Soft measurement modeling and real-time optimization control method for n-butane removal tower of alkylation device

Abstract

The invention discloses a soft measurement modeling and real-time optimization control method for an n-butane removal tower of an alkylation device, and the method comprises the steps: firstly, determining a control variable and a controlled variable of a system based on historical operation data and mechanism analysis, and building a dynamic model of the system; and establishing an LSTM neural network by using off-line sample data of the n-butane removal tower to predict the purity of n-butane at the tower top. And then, on the basis of the dynamic model of the system, an MPC controller is used as an upper controller of a PID controller, and the dynamic adjustment capability and the control precision of the system are improved by predicting and optimizing an operation sequence. And finally, in combination with a real-time optimization (RTO) technology, the maximum n-butane concentration and the minimum tower energy consumption are set as objective functions, and the operation cost and the energy efficiency are optimized. The control precision of the n-butane removal tower can be effectively improved, the product quality is optimized, the energy consumption is reduced, the adaptability of the system to the working condition change is enhanced, and the method is suitable for industrial application.

Description

Technical Field

[0001] The present invention relates to the field of petrochemical industry, in particular to a modeling and advanced control technology of a de-butanizer in an alkylation oil unit, and specifically to a control method based on model predictive control (MPC) and real-time optimization (RTO). Background Art

[0002] In the alkylation unit, the de-n-butanizer is the last separation tower, and its top product is n-butane, and its bottom product is alkylate oil. The feed of this tower comes from the bottom output of the de-isobutanizer, and its main components include n-butane and C8 components. The existing control strategy mainly relies on traditional PID control, which is difficult to deal with problems such as multi-variable coupling and operating condition fluctuations, resulting in difficulty in optimizing product quality and energy consumption.

[0003] Specifically, the problems existing in the de-n-butanizer are: (1) Unstable tower operation: Due to the unstable feed composition of the device and the changes in the environment during operation, the operation of the de-butanizer tower often fluctuates, resulting in unstable product quality (the concentration of n-butane is often lower than 80%). (2) Single control strategy: Traditional PID control has poor adaptability to complex nonlinear and time-varying systems. It takes a long time to adjust under disturbance conditions, which can easily lead to reduced product quality and increased energy consumption. (3) Insufficient energy consumption optimization: Existing control methods mostly focus on single variable regulation and lack global optimization of overall energy consumption and operating costs, making it difficult to meet the needs of industrial energy conservation and consumption reduction. Summary of the invention

[0004] Aiming at the problems of unstable operation, single control strategy and insufficient energy efficiency optimization of a debutanizer, the present invention proposes a soft measurement modeling and real-time optimization control method for a debutanizer. By establishing a soft measurement model and an MPC control strategy and combining RTO to calculate the optimal control parameters, the purity control of the n-butane at the tower top and the optimization of the tower energy consumption are achieved.

[0005] The present invention achieves the above object through the following technical solutions: A soft sensor modeling and real-time optimization control method for a de-n-butanizer of an alkylation unit, comprising: (1) Soft measurement modeling: collect historical operation data and offline analysis data of the dealanizer in the alkylation unit, filter and normalize the collected data, generate a data set for modeling, and divide it into a training set and a validation set; train the neural network model to obtain a soft measurement model; The trained soft-sensing model is put into use during operation, and the actual operating parameters of the alkylation unit are collected to predict the current n-butane concentration in real time, thus realizing online monitoring of n-butane concentration. (2) Controller design and optimization: A model predictive control (MPC) controller is used to provide set values ​​for the PID controller. The MPC controller uses the dynamic response model of the debutanizer unit as its internal model. Before use, a genetic algorithm is used to optimize the parameters of the MPC controller to improve the dynamic response performance and regulation accuracy of the controller. During operation, the optimized MPC controller receives the reference value of the controlled variable given by the engineer and the RTO, calculates the optimal control solution through the internal model, and transmits the result as a reference value to the PID controller to achieve precise regulation of the device. (3) Real-time optimization control, including the following three steps: (a) Real-time data collection: real-time collection of operating data of the alkylation unit, including control variables, controlled variables and n-butane concentration, and filling the data into the database. When the database reaches the upper limit, the latest data will replace the oldest data; (b) Parameter update: With a period of 10 min, perform multivariate quadratic fitting on the data in the cyclic array, establish the relationship function between the control variable, the controlled variable and the n-butane concentration, and use the newly obtained function parameters to replace the function parameters of the previous period; (c) Optimization solution: Use quadratic programming to solve the relationship function to obtain the optimal control point in this cycle, and pass this point as a reference value to the MPC controller; Through the cycle of three steps, a trend of continuous optimization of n-butane concentration and device energy consumption is formed, thereby achieving real-time optimization control of the de-n-butaneizer.

[0006] Preferably, in the soft sensor modeling, the data used are the offline sampling data of the n-butane concentration in the dealanizer, and the historical operation data of the corresponding time control variables and the controlled variables.

[0007] Preferably, the control variables required for the de-n-butanizer tower are: feed flow rate, reflux flow rate, reboiler steam flow rate, condensed water flow rate, and the controlled variables are tower top pressure, tower top temperature, tower sensitive plate temperature, tower bottom temperature, tower bottom alkylate flow rate, tower top external n-butane flow rate, and tower bottom pressure.

[0008] Preferably, the neural network model is a long short-term memory (LSTM) neural network, which includes an input layer, an LSTM layer, a fully connected layer, an activation function layer, a forgetting layer, and an output layer.

[0009] Preferably, based on the existing PID control loop of the debutanizer, an MPC controller, an RTO module and soft measurement are added.

[0010] Preferably, the internal system dynamic model of the MPC controller is a transfer function model of the system input and output, and the transfer function model is as follows:

[0011] in, is the dynamic model of the system, used to represent the control variables and the controlled variable the relationship between; is the gain of the system, which represents the overall amplification or attenuation effect of the system; is the time constant of the system, which is related to the inertia or energy storage of the system; It is the pure lag time of the system, which indicates the delay time that the change of system input affects the output.

[0012] Preferably, the MPC controller uses an internal model to predict future system responses based on real-time measured process parameters, and forms a closed-loop control with the PID control loop, thereby achieving rapid response and precise regulation of the debutanizer process.

[0013] Preferably, the objective function of the genetic algorithm is as follows: ; ;

[0014] Where: F is the objective function, OP1 is the weighted root mean square error between the controlled variable and the reference value, OP2 is the weighted root mean square error of the change value of the controlled variable, is the calculation weight of the control variable and the controlled variable, is the observed value of the controlled variable U and the controlled variable Y at time i, is the reference value of the controller’s controlled variable; Through crossover and mutation operations, the optimal operating points of a set of controllers are solved, the Pareto front is drawn, the optimal operating point is selected according to actual needs, and its corresponding parameter values ​​are assigned to the MPC controller, thereby improving the overall dynamic performance and stability of the system.

[0015] Preferably, in the real-time optimization control, the real-time data collection step is to continuously collect the current working conditions, the parameter updating step uses the real-time data to compare and correct the errors with the preset process model, and the real-time optimization is to optimize and adjust the control strategy in real time based on the online operation status to achieve the goals of reducing operating costs and optimizing energy consumption. The optimization goals are as follows:

[0016] In the formula, is the optimization objective function of RTO, is the relationship function between n-butane concentration c and control variable u, It is the relationship function between the energy consumption E of the debutanizer and the control variable u.

[0017] (1) The present invention can make full use of the historical operation data of the device and can process data sets with a long time and large data volume; (2) The LSTM neural network model used in the present invention has strong prediction capabilities on both linear and nonlinear data sets, and its own model structure and hyperparameters can be set by itself, with strong versatility; (3) The multi-objective genetic algorithm used in the present invention optimizes the MPC controller parameters, which can achieve setting goals according to needs and can intuitively select the optimal solution from the Pareto front as the controller parameters; (4) The real-time optimization of the present invention can update its own internal model by collecting the operation data of the device in real time, so that its adaptability to various working conditions is greatly improved; (5) The real-time optimization of the present invention uses multivariate quadratic functions to fit the internal model, which can greatly reduce the amount of computer calculations while ensuring the accuracy of the model, thereby lowering the threshold for practical application; (6) The double-layer optimization control designed by the present invention can minimize the energy consumption of the device while ensuring the control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the soft measurement and real-time optimization control process of the de-n-butanizer tower of the present invention.

[0019] Figure 2 This is a flow chart of the LSTM neural network training algorithm of the present invention.

[0020] Figure 3 This is a flow chart of the MPC parameter optimization algorithm of the present invention.

[0021] Figure 4 The invention provides a real-time optimization control of the n-butane concentration response curve under step disturbance.

[0022] Figure 5 The present invention optimizes the energy consumption response curve under step disturbance in real time. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto: Reference Figure 1 , the soft sensor modeling and real-time optimization control method of the de-butanizer tower includes the following steps: (1) Analyze the process flow of the de-n-butane tower unit to determine the control variables and controlled variables required for modeling. The control variables are: feed flow, reflux flow, reboiler steam flow, condensate flow, and the controlled variables are: tower top pressure, tower top temperature, tower sensitive plate temperature, tower bottom temperature, tower bottom alkylation oil flow, tower top outbound n-butane flow, and tower bottom pressure. After determining the variables, collect data from the unit's historical operation database when each control variable steps and the other control variables remain unchanged, collect n-butane concentration data from the unit's offline sampling database, and collect data on control variables and controlled variables based on time. Preprocess all data and save them for later use.

[0024] (2) Use the step data of the control variable and the controlled variable to establish a transfer function. Calculate the open-loop gain between the control variable and the controlled variable through the transfer function, and then calculate the relative gain, so as to determine the control variable corresponding to each controlled variable.

[0025] (3) Reference Figure 2 , build an LSTM neural network model, set hyperparameters, use the n-butane concentration collected in step (1) as the output variable, and the remaining variables as input variables. Divide the data into a training set and a test set in a ratio of 8:2, and use the training set to train the LSTM neural network. After training, use the test set to test the model prediction effect, and use the correlation coefficient (r), root mean square error (RMSE) and determination coefficient (R 2 ) is used as the effect indicator, and the calculation formula is as follows: ; ;

[0026] Where: n is the number of samples; is the observed value of input variable X and output variable Y; is the mean of the input variable X and the output variable Y; is the predicted value; Train the model multiple times and select the one with the best indicator as the soft measurement model.

[0027] (4) Design an MPC controller based on the system dynamic model established in step (2), referring to Figure 5 , write its parameter optimization script, and optimize the parameters as the control variable weight, control variable change weight, controlled variable weight, prediction time domain and control time domain. The optimization target selects the root mean square error of the difference between the controlled variable and the set value and the root mean square error of the change of the control variable. The calculation formula is as follows: ; ;

[0028] Where: F is the objective function of the genetic algorithm, OP1 is the weighted root mean square error between the controlled variable and the reference value, OP2 is the weighted root mean square error of the change value of the controlled variable, is the calculation weight of the control variable and the controlled variable, is the observed value of the controlled variable U and the controlled variable Y at time i, is the reference value of the controller's controlled variable.

[0029] Reference Figure 3 Genetic algorithm is used to optimize the MPC controller parameters, the population size is set to 70, the maximum evolutionary number is set to 60, and the upper and lower limits of each parameter are set. After the settings are completed, the genetic algorithm is run, and the Pareto front is drawn after normalization based on the calculation results of the genetic algorithm. The point with the smallest sum of the squares of the two target values ​​in the frontier solution is selected as the optimal solution, and the parameters of this point are used as the final parameters of the controller.

[0030] (5) Design a real-time optimization layer based on the MPC control loop established in step (4). The real-time optimization layer is divided into three steps according to its function: data acquisition, parameter update, and optimization solution.

[0031] The data collection step is responsible for collecting and saving the real-time operation data of the device, and outputting the saved data and updated parameters to the parameter update part. The sampling interval is set to 6s, and the storage data set size is 300 groups. Every 6s, the system writes the latest operation data in the storage. When the stored data reaches 300 groups, the earliest group in the stored data is replaced with the latest collected data to ensure continuous data update; The parameter update step is responsible for fitting the relationship between the control variable, the controlled variable, and the n-butane concentration in the stored data using a multivariate quadratic equation at the beginning of each control cycle based on the input update parameters, and then outputting the equation parameters to the optimization solution part. Set the control cycle to 600s. When the input update parameters meet the requirements, run the equation fitting command to update and output the equation parameters. When the update parameters do not meet the requirements, output the equation parameters of the last solution. The optimization solution step is responsible for solving the current optimal solution based on the input equation parameters and the set objective function and outputting the control variables at the optimal solution to the MPC controller. The set objective function is:

[0032] In the formula, is the optimization objective function of RTO, is the relationship function between n-butane concentration c and control variable u, It is the relationship function between the energy consumption E of the debutanizer and the control variable u.

[0033] Add constraints as: ;

[0034] In the formula, is the control variable, is the controlled variable.

[0035] The quadratic programming method is used to solve the formula, and the optimal value of the control variable is obtained and output to the MPC controller as a reference value.

[0036] In order to compare the control effects of the soft sensor modeling and real-time optimization control method, PID control method and MPC control method of the de-n-butanizer in the present invention, this embodiment conducted a set of simulation experiments: the initial n-butane concentration at the top of the device was 80%, and a 10% step disturbance was added to the control variable at t=500s.

[0037] like Figure 4 and Figure 5 As shown, in the case of adding disturbance, although there are certain fluctuations in the later stage of the real-time optimization control of the present invention, it can be seen that it is within a certain range. The oscillation of the curve around a certain value is regarded as reaching a stable state. By observing the trends of concentration and energy consumption, it can be seen that the concentration fluctuates around 90% after 6000s, and the energy consumption fluctuates around 3.01 after 7000s. Compared with single MPC control, the concentration of n-butane can return to more than 80% faster and the energy consumption after stabilization is reduced from 3.06tce / h (tons of standard coal / hour) at the beginning to 3.01tce / h, a reduction of about 1.6%.

[0038] The soft measurement modeling and real-time optimization control method of the dealbutanizer of the present invention is to establish a soft measurement model of the top butane concentration of the dealbutanizer through an LSTM neural network, and to establish MPC and RTO control loops in sequence. On the basis of the existing PID control, the method improves the control effect of the device and reduces energy consumption while ensuring product quality.

[0039] Although the present invention has been illustrated and described with respect to the preferred embodiments, it will be appreciated by those skilled in the art that various changes and modifications may be made to the present invention without departing from the scope of the present invention as defined by the claims.

Claims

1. A soft sensor modeling and real-time optimization control method for a de-butanizer of an alkylation unit, characterized in that: include: (1) Soft sensor modeling: collect historical operating data and offline analysis data of the dealanizer in the alkylation unit, filter and normalize the collected data, generate a data set for modeling, and divide it into a training set and a validation set; Train the neural network model to obtain the soft measurement model; The trained soft-sensing model is put into use during operation, and the actual operating parameters of the alkylation unit are collected to predict the current n-butane concentration in real time, thus realizing online monitoring of n-butane concentration. (2) Controller design and optimization: A model predictive control (MPC) controller is used to provide set values ​​for the PID controller. The MPC controller uses the dynamic response model of the debutanizer unit as its internal model. Before use, a genetic algorithm is used to optimize the parameters of the MPC controller to improve the dynamic response performance and regulation accuracy of the controller. During operation, the optimized MPC controller receives the reference value of the controlled variable given by the engineer and the RTO, calculates the optimal control solution through the internal model, and transmits the result as a reference value to the PID controller to achieve precise regulation of the device. (3) Real-time optimization control, including the following three steps: (a) Real-time data collection: real-time collection of operating data of the alkylation unit, including control variables, controlled variables and n-butane concentration, and filling the data into the database. When the database reaches the upper limit, the latest data will replace the oldest data; (b) Parameter update: With a period of 10 min, perform multivariate quadratic fitting on the data in the cyclic array, establish the relationship function between the control variable, the controlled variable and the n-butane concentration, and use the newly obtained function parameters to replace the function parameters of the previous period; (c) Optimization solution: Use quadratic programming to solve the relationship function to obtain the optimal control point in this cycle, and pass this point as a reference value to the MPC controller; Through the cycle of three steps, a trend of continuous optimization of n-butane concentration and device energy consumption is formed, thereby achieving real-time optimization control of the de-n-butaneizer.

2. The soft sensor modeling and real-time optimization control method of a de-butanizer of an alkylation unit according to claim 1, characterized in that: In the soft sensor modeling, the data used are the offline sampling data of the n-butane concentration in the dealanizer and the historical operation data of the corresponding time control variables and controlled variables.

3. The soft sensor modeling and real-time optimization control method of a de-butylation tower of an alkylation unit according to claim 1, characterized in that: The controlled variables are: feed flow, reflux flow, reboiler steam flow, condensate flow; the controlled variables are tower top pressure, tower top temperature, tower sensitive plate temperature, tower bottom temperature, tower bottom alkylate flow, tower top external n-butane flow, tower bottom pressure.

4. The soft sensor modeling and real-time optimization control method of a de-butanizer of an alkylation unit according to claim 1, characterized in that: The neural network model is a long short-term memory (LSTM) neural network, which includes an input layer, an LSTM layer, a fully connected layer, an activation function layer, a forgetting layer, and an output layer.

5. The soft sensor modeling and real-time optimization control method of a de-butylation tower of an alkylation unit according to claim 1, characterized in that: The internal system dynamic model of the MPC controller is the transfer function model of the system input and output. The transfer function model is as follows: in, is the dynamic model of the system, used to represent the control variables and the controlled variable the relationship between; is the gain of the system, which represents the overall amplification or attenuation effect of the system; is the time constant of the system, which is related to the inertia or energy storage of the system; It is the pure lag time of the system, which indicates the delay time that the change of system input affects the output.

6. The soft sensor modeling and real-time optimization control method of a de-butanizer of an alkylation unit according to claim 1, characterized in that: The MPC controller uses an internal model to predict future system responses based on real-time measured process parameters, and forms a closed-loop control with the PID control loop, thereby achieving rapid response and precise regulation of the debutanizer process.

7. The soft sensor modeling and real-time optimization control method of a de-butylation tower of an alkylation unit according to claim 1, characterized in that: The objective function of the genetic algorithm is as follows: ； ； Where: F is the objective function, OP1 is the weighted root mean square error between the controlled variable and the reference value, OP2 is the weighted root mean square error of the change value of the controlled variable, is the calculation weight of the control variable and the controlled variable, is the observed value of the controlled variable U and the controlled variable Y at time i, is the reference value of the controller’s controlled variable; Through crossover and mutation operations, the optimal operating points of a set of controllers are solved, the Pareto front is drawn, the optimal operating point is selected according to actual needs, and its corresponding parameter values ​​are assigned to the MPC controller, thereby improving the overall dynamic performance and stability of the system.

8. The soft sensor modeling and real-time optimization control method of a de-butanizer of an alkylation unit according to claim 1, characterized in that: In real-time optimization control, the real-time data collection step is to continuously collect the current working conditions, the parameter update step uses the real-time data to compare and correct the errors with the preset process model, and the real-time optimization is to optimize and adjust the control strategy in real time based on the online operation status to achieve the goals of reducing operating costs and optimizing energy consumption. The optimization goals are as follows: In the formula, is the optimization objective function of RTO, is the relationship function between n-butane concentration c and control variable u, It is the relationship function between the energy consumption E of the debutanizer and the control variable u.

Images