment energy consumption data, production efficiency, and aerogel drying effect data after the aerogel preparation process is completed can be determined as the data in the objective function.

[0101] It is understandable that the objective function includes data from multiple angles for evaluating the aerogel preparation effect, and each angle can be represented by an objective item. For example, in one embodiment, in combination with the above embodiments, the objective items can include equipment energy consumption, production efficiency, and aerogel drying effect.

[0102] Furthermore, the production optimization model is used to comprehensively consider all objective items and generate the first optimal control parameter under the premise of balancing all objective items.

[0103] Furthermore, the first optimal control parameter is the best control parameter for preparing aerogels based on the objective function. It can be understood that preparing aerogels according to the first optimal control parameter can ensure that the target items at various angles in the objective function are well achieved.

[0104] A specific example is given below to introduce the preparation process of the objective function in detail:

[0105] Optionally, the first importance of the target items at the multiple different angles is determined based on product performance, and weight values ​​are set according to the first importance to establish the objective function, and the first importance of the target items at the different angles is different.

[0106] Specifically, the first importance is an indicator of the degree of influence of the target item on aerogel preparation reflected by product performance.

[0107] Furthermore, product performance is data that characterizes the performance of aerogels. It is understandable that the greater the impact on product performance, the greater the first importance of the target item, and the higher the corresponding weight value. It should be noted that during the preparation of aerogels, different angles have different impacts on product performance, and target items at different angles can be set to different weight values.

[0108] The control parameter optimization method for a multi-stage reactor provided in this embodiment uses weight values ​​to characterize the first importance of target items at different angles, and can obtain the first optimal control parameters in a targeted manner through a production optimization model. In this way, different weight values ​​are set for target items at different angles, ensuring the accuracy and reliability of the first optimal control parameters.

[0109] A specific embodiment is given below to introduce in detail the process of obtaining the first optimal control parameter by the production optimization model:

[0110] (1) Initializing a target particle swarm and updating particle parameters of the target particle swarm based on the objective function constraint; wherein the target particle swarm represents a set of multiple groups of control parameters, and each particle in the target particle swarm represents a group of the control parameters.

[0111] Specifically, the target particle swarm is a set of potential solutions, and each particle in the target particle swarm represents a set of control parameters.

[0112] Furthermore, the objective function is used as a constraint to continuously iterate and update the position and velocity of each particle in the target particle swarm to search for the optimal solution.

[0113] It should be noted that in the particle swarm optimization algorithm, each particle has a set of parameters, which correspond to the control parameters in the aerogel production process. The parameters of the particles will be continuously updated during the optimization process to find the optimal combination of control parameters.

[0114] In this step, a certain number of particles need to be randomly generated in the solution space to form a target particle swarm and complete the initialization of the target particle swarm.

[0115] (2) Calculating a global optimal solution based on the particle parameters updated a specified number of times, and determining the global optimal solution as the first optimal control parameter.

[0116] Specifically, the specific value of the specified quantity is determined according to actual needs, and this embodiment does not limit this. For example, in one embodiment, the specific value of the specified quantity is determined based on historical experience.

[0117] Furthermore, the global data is comprehensively analyzed by updating the target particle swarm to obtain a global optimal solution, and the global optimal solution is determined as the first optimal control parameter.

[0118] The control parameter optimization method of the multi-stage reactor provided in this embodiment comprehensively considers the energy consumption factor, finds the global optimal solution that can both meet the production demand and minimize the energy consumption through the target particle group, and then obtains the first optimal control parameter, which can effectively reduce the energy consumption in the production process and meet the preparation requirements of the aerogel. Furthermore, the target particle group can be dynamically adjusted, so it can better cope with the uncertainty and change factors in the production process and ensure the stability and reliability of the production process.

[0119] Optionally, an optimization association relationship between the objective function and the first optimal control parameter is constructed, and the production optimization model is established based on the optimization association relationship; wherein the objective function includes power data, energy consumption data and product quality data; and the maximum weight of the optimization association relationship corresponds to the product quality data.

[0120] Specifically, the product quality data reflects the effect of aerogel preparation. In specific implementation, the product quality data can be determined based on the drying effect data of the aerogel.

[0121] Furthermore, by optimizing the correlation relationship to establish a production optimization model, the corresponding first optimal control parameter can be obtained based on the objective function through the production optimization model, which provides strong support for the determination of control parameters in the aerogel production process.

[0122] S204. Input the first optimal control parameter into the production prediction model, simulate the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generate a prediction result; wherein the prediction result includes the value of each item of the objective function.

[0123] In this step, when it is specifically implemented, the production prediction model simulates the aerogel preparation process under the first optimal control parameter to obtain a prediction result. It should be noted that the prediction result represents the result obtained by preparing the aerogel under the first optimal control parameter.

[0124] S205. Determine whether each item in the objective function meets the preset requirements based on the prediction result. If so, use the first optimal control parameter as the final control parameter. If not, determine the gap between the prediction result and the preset requirements, adjust the objective function based on the gap, and return to the step of calculating the first optimal control parameter based on the production optimization model.

[0125] Specifically, the specific content of the preset requirements is determined according to actual needs, and is not limited in this embodiment. For example, the preset requirements for aerogel preparation can be determined based on historical experience.

[0126] Furthermore, when the prediction result meets the preset requirements, the first control parameter at this time can meet the needs of actual aerogel preparation, so the first control parameter can be directly determined as the final control parameter; when the prediction result does not meet the preset requirements, the part that does not meet the preset requirements will be recorded and the objective function will be adjusted. In this way, the first optimal control parameter can be obtained again through the adjusted objective function, which can overcome the problems encountered in this simulation and ensure the accuracy of the results.

[0127] A specific embodiment is given below to introduce in detail the process of judging whether each item in the objective function meets the preset requirements:

[0128] Optionally, the second importance of the target items at the multiple different angles is determined based on the production process, and the component items of the preset requirements are determined according to the second importance, and the most important target items in the first importance ranking and the second importance ranking are different.

[0129] Specifically, the second importance is an indicator of the degree of influence of the target item determined by the production process on the aerogel preparation.

[0130] It should be noted that the preset requirement may correspond to one target item or multiple target items, and this is not limited in this embodiment.

[0131] Furthermore, the production process of aerogels determines the importance of different aspects of aerogels. In specific implementation, in one production process, the energy consumption of the equipment can be the most important target item of this production process; in another production process, the drying effect of the aerogel can be the most important target item of this production process.

[0132] Furthermore, the results of representative items in the predicted results are compared with the data of representative items in the preset results to determine whether they meet or are better than the preset results.

[0133] Furthermore, the representative term is different from the term with the largest weight value in the objective function. It should be noted that the term with the largest weight value is the term that the production process must satisfy, so there is no need to consider the preparation effect of the term with the largest weight value, and only the preparation of the target term needs to be judged. For example, in one embodiment, the term with the largest weight value is the drying effect. At this time, the drying effect of the aerogel must satisfy the preset conditions, so it is necessary to judge whether the target term (energy consumption of the equipment) is lower than the equipment energy consumption required by the preset conditions through the prediction results.

[0134] The control parameter optimization method and device of the multi-stage reactor provided in this embodiment first obtains the aerogel production data to generate a time series data matrix; then trains the benchmark model based on the time series data matrix to obtain the production prediction model; then obtains the objective function, establishes the production optimization model, and calculates the first optimal control parameter based on the production optimization model; then inputs the first optimal control parameter into the production prediction model, simulates the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generates a prediction result; finally, based on the prediction result, it is judged whether each item in the objective function meets the preset requirements. If it does, the first optimal control parameter is used as the final control parameter; if it does not meet, the gap between the prediction result and the preset requirements is determined, the objective function is adjusted based on the gap, and the step of calculating the first optimal control parameter based on the production optimization model is returned. In this way, the relationship between variables at different times in the aerogel preparation process is captured by the production prediction model, and the accurate prediction of the production status is achieved. Furthermore, the production optimization model comprehensively considers multiple indicators, making the optimization process more comprehensive and scientific. Furthermore, by combining the first optimal control parameter obtained by the production optimization model with the production prediction model, the production prediction model can be used to determine whether the result output by the production optimization model meets the preset requirements. In this way, by continuously jointly correcting the two models, it is ensured that accurate and reliable optimal control parameters can be obtained in the end. Furthermore, by judging whether the prediction results meet the preset requirements through representative items, it is ensured that the energy consumption of the equipment is saved under the premise of successful preparation of the aerogel, and the preparation effect of the aerogel can be improved. In this way, through accurate prediction models and real-time regulation, the key parameters in the aerogel drying process can be better controlled, the variability and uncertainty in the production process can be reduced, thereby improving the uniformity and stability of the product, and finally reliably preparing the aerogel.

[0135] A specific example is given below to introduce in detail the process of obtaining aerogel production data:

[0136] Specifically, the production data and control parameters of the aerogel drying process are collected, and the collected data are preprocessed. In specific implementation, the aerogel drying process adopted by the aerogel can be: the modified aerogel is sent into the reactor, and the ethanol in the aerogel is extracted under high pressure by using the carbon dioxide supercritical extraction technology, and the aerogel is sent out of the reactor after the reaction. The extracted gas enters the separation kettle to separate the carbon dioxide and ethanol.

[0137] Furthermore, the categories of control variable data recorded by the control program are shown in Table 1:

[0138] Table 1

[0139] Data Category describe time Data collection time Drying kettle electric heating water tank temperature (℃) Controlling the extraction reaction temperature Temperature of electric heating water tank of separation kettle (℃) Controlling separation reaction temperature Water cooling unit temperature (℃) Maintain constant temperature Booster pump (mPa) Controlling the CO2 flow rate Pressure of carbon dioxide circulation tank (mPa) Controlling extraction reaction conditions Temperature of carbon dioxide circulation tank (℃) Controlling extraction reaction conditions Dioxide booster pump frequency (Hz) Ensure constant extraction reaction pressure CO2 Booster Pump Flow Ensure a constant carbon dioxide content in the extraction

[0140] Furthermore, the environmental parameters of the drying process collected by high-precision sensors are shown in Table 2:

[0141] Table 2

[0142] Data Category describe Drying kettle 1 pressure (mPa) Extraction reaction real-time pressure Drying kettle 1 temperature (℃) Extraction reaction real-time temperature Drying kettle 2 temperature (℃) Extraction reaction real-time temperature Drying kettle 2 pressure (mPa) Extraction reaction real-time pressure Drying kettle outlet temperature (℃) Temperature of carbon dioxide and ethanol mixed gas Drying kettle outlet pressure (mPa) Carbon dioxide and ethanol mixed gas pressure Separation preheater pressure (mPa) Separation reaction pressure lower limit Separation preheater temperature (℃) Lower limit of separation reaction temperature Separation kettle 1 pressure (mPa) Separation reaction real-time pressure Separation kettle 1 temperature (℃) Separation reaction real-time temperature Separation kettle 2 pressure (mPa) Separation reaction real-time pressure Separation kettle 2 temperature (℃) Separation reaction real-time temperature Liquid level of carbon dioxide circulation tank (mm) Reflects the carbon dioxide separation content Separation kettle 1 ethanol liquid level height (cm) Drying effect index Ethanol liquid level height in separation kettle 2 (cm) Drying effect index power consumption Energy consumption index

[0143] Furthermore, after preprocessing the data in Table 1 and Table 2, neural network technology is used to construct a functional relationship model between environmental parameters, control variables, energy consumption, and drying effects. The constructed production prediction model includes a data processing module and a prediction module.

[0144] It should be noted that the data processing module is responsible for reconstructing disordered data to assist in the training step and effectively learn the key information of the data; the prediction module is responsible for effectively capturing the internal dependencies of the sequence data, and the prediction module is divided into two mechanisms: encoding and decoding. The encoding mechanism receives the reconstructed time series data and encodes it into a vector matrix containing many change rules. The decoding mechanism predicts the data at the next moment based on the vector matrix encoded at the current moment. The parameter weights of the two mechanisms are continuously improved through back propagation during the training process to achieve the fit between the predicted value and the true value.

[0145] Furthermore, the production forecast model can include a forecast module, which captures many change rules and encodes them together with the input data into a vector matrix. Furthermore, the decoding process is an autoregressive generation process, that is, predicting the data at the next moment based on the encoded matrix in the current minimum time unit.

[0146] A specific embodiment is given below to introduce the modeling process of the prediction module in the production prediction model in detail:

[0147] Specifically, the prediction module is modeled based on the following formula:

[0148] ;

[0149] ;

[0150] Where x(t) is the input value, h(t) is the state vector of the system at time t, and h'(t) is the derivative of the state vector h(t) with respect to time t.

[0151] It should be noted that h(t) can represent the changing pattern of data at the current time; h'(t) can represent the trend of the changing pattern.

[0152] Furthermore, It is a state equation, which is used to describe the regularity of the data at the current time point and the trend of regular changes.

[0153] In specific implementation, in the linear time-varying equation, A(t) is a matrix that changes with time and is used to describe the internal dynamic relationship between system states. In the present invention, by fixing the parameter A, the state change is limited to a fixed time scale, that is, when it fits actual production, the law of data change within a period of time remains unchanged, ensuring that the short-term data law is fully mined during the training process.

[0154] Furthermore, h(t-1) represents the state vector of the system at time t-1, which indicates that the current state of the system depends on the state at the previous moment. That is, during the training process, the dependency relationship between data can be established through time changes; and the current state change depends on the state at the previous moment, realizing mathematical modeling of the change transmission properties between data. In the process of model training, the inherent trends between data change laws can be fully explored.

[0155] Furthermore, B(t) is a time-varying matrix that describes how the input vector x(t) affects the state of the system and represents the impact of current data on short-term data patterns.

[0156] Furthermore, the prediction results can be obtained by the following formula:

[0157] ;

[0158] Where x(t) is the input vector at time t, y(t) is the output vector at time t, C(t) is a time-varying matrix, h(t) is the state vector at time t, and D(t) is a time-varying matrix.

[0159] Furthermore, C(t) is used to describe how the system state is mapped to the output. That is, the data change law is mapped to the data of the current time node, ensuring that the data of the current node not only contains the existing data information, but also has the vector information of the current data change law. h(t) is obtained by the previous state equation and represents the change law in the current time period. D(t) defines the impact of the input data x(t) on the predicted value y(t). D(t) ensures that the prediction not only fully considers the information representing the change law, but also considers the characteristics of the input data, so that it will not be covered by a large amount of regular information.

[0160] Furthermore, the objective function needs to comprehensively consider multiple indicators such as production efficiency, energy consumption and product quality. The objective function can be calculated by the following formula:

[0161] ;

[0162] Where x is the control parameter vector, W 1 , W 2 , W 3is the weight coefficient, Efficiency, EnergyConsumption and Quality are the evaluation functions of production efficiency, energy consumption and product quality respectively.

[0163] Furthermore, after initializing the particle swarm, each particle represents a potential control parameter solution. The limiting conditions of the position information using the process conditions are shown in Table 3:

[0164] Table 3

[0165] Data Category Process requirements Drying kettle temperature Around 64.5℃ Drying kettle pressure strength 16.4MPa up and down Drying kettle total outlet valve opening 65 degrees Separation kettle temperature 5.3MPa up and down Separation kettle pressure Around 40℃ Carbon dioxide circulation tank temperature Around 12℃ Carbon dioxide circulation tank pressure About 5MPa CO2 Booster Pump Frequency 18Hz up and down

[0166] Furthermore, iterative updating of each example in the target particle group may include the following steps: calculating the objective function value F(x) of each particle and evaluating the performance of the current particle; then updating the individual optimal solution of each particle, that is, the optimal objective function value achieved by each particle in its historical position; then updating the global optimal solution, that is, the optimal objective function value achieved by all particles in all iterations; finally, updating the velocity and position of the particle according to the individual optimal solution and the global optimal solution.

[0167] Furthermore, the trained model is deployed to the edge control equipment of the aerogel drying production line. In specific implementation, the equipment can use the SX5 series of industrial intelligent machines from Zhongke Times, which have two types of computing modules: non-real-time system and real-time system. The real-time system deploys the PLC control program, and the non-real-time system deploys the production prediction model and production optimization model.

[0168] A specific embodiment is given below to introduce the specific production process of the edge control device in detail:

[0169] Specifically, the high-precision sensors in the reactor interact with the PLC control program in the real-time system of the IIoT machine to collect the control parameters and environmental parameters of the current production environment in real time; the relevant parameters are transmitted to the production optimization model in the non-real-time system through the TCP transmission protocol and shared memory technology; the production optimization model is used to solve the optimization control parameters based on the existing environmental parameters and the process conditions of the aerogel drying process as restrictions; the quality prediction and risk warning of the current optimization parameters are carried out in conjunction with the prediction model, and it is confirmed whether the production optimization model needs to be re-called based on the feedback information. The prediction model is used as the verification standard to obtain the optimal control parameters; based on the calculated optimal parameters, the non-real-time system interacts with the PLC program through the TCP transmission protocol and shared memory technology to adjust the temperature, pressure and other control parameters of the supercritical multi-stage reactor; the adjusted process parameters and product extraction quality are recorded for further optimization of subsequent models.

[0170] Corresponding to the aforementioned embodiment of a method for optimizing control parameters of a multi-stage reactor, the present application also provides an embodiment of a device for optimizing control parameters of a multi-stage reactor.

[0171] The embodiment of the control parameter optimization device for a multi-stage reactor of the present application can be applied to the control parameter optimization equipment for a multi-stage reactor. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, the processor of the control parameter optimization equipment for the multi-stage reactor in which it is located reads the corresponding computer program instructions in the non-volatile memory into the memory and runs them. From the hardware level, if Figure 3 As shown, it is a hardware structure diagram of the control parameter optimization device of the multi-stage reactor in the present application, except Figure 3 In addition to the processor, memory, network interface, and non-volatile memory shown, the control parameter optimization device of the multi-stage reactor in which the device is located in the embodiment can also include other hardware, which will not be described in detail.

[0172] Figure 4 This is a schematic diagram of the structure of the first embodiment of the control parameter optimization device for a multi-stage reactor provided in this application. Figure 4 , the device provided in this embodiment includes an acquisition module 410, a training module 420, a calculation module 430, a generation module 440 and a processing module 450; wherein;

[0173] The acquisition module 410 is used to acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor;

[0174] The training module 420 is used to train a benchmark model based on the time series data matrix to obtain a production prediction model;

[0175] The calculation module 430 is used to obtain an objective function, establish a production optimization model, and calculate a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles;

[0176] The generating module 440 is used to input the first optimal control parameter into the production prediction model, simulate the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generate a prediction result; wherein the prediction result includes the value of each item of the objective function;

[0177] The processing module 450 is used to determine whether each item in the objective function meets the preset requirements based on the prediction results. If so, the first optimal control parameter is used as the final control parameter; if not, the gap between the prediction results and the preset requirements is determined, the objective function is adjusted based on the gap, and the step of calculating the first optimal control parameter based on the production optimization model is returned.

[0178] The device of this embodiment can be used to perform Figure 1 The steps, specific implementation principles and implementation processes of the method embodiment shown are similar and will not be repeated here.

[0179] Optionally, the calculation module 430 is specifically configured to determine the first importance of the target items at the multiple different angles based on product performance, and to set weight values ​​according to the first importance to establish the objective function, wherein the first importance of the target items at the different angles is different;

[0180] The calculation module 430 is also specifically used to determine the second importance of the target items of the multiple different angles based on the production process, and determine the component items of the preset requirements according to the second importance, and the most important target items in the first importance ranking and the second importance ranking are different.

[0181] Optionally, the acquisition module 410 is specifically used to acquire the aerogel production data based on multiple preset time step data, and collect data at multiple time points;

[0182] The acquisition module 410 is further specifically used to generate a plurality of reconstruction rules and corresponding weight matrices based on the preset time step data; wherein each of the reconstruction rules corresponds to a weight vector of one dimension in the weight matrix;

[0183] The acquisition module 410 is further specifically used to transform the weight matrix based on the trend law corresponding to the time point data to obtain the target weight, determine the target reconstruction rule based on the target weight, and generate the time series data matrix based on the target reconstruction rule.

[0184] Optionally, the acquisition module 410 is further specifically used to acquire the trend law of the data at the time point, and generate a transformation sample in combination with the trend law; wherein the trend law includes the frequency distribution of the data of the same dimension at the same position; and the transformation sample includes the target input parameter and the target output parameter;

[0185] The acquisition module 410 is also specifically used to transform the weight matrix based on the transformed sample, determine the target weight based on the numerical value of the weight vector in the transformed weight matrix, and determine the target reconstruction rule based on the target weight; wherein the target reconstruction rule limits the time interval between two adjacent data vectors.

[0186] Optionally, the acquisition module 410 is further specifically used to transform the aerogel production data to acquire frequency data;

[0187] The acquisition module 410 is further specifically used to remove the high-frequency data in the frequency data to obtain low-frequency data, wherein the high-frequency data is data with a frequency higher than a preset value, and the low-frequency data is data with a frequency less than or equal to a preset value;

[0188] The acquisition module 410 is further specifically used to extract the trend characteristics of the low-frequency data and generate trend data;

[0189] The acquisition module 410 is further specifically configured to determine the value of the target weight based on the high-frequency data and the trend data, and calculate the target reconstruction rule based on the value of the target weight.

[0190] Optionally, the calculation module 430 is specifically used to initialize the target particle group and update the particle parameters of the target particle group based on the objective function constraint; wherein the target particle group represents a set of multiple groups of control parameters, and each particle in the target particle group represents a group of the control parameters;

[0191] The calculation module 430 is further specifically configured to calculate a global optimal solution based on the particle parameters updated a specified number of times, and determine the global optimal solution as the first optimal control parameter.

[0192] Optionally, the calculation module 430 is also specifically used to construct an optimization association relationship between the objective function and the first optimal control parameter, and establish the production optimization model based on the optimization association relationship; wherein the objective function includes power data, energy consumption data and product quality data; the maximum weight of the optimization association relationship corresponds to the product quality data.

[0193] Optionally, the acquisition module 410 is also specifically used to obtain the predictive correlation relationship based on the trend law, train the benchmark model based on the predictive correlation relationship, and obtain the production prediction model; wherein the predictive correlation relationship represents the association between the control parameters at different time points; and the maximum weight of the predictive correlation relationship corresponds to the energy consumption data in the objective function.

[0194] Optionally, the production data of each position of the multi-stage reactor includes: dryer preheater data, dryer data, dryer outlet data, separation preheater data, first separator data, second separator data, circulation storage tank data, booster pump data and ethanol data.

[0195] Please continue to refer to Figure 3 The present application also provides a control parameter optimization device for a multi-stage reactor, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any one of the methods provided in the first aspect of the present application are implemented.

[0196] The present application also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of any one of the methods provided in the present application are implemented.

[0197] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0198] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0199] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A control parameter optimization method for a multi-stage reactor, characterized in that: The method comprises: Acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor; The step of obtaining aerogel production data to generate a time series data matrix comprises: Acquire the aerogel production data based on multiple preset time step data, and collect data at multiple time points; Generate multiple reconstruction rules and corresponding weight matrices based on the preset time step data; wherein each of the reconstruction rules corresponds to a weight vector of one dimension in the weight matrix; Transform the weight matrix based on the trend law corresponding to the time point data to obtain a target weight, determine a target reconstruction rule based on the target weight, and generate the time series data matrix based on the target reconstruction rule; The step of transforming the weight matrix based on the trend law corresponding to the data at the time point to obtain the target weight includes: Transforming the aerogel production data to obtain frequency data; Removing high-frequency data from the frequency data to obtain low-frequency data, wherein the high-frequency data is data with a frequency higher than a preset value, and the low-frequency data is data with a frequency less than or equal to a preset value; Extracting trend features of the low-frequency data to generate trend data; Determining the value of the target weight based on the high-frequency data and the trend data, and calculating the target reconstruction rule based on the value of the target weight; Training a benchmark model based on the time series data matrix to obtain a production prediction model; Obtaining an objective function, establishing a production optimization model, and calculating a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles; Inputting the first optimal control parameter into the production prediction model, simulating the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generating a prediction result; wherein the prediction result includes the value of each item of the objective function; Based on the prediction result, determine whether each item in the objective function meets the preset requirements. If so, use the first optimal control parameter as the final control parameter. If not, determine the gap between the prediction result and the preset requirements, adjust the objective function based on the gap, and return to the step of calculating the first optimal control parameter based on the production optimization model.

2. The method according to claim 1, characterized in that The obtaining of the objective function, establishing a production optimization model, and calculating the first optimal control parameter based on the production optimization model include: Determining first importances of the target items at different angles based on product performance, and setting weight values ​​according to the first importances to establish the objective function, wherein the first importances of the target items at different angles are different; Judging whether each item in the objective function meets preset requirements based on the prediction result includes: The second importance of the target items at the multiple different angles is determined based on the production process, and the component items of the preset requirements are determined according to the second importance, and the most important target items in the first importance ranking and the second importance ranking are different.

3. The method according to claim 1, characterized in that The step of transforming the weight matrix based on the trend law corresponding to the data at the time point to obtain the target weight includes: Obtain the trend law of the data at the time point, and generate a transformation sample in combination with the trend law; wherein the trend law includes the frequency distribution of data of the same dimension at the same position; the transformation sample includes a target input parameter and a target output parameter; The weight matrix is ​​transformed based on the transformed samples, the target weight is determined based on the value of the weight vector in the transformed weight matrix, and the target reconstruction rule is determined based on the target weight; wherein the target reconstruction rule limits the time interval between two adjacent data vectors.

4. The method according to claim 1, characterized in that: The obtaining of the objective function, establishing a production optimization model, and calculating the first optimal control parameter based on the production optimization model include: Initializing a target particle swarm, and updating particle parameters of the target particle swarm based on the objective function constraint; wherein the target particle swarm represents a set of multiple groups of control parameters, and each particle in the target particle swarm represents a group of the control parameters; A global optimal solution is calculated based on the particle parameters updated a specified number of times, and the global optimal solution is determined as the first optimal control parameter.

5. The method according to claim 1, characterized in that The obtaining of the objective function, establishing a production optimization model, and calculating the first optimal control parameter based on the production optimization model include: Construct an optimization association relationship between the objective function and the first optimal control parameter, and establish the production optimization model based on the optimization association relationship; wherein the objective function includes power data, energy consumption data and product quality data; and the maximum weight of the optimization association relationship corresponds to the product quality data.

6. The method according to claim 3, characterized in that The training of the benchmark model based on the time series data matrix to obtain the production prediction model includes: The prediction correlation is obtained based on the trend law, and the benchmark model is trained based on the prediction correlation to obtain the production prediction model; wherein the prediction correlation represents the correlation between the control parameters at different time points; and the maximum weight of the prediction correlation corresponds to the energy consumption data in the objective function.

7. The method according to claim 1, characterized in that The production data of each position of the multi-stage reactor includes: drying preheater data, drying kettle data, drying kettle outlet data, separation preheater data, first separator data, second separator data, circulation storage tank data and booster pump data.

8. A control parameter optimization device for a multi-stage reactor, characterized in that: The device includes an acquisition module, a training module, a calculation module, a generation module and a processing module; wherein, The acquisition module is used to acquire aerogel production data to generate a time series data matrix; wherein the amount of data in the time series data matrix is ​​less than the aerogel production data, the values ​​of the corresponding positions of the data vectors in the aerogel production data vary periodically, each of the data vectors in the time series data matrix is ​​arranged in chronological order, and the data vectors include the production data of each position of the multi-stage reactor; The step of obtaining aerogel production data to generate a time series data matrix comprises: Acquire the aerogel production data based on multiple preset time step data, and collect data at multiple time points; Generate multiple reconstruction rules and corresponding weight matrices based on the preset time step data; wherein each of the reconstruction rules corresponds to a weight vector of one dimension in the weight matrix; Transform the weight matrix based on the trend law corresponding to the time point data to obtain a target weight, determine a target reconstruction rule based on the target weight, and generate the time series data matrix based on the target reconstruction rule; The step of transforming the weight matrix based on the trend law corresponding to the data at the time point to obtain the target weight includes: Transforming the aerogel production data to obtain frequency data; Removing high-frequency data from the frequency data to obtain low-frequency data, wherein the high-frequency data is data with a frequency higher than a preset value, and the low-frequency data is data with a frequency less than or equal to a preset value; Extracting trend features of the low-frequency data to generate trend data; Determining the value of the target weight based on the high-frequency data and the trend data, and calculating the target reconstruction rule based on the value of the target weight; The training module is used to train a benchmark model based on the time series data matrix to obtain a production prediction model; The calculation module is used to obtain an objective function, establish a production optimization model, and calculate a first optimal control parameter based on the production optimization model, wherein the objective function includes multiple objective items at different angles; The generating module is used to input the first optimal control parameter into the production prediction model, simulate the aerogel production process of the multi-stage reactor based on the production prediction model and the first optimal control parameter, and generate a prediction result; wherein the prediction result includes the value of each item of the objective function; The processing module is used to determine whether each item in the objective function meets the preset requirements based on the prediction results. If so, the first optimal control parameter is used as the final control parameter; if not, the gap between the prediction results and the preset requirements is determined, the objective function is adjusted based on the gap, and the step of calculating the first optimal control parameter based on the production optimization model is returned.

Images