LSTM-based rectification tower sensitive plate temperature model prediction control method and device, processor and storage medium thereof

Abstract

The invention relates to an LSTM-based model prediction control method for the temperature of a sensitive plate of a rectifying tower, and the method comprises the steps: collecting the historical data of a plurality of variables in the production process of the rectifying tower, and obtaining the original process data arranged according to a time axis; calculating the maximum information coefficient MIC of the plurality of variables under different delay time; taking the delay time corresponding to the maximum MIC value of the auxiliary variable as the optimal delay time between the auxiliary variable and the temperature of the sensitive plate; performing time delay reconstruction on the original data based on the optimal delay time to obtain a new data set; a data set obtained after time delay reconstruction is used as a model training sample, and an LSTM model is built and trained; and the trained LSTM model is used as a real-time prediction model of model prediction control MPC, so that a sensitive plate temperature controller is designed, a nonlinear constraint optimization problem is solved by establishing a target optimization function, and a model prediction optimal control quantity is obtained. The invention also relates to a corresponding device, a processor and a computer readable storage medium.

Description

Technical Field

[0001] The present invention relates to the field of distillation tower temperature control, and specifically refers to a model predictive control method, device, processor and computer-readable storage medium of the temperature of a distillation tower sensitive plate based on LSTM. Background Art

[0002] The distillation process is one of the common and important separation operations in chemical production, and is widely used in petrochemical, pharmaceutical, food and other industries. The operating conditions of the distillation tower directly affect the purity and yield of the product, among which temperature control is a key parameter in the distillation process. The traditional distillation tower temperature control method usually adopts PID control, but due to the high nonlinearity, strong coupling and large hysteresis of the distillation process, PID control often fails to achieve ideal control effects when facing complex working conditions.

[0003] Traditional MPC controllers mostly rely on complex physical models. In the actual production process, due to the dynamic properties of the fluid in the distillation tower, there are very complex nonlinear relationships between different variables, which leads to insufficient prediction accuracy of the prediction model, making the actual effect of the traditional MPC controller fail to meet the required accuracy and effect.

[0004] Based on this, it is necessary to propose a solution to overcome the shortcomings of the existing technology. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a model predictive control method, device, processor and computer-readable storage medium of the temperature of a sensitive plate of a distillation tower based on LSTM.

[0006] In order to achieve the above-mentioned object, the model predictive control method, device, processor and computer-readable storage medium of the temperature of the sensitive plate of the distillation tower based on LSTM of the present invention are as follows:

[0007] The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM is mainly characterized in that the method comprises the following steps:

[0008] (1) Collect historical data of multiple variables in the distillation tower production process from the factory to obtain original process data arranged along the time axis;

[0009] (2) Calculate the maximum information coefficient (MIC) of multiple variables at different delay times, use the MIC value as an evaluation index of the correlation between the auxiliary variable and the sensitive plate temperature of the distillation tower, and determine the auxiliary variables required for predicting the sensitive plate temperature through variable screening;

[0010] (3) The delay time corresponding to when the auxiliary variable takes the maximum MIC value is taken as the optimal delay time between the auxiliary variable and the sensitive plate temperature;

[0011] (4) reconstructing the original data with a time delay based on the optimal delay time to obtain a new data set;

[0012] (5) The data set obtained after time delay reconstruction is used as a model training sample to build and train the LSTM model;

[0013] (6) The trained LSTM model is used as the real-time prediction model of the model predictive control (MPC) to design the sensitive plate temperature controller;

[0014] (7) A target optimization function is established based on the predicted output of the LSTM model and the reference value in the prediction time domain, and the optimal control quantity predicted by the model is obtained by solving the nonlinear constrained optimization problem.

[0015] Preferably, the step (1) is specifically as follows:

[0016] The historical data of the distillation tower production process is collected from the factory, including multiple variable data of the distillation tower inlet flow, outlet flow, reboiler heat carrier flow, top reflux flow and distillation tower sensitive plate temperature, and sampling is performed according to the sampling period to obtain the original process data arranged along the time axis.

[0017] Preferably, the step (2) specifically comprises the following steps:

[0018] (2.1) Calculate the maximum information coefficient MIC between the characteristic variables and the target variables in the collected production process historical data at different delay times;

[0019] (2.2) The MIC between each characteristic variable and the target variable is taken as the correlation index, and the characteristic variables with MIC lower than the set threshold are eliminated. A preliminary variable screening is performed on multiple characteristic variables in the production data to obtain auxiliary variables for predicting the temperature of the sensitive plate of the distillation tower.

[0020] Preferably, the step (2.1) specifically includes the following steps:

[0021] Assume that there are two variables X and Y, each containing n measured values.

[0022] (2.1.1) Divide the variables X and Y into several intervals and form a two-dimensional grid to achieve data discretization processing;

[0023] (2.1.2) For each grid, count the number of data points in each cell to form a two-dimensional frequency table and calculate the mutual information;

[0024] (2.1.3) normalizing the mutual information to obtain a normalized mutual information value;

[0025] (2.1.4) Traverse all network partitioning methods and calculate the normalized mutual information value under each partitioning method, so as to calculate the maximum information coefficient MIC.

[0026] Preferably, the step (3) is specifically as follows:

[0027] Define the sample data set E = {X1, X2, ..., X n ,Y}, where X i is the dimension of the input variable after feature screening, Y is the target variable distillation tower sensitive plate temperature, and the delay time range is set to τ={0,1,2,...,τ max}, where τ max is the maximum delay time; for each feature variable X in the sample data set E i (i=1,2,...,n), calculate its mutual information coefficient I(X i (t-τ); Y(t)), the delay time corresponding to the maximum information coefficient MIC between the auxiliary variable and the target variable is selected as the optimal delay time.

[0028] Preferably, the step (4) is specifically as follows:

[0029] The number of delayed samples of each variable is calculated, and the optimal delay time of each auxiliary variable is divided by the sampling time to obtain the number of delayed samples; and the original time series data is converted into a data matrix, in which the number of rows is the number of rows of the original data minus the maximum number of delayed samples, and the number of columns is the number of auxiliary variables plus the number of target variables. The original data is shifted according to the calculated number of delayed samples and filled into the new matrix. The data of the target variable is filled into the new matrix starting from the row with the maximum number of delayed samples to obtain a new data set.

[0030] Preferably, the step (5) specifically comprises the following steps:

[0031] (5.1) Define the input layer, LSTM layer, activation layer, fully connected layer, and regression layer of the LSTM model;

[0032] (5.2) Select the optimizer, set the maximum number of iterations, set the initial learning rate, configure the learning rate reduction strategy, and perform model parameter configuration;

[0033] (5.3) The acquired data is randomly divided into training set and test set, and the data is divided according to the preset ratio.

[0034] (5.4) Use the configured training parameters and preprocessed training data to train the model and minimize the loss function of the model through iterative optimization;

[0035] (5.5) Calculate the evaluation indicators on the test set to evaluate the prediction accuracy and performance of the model and complete the model verification.

[0036] Preferably, the step (6) specifically comprises the following steps:

[0037] (6.1) Use the LSTM model to output the predicted value of the distillation tower sensitive plate temperature at the future moment;

[0038] (6.2) Design the objective function J(k) of the model predictive control MPC:

[0039]

[0040] Among them, y k represents the predicted temperature at the kth moment, y ref Indicates the set temperature, △u k represents the control input change at the kth moment, λ represents the weight coefficient of the control input change, N p Indicates the length of the prediction time domain, N c Indicates the length of the control time domain.

[0041] (6.3) Using the LSTM temperature prediction model, the temperature prediction value of the MPC controller in the prediction time domain is calculated and substituted into the objective function for optimization to obtain the optimal control input sequence.

[0042] Preferably, the step (7) is specifically as follows:

[0043] The particle swarm optimization algorithm is used to solve the nonlinear constrained optimization problem, and the optimal control sequence is obtained through real-time rolling optimization. The first control input u in the optimal control input sequence is t It is applied to the system, updates the system status, and enters the next control cycle to complete the predictive control process.

[0044] The model predictive control device for the temperature of the sensitive plate of the distillation tower based on LSTM has the following main features: the device comprises:

[0045] a processor configured to execute computer executable instructions;

[0046] A memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the above-mentioned LSTM-based model predictive control method for the temperature of the sensitive plate of the distillation tower are implemented.

[0047] The LSTM-based model predictive control processor for the temperature of a distillation tower sensitive plate has the main feature that the processor is configured to execute computer executable instructions, and when the computer executable instructions are executed by the processor, the steps of the above-mentioned LSTM-based model predictive control method for the temperature of a distillation tower sensitive plate are implemented.

[0048] The main feature of the computer-readable storage medium is that a computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the above-mentioned LSTM-based model predictive control method for the temperature of the sensitive plate of the distillation tower.

[0049] The LSTM-based model predictive control method, device, processor and computer-readable storage medium of the distillation tower sensitive plate temperature of the present invention are adopted. By building a distillation tower sensitive plate temperature model predictive controller, the problems of poor overall effect and poor anti-interference of the traditional PID controller are solved. Through the simulation analysis of the joint dynamic operation of Simulink and Aspen Dynamic, the results show that the designed control method has very good dynamic characteristics, thereby verifying the effectiveness and practicality of the control method. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention is a flow chart of the model predictive control method of the temperature of the sensitive plate of the distillation tower based on LSTM.

[0051] Figure 2 It is a distribution diagram of MIC values ​​of main process variables under different delay times in the LSTM-based model predictive control method of the sensitive plate temperature of the distillation tower of the present invention.

[0052] Figure 3 This is a test set prediction result diagram of the LSTM prediction model trained by the LSTM-based model prediction control method for the temperature of the sensitive plate of the distillation tower of the present invention.

[0053] Figure 4 It is a schematic diagram of the control strategy of the model predictive control method in the model predictive control method of the temperature of the sensitive plate of the distillation tower based on LSTM of the present invention.

[0054] Figure 5 This is a simulation result diagram of the model predictive control method of the temperature of the sensitive plate of the distillation tower based on LSTM of the present invention. DETAILED DESCRIPTION

[0055] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0056] Before describing in detail embodiments according to the present invention, it should be noted that, hereinafter, the terms "comprises", "includes" or any other variations are intended to cover non-exclusive inclusion, whereby a process, method, article or apparatus comprising a series of elements includes not only these elements, but also other elements not explicitly listed or inherent to such process, method, article or apparatus.

[0057] See also Figure 1 As shown, the model predictive control method of the temperature of the sensitive plate of the distillation tower based on LSTM, wherein the method comprises the following steps:

[0058] Step 1: Sample the data of the actual production process according to the sampling period to obtain the production history data of the distillation tower in the actual factory production.

[0059] Step 2: Calculate the maximum information coefficient (MIC) of multiple variables and the sensitive plate temperature at different delay times, use the MIC value of the data as a measure of the correlation between the characteristic variable and the sensitive plate temperature of the distillation tower, perform variable screening, eliminate the characteristic variables with a maximum information coefficient lower than the set threshold, and determine the auxiliary variables used for sensitive plate temperature prediction. Take the delay time corresponding to the auxiliary variable when the maximum MIC is obtained as the optimal delay time between the auxiliary variable and the sensitive plate temperature, and then reconstruct the original data based on the optimal delay time;

[0060] Step 3: Use the delayed reconstructed data set as training samples to build and train the LSTM model.

[0061] Step 4: Use the LSTM model trained in step 3 as the prediction model in the model predictive controller, consider the mechanical limitations of the distillation tower and the upper and lower limits of the variables in the process, establish the target optimization function based on the prediction output of the LSTM model and the reference value in the prediction time domain, use the selected auxiliary variables as the input of the LSTM prediction model, predict the future state of the sensitive plate temperature of the distillation tower, use the particle swarm optimization algorithm (PSO) to solve the nonlinear constrained optimization problem, and then obtain the optimal control amount through real-time rolling optimization.

[0062] In a preferred embodiment of the present invention, the step 1 specifically comprises the following steps:

[0063] (1.1) The production data of the distillation process in the factory is obtained through sensors and data acquisition systems, and the original data arranged on the time axis is obtained, including multiple variables such as the distillation tower inlet flow, outlet flow, reboiler heat carrier flow, reflux flow and sensitive plate temperature.

[0064] In a preferred embodiment of the present invention, step 2 specifically comprises the following steps:

[0065] (2.1) Calculate the maximum information coefficient (MIC) between the characteristic variable and the target variable in the production data collected in step 1 at different delay times. The maximum information coefficient (MIC) is calculated in the following way:

[0066] (2.1.1) Data discretization:

[0067] Suppose there are two continuous variables X and Y, each with n observations. Divide the value range of variables X and Y into several intervals of equal width or frequency. Generally speaking, the number of intervals B should be less than n. B is n 0.6 For each variable, different partitioning methods are selected to form different two-dimensional grids.

[0068] (2.1.2) Calculate mutual information:

[0069] For each partitioning method, the data points are assigned to different cells of the two-dimensional grid. Then the number of data points in each cell is calculated to form a two-dimensional frequency table. The mutual information I(X;Y) is calculated using the frequency table:

[0070]

[0071] Here, p(x,y) is the joint probability distribution, and p(x) and p(y) are the marginal probability distributions.

[0072] (2.1.3) Normalized mutual information:

[0073] In order to make the mutual information values ​​under different division methods comparable, the mutual information is normalized. The normalization method is usually to divide the mutual information by the logarithm of the number of grid divisions, that is:

[0074]

[0075] Among them, |X| and |Y| are the value intervals of X and Y respectively.

[0076] (2.1.4) Calculate the maximum information coefficient (MIC):

[0077] Traverse all possible grid division methods and calculate the normalized mutual information value under each division method.

[0078] The maximum value of the normalized mutual information value is selected as the maximum information coefficient (MIC):

[0079]

[0080] (2.2) A variable is randomly generated, and the value of the variable is random. The MIC between the random variable and the target variable is calculated. The MIC of the random variable is used as the threshold, and the characteristic variables whose maximum information coefficient is lower than the threshold are eliminated. A preliminary variable screening is performed on multiple characteristic variables in the production data to obtain auxiliary variables for predicting temperature.

[0081] (2.3) Based on the maximum information coefficient (MIC) between the auxiliary variable and the target variable temperature, determine the auxiliary variable x i The time delay parameter D between the target variable y(k) i , where i=1,2,…,m represents the number of auxiliary variables.

[0082] (2.4) Based on the optimal delay time between the auxiliary variable and the target variable, the collected process production data is reconstructed with a time delay to obtain a data set after screening and time delay reconstruction in step (2.2). The specific steps for time delay reconstruction are as follows:

[0083] For each time step k (from max(D i ) to T, where T is the total number of time steps in the time series data), construct a row containing the auxiliary variable values ​​after all delays. For example, suppose there are m auxiliary variables x1, x2, …, x m and the corresponding delay parameters D1, D2, ... D m and the total number of time steps T, the number of rows in the delay matrix will be T-max(D i )+1, the number of columns is the number of auxiliary variables m. The delay matrix is ​​in the following form:

[0084]

[0085] Among them, each row of the matrix corresponds to a time step, and each column corresponds to a delayed auxiliary variable.

[0086] In a preferred embodiment of the present invention, the LSTM neural network model established in step 3 is as follows:

[0087] (3.1) Model structure definition: define input layer, LSTM layer, activation layer, fully connected layer, regression layer, etc.

[0088] (3.2) Configure model parameters: select optimizer, set maximum number of iterations, set initial learning rate, configure learning rate reduction strategy, etc.

[0089] (3.3) Data partitioning: The data is randomly divided into training set and test set, with 80% training set and 20% test set.

[0090] (3.4) Model training: Use the configured training parameters and preprocessed training data for training. The model is iteratively optimized to minimize the loss function.

[0091] (3.5) Model validation: Calculate the evaluation indicators on the test set, such as mean absolute error (MAE) and mean review error (MBE), to evaluate the prediction accuracy and performance of the model.

[0092] In a preferred embodiment of the present invention, the steps of designing the MPC controller in step 4 are as follows:

[0093] Set the upper and lower limits of the distillation tower temperature, and use the LSTM temperature prediction model trained in step 3 to predict the temperature data after the future time step, and input it into the cost function. In each sliding time window, rolling optimization is performed in the limited time domain of [t, t+k]. The process of solving the optimal control system problem can be converted into the optimization process of the cost function. The PSO algorithm is used to solve the optimization problem and find the best control sequence. These control variables are regarded as particles in the multidimensional search space. The moving speed and direction of the particles are updated iteratively to obtain the optimal solution. The first control input u in the optimal control sequence is t Applied to the system, the system status is updated, and the next control cycle is entered.

[0094] In a specific embodiment of the present invention, the specific implementation process of the method is as follows:

[0095] 1. Data acquisition and processing

[0096] In the embodiment, the original data is the production process data of a certain factory's pressure swing distillation. By analyzing the reaction mechanism of the distillation process, 8 variables affecting the temperature of the sensitive plate of the distillation tower are initially selected as characteristic variables, and the sensitive plate temperature is used as the target variable. The main process variables are shown in Table 1. With a sampling period of 5s, 5000 operation data in the operation process are obtained from the factory.

[0097] Table 1 Main process variables of distillation tower

[0098]

[0099] Calculate the maximum information coefficient (MIC) between the characteristic variables and the target variables in the collected production data at different delay times. The specific steps are as follows:

[0100] (1) The collected production data {X 5000×10 ,Y 5000×1 Each auxiliary variable X i The range of values ​​of and Y is divided into several intervals of equal width or frequency, and the number of intervals B is n 0.6 , n is the total amount of collected data, 5000.

[0101] (2) Calculate mutual information

[0102] For each partitioning method, the data points are assigned to different cells of the two-dimensional grid. Then the number of data points in each cell is calculated to form a two-dimensional frequency table. The mutual information I(X; Y) is calculated by the following formula (1):

[0103]

[0104] Here, p(x,y) is the joint probability distribution, and p(x) and p(y) are the marginal probability distributions.

[0105] (3) Normalized mutual information:

[0106] In order to make the mutual information values ​​under different partitioning methods comparable, the mutual information is normalized using the following formula (2).

[0107]

[0108] Among them, |X| and |Y| are the value intervals of X and Y respectively.

[0109] (4) Calculate the maximum information coefficient (MIC):

[0110] Traverse all possible grid division methods and calculate the normalized mutual information value under each division method.

[0111] The maximum value of the normalized mutual information value is selected as the maximum information coefficient (MIC) by the following formula (3):

[0112]

[0113] (5) Auxiliary variable screening:

[0114] A variable is randomly generated, the value of which is random, and the MIC between the random variable and the target variable is calculated. The calculation result is as follows Figure 2 The result is 0.334. The MIC threshold is set to 0.334, and the characteristic variables below the threshold are eliminated. The characteristic variables in the production data are screened to obtain auxiliary variables for predicting temperature. The MIC of each auxiliary variable after screening and the optimal delay time between each auxiliary variable and the target variable are shown in Table 2.

[0115] Table 2

[0116]

[0117] (6) Delay reconstruction:

[0118] The delay time corresponding to the maximum MIC of the auxiliary variable is taken as the optimal delay time between the auxiliary variable and the sensitive board temperature. Calculate the number of delay samples D for each variable i , the optimal delay time of each auxiliary variable is divided by the sampling time to get the number of delay samples. For converting the original time series data into a data matrix, the number of rows is the number of rows of the original data minus the maximum number of delay samples max(D i ), the number of columns is the number of auxiliary variables, the original data is shifted according to the calculated number of delay samples, and filled into the new matrix, and the data of the target variable is filled into the new matrix starting from the row with the largest number of delay samples. The form of the delay matrix is ​​as follows:

[0119]

[0120] 2. Build and train LSTM prediction model

[0121] (1) Build and train the LSTM neural network model

[0122] The LSTM neural network model includes an input layer, an LSTM layer, a ReLU activation layer, a fully connected layer, and an output layer. The Adam optimizer is selected. After multiple experiments to adjust the parameters, the learning rate of the Adam algorithm is set to 0.01. 80% of the preprocessed data is used as the training set of the model, and 20% is used as the test set of the model. After multiple experiments, it is found that the model has the best effect when it is trained 1000 times, so the maximum number of model training times is set to 1000 times. The prediction effect of the trained model test set is shown in the figure below: Figure 3 .

[0123] 3. Design MPC temperature controller

[0124] The task of the distillation tower sensitive plate temperature control model is to determine N c input sequence, so that the predicted value of the next sampling period [y p (k+1),y p (k+2),…,y p (k+N p )] is constantly approaching the expected value of the distillation tower temperature. Solving the optimal control sequence problem can be transformed into the following equation (8): The objective function is an optimization process with equation (9) as the constraint condition. Here, the PSO algorithm is used to obtain the optimal control sequence u = [u(k), u(k+1),…, u(k+N c -1)], take the first one in the optimal control sequence as the control variable and input it into the system. In this example, the control time domain N p is 5, prediction time domain N c is 15.

[0125]

[0126] 4. Simulation experiment

[0127] The final simulation experiment is based on the production process parameters and process requirements in the actual factory. AspenDynamic is used to build a dynamic simulation of the actual factory to simulate the production process in the actual factory. The simulation verification has the following steps:

[0128] (1) Aspen Dynamic was used to obtain historical data of multiple characteristic variables in the distillation tower production process by adding disturbances and changing set values ​​during operation. The sampling time was 1 s, and 5000 time series data were collected.

[0129] (2) In MATLAB, the obtained time series data is used according to the above steps to calculate MIC, screen auxiliary variables, reconstruct data delays, and build and train the LSTM prediction model.

[0130] (3) Based on the trained LSTM prediction model, design the LSTM model prediction controller and build the control loop in Simulink. The control strategy is as follows: Figure 4 Aspen Dynamic simulation is used as the actual production site for joint simulation and compared with conventional PID cascade control. During the simulation, the PID controller is adjusted to the optimal parameters to obtain the best control effect of the comparison method. The final control effect comparison is as follows: Figure 5 shown.

[0131] The model predictive control device for the temperature of a sensitive plate of a distillation tower based on LSTM, wherein the device comprises:

[0132] a processor configured to execute computer executable instructions;

[0133] A memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the above-mentioned LSTM-based model predictive control method for the temperature of the sensitive plate of the distillation tower are implemented.

[0134] The LSTM-based model predictive control processor for the temperature of a distillation tower sensitive plate, wherein the processor is configured to execute computer executable instructions, and when the computer executable instructions are executed by the processor, the steps of the above-mentioned LSTM-based model predictive control method for the temperature of a distillation tower sensitive plate are implemented.

[0135] The computer-readable storage medium stores a computer program thereon, and the computer program can be executed by a processor to implement the steps of the above-mentioned LSTM-based model predictive control method for the temperature of the sensitive plate of the distillation tower.

[0136] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0137] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device.

[0138] A person of ordinary skill in the art may understand that all or part of the steps of the method for implementing the above-mentioned embodiment may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one of the steps of the method embodiment or a combination thereof.

[0139] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0140] In the description of this specification, the description with reference to the terms "an embodiment", "some embodiments", "example", "specific example", or "embodiment" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0141] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

[0142] The LSTM-based model predictive control method, device, processor and computer-readable storage medium of the distillation tower sensitive plate temperature of the present invention are adopted. By building a distillation tower sensitive plate temperature model predictive controller, the problems of poor overall effect and poor anti-interference of the traditional PID controller are solved. Through the simulation analysis of the joint dynamic operation of Simulink and Aspen Dynamic, the results show that the designed control method has very good dynamic characteristics, thereby verifying the effectiveness and practicality of the control method.

[0143] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it is apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Therefore, the specification and drawings should be regarded as illustrative rather than restrictive.

Claims

1. A model predictive control method for the temperature of a distillation tower sensitive plate based on LSTM, characterized in that: The method comprises the following steps: (1) Collecting historical data of multiple variables in the production process of the distillation tower to obtain original process data arranged along the time axis; (2) Calculate the maximum information coefficient (MIC) of multiple variables at different delay times, use the MIC value as an evaluation index of the correlation between the auxiliary variable and the sensitive plate temperature of the distillation tower, and determine the auxiliary variables required for predicting the sensitive plate temperature through variable screening; (3) The delay time corresponding to when the auxiliary variable takes the maximum MIC value is taken as the optimal delay time between the auxiliary variable and the sensitive plate temperature; (4) reconstructing the original data with a time delay based on the optimal delay time to obtain a new data set; (5) The data set obtained after time delay reconstruction is used as a model training sample to build and train the LSTM model; (6) The trained LSTM model is used as the real-time prediction model of the model predictive control (MPC) to design the sensitive plate temperature controller; (7) A target optimization function is established based on the predicted output of the LSTM model and the reference value in the prediction time domain, and the optimal control quantity predicted by the model is obtained by solving the nonlinear constrained optimization problem.

2. The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM according to claim 1, characterized in that: The step (1) is specifically as follows: The historical data of the distillation tower production process is collected from the factory, including multiple variable data of the distillation tower inlet flow, outlet flow, reboiler heat carrier flow, top reflux flow and distillation tower sensitive plate temperature, and sampling is performed according to the sampling period to obtain the original process data arranged along the time axis.

3. The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM according to claim 2, characterized in that, The step (2) specifically comprises the following steps: (2.1) Calculate the maximum information coefficient MIC between the characteristic variables and the target variables in the collected production process historical data at different delay times; (2.2) The MIC between each characteristic variable and the target variable is taken as the correlation index, and the characteristic variables with MIC lower than the set threshold are eliminated. A preliminary variable screening is performed on multiple characteristic variables in the production data to obtain auxiliary variables for predicting the temperature of the sensitive plate of the distillation tower.

4. The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM according to claim 3, characterized in that: The step (2.1) specifically comprises the following steps: Assume that there are two variables X and Y, each containing n measured values. (2.1.1) Divide the variables X and Y into several intervals and form a two-dimensional grid to achieve data discretization processing; (2.1.2) For each grid, count the number of data points in each cell to form a two-dimensional frequency table and calculate the mutual information; (2.1.3) normalizing the mutual information to obtain a normalized mutual information value; (2.1.4) Traverse all network partitioning methods and calculate the normalized mutual information value under each partitioning method, so as to calculate the maximum information coefficient MIC.

5. The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM according to claim 4, characterized in that: The step (3) is specifically as follows: Define the sample data set E = {X1, X2, ..., X n ,Y}, where n is the dimension of the input variable after feature screening, Y is the target variable distillation tower sensitive plate temperature, and the delay time range is set to τ={0,1,2,...,τ max }, where τ max is the maximum delay time; for each feature variable X in the data set E i (i=1,2,...,n), calculate its mutual information coefficient I(X i (t-τ); Y(t)), the delay time corresponding to the maximum information coefficient MIC between the auxiliary variable and the target variable is selected as the optimal delay time.

6. The model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM according to claim 5, characterized in that: The step (4) is specifically as follows: The number of delayed samples of each variable is calculated, and the optimal delay time of each auxiliary variable is divided by the sampling time to obtain the number of delayed samples; and the original time series data is converted into a data matrix, in which the number of rows is the number of rows of the original data minus the maximum number of delayed samples, and the number of columns is the number of auxiliary variables plus the number of target variables. The original data is shifted according to the calculated number of delayed samples and filled into the new matrix. The data of the target variable is filled into the new matrix starting from the row with the maximum number of delayed samples to obtain a new data set.

7. The model predictive control method for the temperature of the sensitive plate of a distillation tower based on LSTM according to claim 6, characterized in that: The step (5) specifically comprises the following steps: (5.1) Define the input layer, LSTM layer, activation layer, fully connected layer, and regression layer of the LSTM model; (5.2) Select the optimizer, set the maximum number of iterations, set the initial learning rate, configure the learning rate reduction strategy, and perform model parameter configuration; (5.3) Randomly divide the acquired data into training set and test set, and divide the data according to the preset ratio; (5.4) Use the configured training parameters and preprocessed training data to train the model and minimize the loss function of the model through iterative optimization; (5.5) Calculate the evaluation indicators on the test set to evaluate the prediction accuracy and performance of the model and complete the model verification.

8. The model predictive control method for the temperature of the sensitive plate of a distillation tower based on LSTM according to claim 7, characterized in that: The step (6) specifically comprises the following steps: (6.1) Use the LSTM model to output the predicted value of the distillation tower sensitive plate temperature at the future moment; (6.2) Design the objective function J(k) of the model predictive control MPC: Among them, y k represents the predicted temperature at the kth moment, y ref Indicates the set temperature, △u k represents the control input change at the kth moment, λ represents the weight coefficient of the control input change, N p Indicates the length of the prediction time domain, N c Indicates the length of the control time domain; (6.3) Using the LSTM temperature prediction model, the temperature prediction value of the MPC controller in the prediction time domain is calculated and substituted into the objective function for optimization to obtain the optimal control input sequence.

9. The model predictive control method for the temperature of the sensitive plate of a distillation tower based on LSTM according to claim 8, characterized in that: The step (7) is specifically as follows: The particle swarm optimization algorithm is used to solve the nonlinear constrained optimization problem, and the optimal control sequence is obtained through real-time rolling optimization. The first control input u in the optimal control input sequence is t It is applied to the system, updates the system status, and enters the next control cycle to complete the predictive control process.

10. A model predictive control device for the temperature of a distillation tower sensitive plate based on LSTM, characterized in that: The device comprises: a processor configured to execute computer executable instructions; A memory storing one or more computer executable instructions, wherein when the computer executable instructions are executed by the processor, the steps of the model predictive control method for the temperature of a distillation tower sensitive plate based on LSTM described in any one of claims 1 to 9 are implemented.

11. A model predictive control processor for the temperature of a distillation tower sensitive plate based on LSTM, characterized in that: The processor is configured to execute computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the model predictive control method for the temperature of the sensitive plate of the distillation tower based on LSTM described in any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the LSTM-based model predictive control method for the temperature of a distillation tower sensitive plate as described in any one of claims 1 to 9.

Images