A data-driven rectification energy consumption optimization control method

Abstract

The present application relates to the technical field of chemical process control, and discloses a distillation energy consumption optimization control method based on data driving. According to the control cycle, the feed flow, the feed and the light component content of the tower top, the reflux ratio of the tower top and the heat load of the reboiler are synchronously collected, the heat load and the reflux ratio are combined to form a control vector, the lower limit of the product purity is combined to construct the separation efficiency, the unit efficiency heat load and the invalid heat energy, the overall equivalent unit energy consumption index is formed as an optimization target, a tower top purity prediction function based on historical data is established, the heat load and the reflux ratio are taken as inputs to form a purity constraint, a Lagrange multiplier related to the constraint is introduced, the energy consumption index and the purity constraint are unified into a dual optimization framework, the control vector is updated online in a safe range through a sub-gradient correction and a limited iteration, a convergence solution is obtained, and the convergence solution is executed and packaged and stored.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chemical process control, in particular to a distillation energy consumption optimization control method based on data driving. BACKGROUND

[0002] As a core unit operation in petrochemical, coal chemical, fine chemical and other industries, distillation process is widely used in raw material separation, product purification and by-product recovery. Traditional distillation column operation control mainly relies on experience parameter setting, conventional material and energy balance calculation and process optimization method based on mechanism model. In actual engineering application, the control object has the characteristics of high multivariable coupling, strong nonlinearity and variable working condition, and the energy consumption accounts for a high proportion, and there is often a difficult problem to balance between separation efficiency and energy consumption reduction.

[0003] In the prior art, distillation energy consumption optimization methods are roughly divided into optimization methods based on mechanism model and rule-based empirical control methods. Mechanism model optimization usually requires detailed modeling of mass transfer, heat transfer, flooding, wall flow and other physical processes in the tower, with numerous parameters and being easily affected by disturbances, making it difficult to adapt to complex scenarios of multiple categories and multiple processes. At the same time, many control models based on linear or static assumptions cannot accurately reflect the dynamic changes of the system, resulting in regulation lag, large energy consumption fluctuations, and difficulty in achieving dynamic balance between separation efficiency and energy consumption. In actual production, operating parameters such as reflux ratio, tower bottom heating power, feed flow rate, etc. are often adjusted in a rough way with fixed values or static intervals, ignoring the important role of real-time data in energy consumption dynamic optimization. In recent years, some technologies have proposed using data-driven optimization algorithms to manage distillation energy consumption, but existing methods mostly rely on ready-made model structures (such as linear regression, genetic algorithm or general artificial intelligence framework), and have insufficient adaptive modeling capability for key process parameters, and a large number of weights, factors, penalty coefficients, etc. are used in the optimization algorithm, which are difficult to be explained by physics, and have the problems of poor interpretability and insufficient landing. In addition, existing data-driven schemes often only focus on a single target, such as simply minimizing energy consumption or separately improving separation efficiency, and fail to achieve dynamic balance and hierarchical partition optimization of the two, and are also difficult to adapt to the multi-objective and multi-level optimization requirements in actual production, which is easy to cause separation product fluctuation, energy consumption control failure or process stability decline.

[0004] To this end, the case aims to propose a data-driven rectification energy consumption optimization control method, which acquires the key working condition parameters of the rectification tower such as feed flow, feed composition, tower top reflux ratio and reboiler heat load in real time according to a control period, introduces the coupling of'separation efficiency-heat load-inefficient energy' to construct an overall equivalent unit energy consumption index, and then takes the index as the optimization target, combines the tower top light component mole fraction prediction function and the purity constraint, converts the optimization problem into a dual problem of the Lagrange energy consumption optimization function, iteratively updates the Lagrange multiplier and the control variable vector through the sub-gradient method, and finally outputs the optimal control solution that can directly act on the reboiler and the condensation reflux adjusting device. SUMMARY

[0005] The present application provides a data-driven rectification energy consumption optimization control method, which solves the problems mentioned in the background art.

[0006] The present application provides the following technical scheme: a data-driven rectification energy consumption optimization control method, comprising:

[0007] The rectification feed mass flow, feed light component mole fraction, tower top reflux ratio, reboiler heat load and tower top light component mole fraction are acquired according to a control period, the control variable vector composed of the reboiler heat load and the tower top reflux ratio is set, and the lower limit of the target product purity at the tower top is obtained;

[0008] The separation efficiency function is constructed according to the feed light component mole fraction and the tower top light component mole fraction, and the unit separation efficiency heat load function and the inefficient heat energy term are established in combination with the reboiler heat load;

[0009] The overall equivalent unit energy consumption index is constructed based on the unit separation efficiency heat load and the inefficient heat energy term, the index is taken as the optimization objective function, and the control variable vector is minimized;

[0010] The tower top light component mole fraction prediction function is constructed, the reboiler heat load and the tower top reflux ratio are input, the tower top light component mole fraction is output, and the purity constraint function is formed by the predicted value and the lower limit of the target product purity at the tower top;

[0011] The Lagrange multiplier related to the tower top purity constraint is introduced, the optimization objective function and the tower top purity constraint function are combined into the Lagrange energy consumption optimization function, and the dual problem is constructed;

[0012] The initial value of the Lagrange multiplier and the control variable vector is set, the multiplier sub-gradient correction mechanism is constructed based on the prediction function and the purity lower limit, and the multiplier is iteratively updated in the sub-gradient direction;

[0013] Within the constraint range of the reboiler heat load and the column top reflux ratio, the control variable vector is iteratively updated based on the Lagrange energy consumption optimization function, and the convergence criterion is set according to the function and the multiplier change to obtain a converged solution or a solution reaching the preset maximum iteration number;

[0014] The control variable vector at the time of convergence is output to the reboiler heating adjustment device and the condensate reflux adjustment device as the optimal control solution of the current control period, and the working condition parameters, the optimal control variable vector, the column top light component mole fraction, the lower limit of the column top target product purity, the overall equivalent unit energy consumption index and the Lagrange multiplier are packaged and stored.

[0015] Optionally, the rectification feed mass flow, the feed light component mole fraction, the column top reflux ratio, the reboiler heat load, and the column top light component mole fraction are collected according to the control period, and the control variable vector composed of the reboiler heat load and the column top reflux ratio is set, and the lower limit of the column top target product purity is obtained, specifically including:

[0016] The system control period is set to collect data every five minutes, the control period number is generated starting from zero and increasing sequentially according to the control period setting, and the starting time corresponding to the control period number is specified for each control period, and the upper limit of the control period number is preset;

[0017] At the starting time of each control period, the rectification feed mass flow, the rectification feed light component mole fraction, the current column top reflux ratio, the current reboiler heat load and the current column top light component mole fraction are collected synchronously, and the quantified physical quantities of the rectification tower feed mass flow, the feed composition, the column top reflux state, the reboiler heating state and the column top composition under the current control period are formed correspondingly;

[0018] In each control period, the reboiler heat load and the column top reflux ratio are taken as components to construct the control variable vector, and the control variable vector represents the combined state of heat input and reflux operation under the current control period;

[0019] The lower limit of the column top target product purity is obtained, and the lower limit of the column top target product purity is taken as the target value of the subsequent column top purity constraint.

[0020] Optionally, the separation efficiency function is constructed according to the feed light component mole fraction and the column top light component mole fraction, and the unit separation efficiency heat load function and the invalid heat energy term are established in combination with the reboiler heat load, specifically including:

[0021] In each control cycle, a light component separation efficiency function is constructed according to the collected feed light component molar fraction and the column top light component molar fraction, so that the light component separation efficiency increases with the increase of the difference between the column top light component molar fraction and the feed light component molar fraction, and the effective separation efficiency is defined under the condition that the feed light component molar fraction and the column top light component molar fraction are both greater than or equal to zero and less than one, and the column top light component molar fraction is greater than or equal to the feed light component molar fraction;

[0022] In each control cycle, a unit separation efficiency heat load function is constructed according to the reboiler heat load and the separation efficiency, so that the reboiler heat load in each control cycle is combined with the separation efficiency and the positive offset constant to generate an equivalent heat load index corresponding to each unit separation efficiency;

[0023] When there is at least one control cycle with a separation efficiency greater than zero, the minimum positive separation efficiency is obtained in all control cycles with a separation efficiency greater than zero, and the positive offset constant is set to five percent of the minimum positive separation efficiency;

[0024] When all separation efficiencies in the control cycle numbers less than or equal to the preset upper limit are zero, the positive offset constant is set to a fixed small positive value, and the fixed small positive value is one percent;

[0025] In each control cycle, the reboiler heat load is multiplied by one minus the separation efficiency to form an invalid heat energy item, and the invalid heat energy item is regarded as the heat load part of the reboiler heat load that does not form effective separation efficiency.

[0026] Optionally, the overall equivalent unit energy consumption index is constructed based on the unit separation efficiency heat load and the invalid heat energy item, and the index is taken as an optimization objective function to perform minimization on the control variable vector, specifically including:

[0027] In each control cycle, the reboiler heat load and the invalid heat energy item are added to obtain an equivalent total heat load, and the overall equivalent unit energy consumption index is constructed in combination with the separation efficiency, so that the overall equivalent unit energy consumption index reflects the ratio relationship between the equivalent total heat load and the separation efficiency;

[0028] In each control cycle, the overall equivalent unit energy consumption index is taken as an optimization objective function to perform optimization search on the reboiler heat load and the column top reflux ratio in the control variable vector, so that the overall equivalent unit energy consumption index obtains a minimum value.

[0029] Optionally, the column top light component molar fraction prediction function is constructed, the reboiler heat load and the column top reflux ratio are input, and the column top light component molar fraction is output, and the predicted value and the lower limit of the column top target product purity form a column top purity constraint function, specifically including:

[0030] In each control cycle, the mole fraction of light components of the rectification feed in the next control cycle is taken as an amplitude parameter, the reboiler heat load and the reflux ratio of the column top in the current control cycle are taken as independent variables, a prediction function of the mole fraction of light components of the column top is constructed, the change rule of the mole fraction of light components of the column top with the reboiler heat load is described in the form of exponential decay, so that the predicted value of the mole fraction of light components of the column top changes between zero and the mole fraction of light components of the feed, and gradually approaches the upper limit of the mole fraction of light components of the feed with the increase of the reboiler heat load;

[0031] When the control cycle number is not less than two, the reboiler heat loads of the previous control cycle and the previous two control cycles are not equal, and the corresponding mole fraction of light components of the column top is between zero and one, the reboiler heat loads of the previous control cycle and the previous two control cycles and the mole fraction of light components of the column top are selected, the heat-driven sensitivity coefficient of the current control cycle is calculated based on the exponential decay form of the prediction function of the mole fraction of light components of the column top by making the outputs of the prediction function at the two sets of working conditions consistent with the actual mole fraction of light components of the column top at the same time;

[0032] When the control cycle number is less than two or the reboiler heat loads of the previous control cycle and the previous two control cycles are equal, two sets of steady-state calibration working conditions are selected before the system is put into operation, different reboiler heat load values and corresponding mole fractions of light components of the column top are given, and the mole fraction of light components of the column top is between zero and one, the heat-driven sensitivity coefficient is calculated by making the outputs of the prediction function of the mole fraction of light components of the column top at the two sets of steady-state calibration working conditions consistent with the corresponding mole fraction of light components of the column top;

[0033] In each control cycle, the reboiler heat load and the reflux ratio of the column top of the current control cycle are input into the prediction function of the mole fraction of light components of the column top, and the predicted value of the mole fraction of light components of the column top is obtained in combination with the heat-driven sensitivity coefficient;

[0034] In each control cycle, the predicted value of the mole fraction of light components of the column top is subtracted from the lower limit of the purity of the target product at the column top to construct a column top purity constraint function, and the column top purity constraint function is not less than zero as a constraint condition for meeting the purity requirement of the product at the column top.

[0035] Optionally, a Lagrange multiplier related to the column top purity constraint is introduced, the optimization objective function and the column top purity constraint function are combined into a Lagrange energy consumption optimization function, and a dual problem is constructed, specifically including:

[0036] The Lagrange multiplier related to the column top purity constraint is introduced, the overall equivalent unit energy consumption index and the column top purity constraint function are combined through the Lagrange multiplier and a penalty weight parameter, the penalty weight parameter is a preset non-negative constant, and a Lagrange energy consumption optimization function in each control cycle is formed;

[0037] A dual optimization problem is constructed for the Lagrange energy consumption optimization function, and under the premise that the Lagrange multiplier is not less than zero, an inner layer is adopted to perform a minimum operation on the control variable vector, and an outer layer is adopted to perform a maximum operation on the Lagrange multiplier, so as to obtain the minimum value of the Lagrange energy consumption optimization function under the condition of a given penalty weight parameter.

[0038] Optionally, the setting of the Lagrange multiplier and the control variable vector initial value, the construction of the multiplier sub-gradient correction mechanism based on the prediction function and the purity lower limit, and the iterative updating of the multiplier in the sub-gradient direction are specifically as follows:

[0039] In each control period, the maximum number of inner-layer iterations of the multiplier is set to fifty, the initial value of the Lagrange multiplier is set to zero, and the reboiler heat load and the column top reflux ratio measured at the starting moment of the current control period are taken as the iteration starting point of the control variable vector;

[0040] In each iteration, the top light component mole fraction prediction value is calculated through the top light component mole fraction prediction function according to the current iteration control variable vector, and the top light component mole fraction prediction value of the current iteration is formed;

[0041] The first-order partial derivative of the top light component mole fraction prediction value with respect to the reboiler heat load is obtained according to the structure of the top light component mole fraction prediction function, and the first-order partial derivative is taken as an intermediate quantity for subsequent gradient calculation;

[0042] A projection function is constructed for taking the non-negative part of any real number, the difference between the top target product purity lower limit and the current iteration top light component mole fraction prediction value is taken as the input of the projection function, and a non-negative purity deviation is obtained;

[0043] The non-negative purity deviation is taken as the Lagrange multiplier increment, and the Lagrange multiplier value is updated according to the rule of adding the multiplier increment to the current Lagrange multiplier value at the end of each iteration, so as to obtain the initial value of the Lagrange multiplier corresponding to the next iteration.

[0044] Optionally, the control variable vector is iteratively updated based on the Lagrange energy consumption optimization function within the constraint range of the reboiler heat load and the column top reflux ratio, and a convergence criterion is set according to the function and the multiplier change, so as to obtain a converged solution or a solution reaching a preset maximum iteration number, which specifically includes:

[0045] The safe lower limit value and the safe upper limit value of the reboiler heat load are obtained from the reboiler nameplate information, and the allowable upper limit of the column top reflux ratio is obtained;

[0046] In each iteration, the reboiler heat load and the column top reflux ratio are updated according to the first-order partial derivative direction of the current Lagrange energy consumption optimization function with respect to the reboiler heat load and the column top reflux ratio, so as to obtain the updated control variable vector candidate value;

[0047] According to the first-order partial derivative values of the Lagrange energy consumption optimization function with respect to the reboiler heat load and the column top reflux ratio at the initial iteration, in combination with the lower and upper safety limits of the reboiler heat load and the upper limit of the column top reflux ratio, step coefficients of the reboiler heat load and the column top reflux ratio are calculated respectively, so that the step coefficients are related to the respective allowed adjustment ranges and the initial gradient magnitudes;

[0048] The updated reboiler heat load and column top reflux ratio are subjected to a truncation operation according to the respective allowed ranges, and the part of the reboiler heat load less than the lower safety limit is corrected to the lower safety limit, the part of the reboiler heat load greater than the upper safety limit is corrected to the upper safety limit, the part of the column top reflux ratio less than zero is corrected to zero, and the part of the column top reflux ratio greater than the upper limit is corrected to the upper limit;

[0049] After the end of each iteration, the current Lagrange energy consumption optimization function value is calculated according to the updated control variable vector and the Lagrange multiplier, and the difference between the Lagrange energy consumption optimization function values of the current iteration and the previous iteration and the difference between the Lagrange multiplier values are calculated;

[0050] The convergence tolerances of the difference between the Lagrange energy consumption optimization function values and the difference between the Lagrange multiplier values are set, wherein the convergence tolerance of the difference between the Lagrange energy consumption optimization function values is one percent of the lower safety limit of the reboiler heat load, and the convergence tolerance of the difference between the Lagrange multiplier values is one thousandth, and when the difference between the Lagrange energy consumption optimization function values and the difference between the Lagrange multiplier values of two consecutive iterations are both less than the corresponding convergence tolerances, it is determined that the optimization calculation of the current control period converges;

[0051] When the convergence determination condition is met before the number of iterations reaches the preset maximum number of iterations, the control variable vector and the Lagrange multiplier corresponding to when the convergence determination condition is met are taken as the iteration results of the current control period, and when the convergence determination condition is not met when the number of iterations reaches the preset maximum number of iterations, the control variable vector and the Lagrange multiplier corresponding to the maximum number of iterations are taken as the iteration results of the current control period.

[0052] Optionally, the control variable vector at the time of convergence is taken as the optimal control solution of the current control period, and is output to the reboiler heat supply adjustment device and the condensate reflux adjustment device, and the working condition parameters, the optimal control variable vector, the column top light component mole fraction, the lower limit of the column top target product purity, the overall equivalent unit energy consumption index, and the Lagrange multiplier are subjected to data packaging and storage, specifically including: when the multiplier iteration update and the control variable optimization iteration meet the convergence determination condition, the number of convergence iterations is obtained, and the reboiler heat load and the column top reflux ratio corresponding to the number of convergence iterations are taken as the optimal control solution of the current control period;

[0053] Output the optimal reboiler heat duty of the current control cycle to the reboiler heat supply adjusting device, and output the optimal column top reflux ratio of the current control cycle to the condensate reflux adjusting device.

[0054] Write the distillation feed mass flow, the distillation feed light component molar fraction, the optimal reboiler heat duty, the optimal column top reflux ratio, the column top light component molar fraction, the column top target product purity lower limit, the overall equivalent unit energy consumption index and the Lagrange multiplier value after convergence under the current control cycle into the data recording module, and continue to perform the working condition parameter data acquisition, function construction, optimization calculation and control output based on the recorded data and new collected data in the next control cycle.

[0055] The present application has the following beneficial effects:

[0056] 1. The separation efficiency function based on the difference in the molar fraction of light components of the raw materials is proposed in relation to the relationship between the raw material composition and the output purity, and the unit separation efficiency heat load model and the invalid heat energy item are constructed based on the separation efficiency function and the reboiler heat load. This innovative method couples separation efficiency and heat load, so that the energy consumption optimization can consider the actual utilization rate of heat energy and the separation workload at the same time, thereby avoiding the one-sidedness of only focusing on the absolute size of the heat load or simply empirical adjustment. Compared with the prior art, the present model can automatically update the offset constant under different working conditions, ensuring that a reasonable energy consumption benchmark can be obtained when the separation efficiency is low; when the efficiency is high, the marginal revenue of a small amount of additional heat energy on the purity improvement can be truly reflected, improving the accuracy of energy consumption evaluation and realizing the self-adaptive ability of the model to working condition changes.

[0057] 2. The overall equivalent unit energy consumption index constructed by the present scheme integrates heat load and efficiency model to form a comprehensive target function that can be measured and is proportional to the actual energy consumption and separation efficiency. Compared with the traditional optimization target that only minimizes heat load or meets the purity standard, the present index takes into account the balance of both, which can ensure separation effect while pursuing the lowest energy consumption. This innovation is beneficial to clear understanding of the control target, easy implementation of calculation and comparison, and avoids the problem of difficult weight determination in multi-objective optimization. Compared with the prior art of directly setting a fixed reflux ratio or reboiler heat load, the present scheme provides a unified quantitative standard, simplifies the difficulty of control strategy design, and improves the operability and interpretability of optimization.

[0058] 3、In view of the nonlinear characteristics of the purity of the overhead product changing with the working conditions, the present scheme proposes an exponential decay form of the light component mole fraction prediction function, and obtains the heat drive sensitivity coefficient through historical calibration and online updating, so as to realize the consideration of dynamic and steady state conditions by the model. This method innovatively introduces the exponential decay characteristics into the online model, which can effectively simulate the diminishing effect of heat load on purity improvement, and avoids the distortion of common linear regression or black box model in a wide range of working conditions. In practical application, the prediction function can quickly respond to the purity change caused by changes in raw material composition or equipment, compared with the existing technology relying on manual experience adjustment, which improves the adaptability and control accuracy of the rectification system to changes.

[0059] 4、The present scheme unifies the energy consumption target and the purity constraint into the same function framework through the Lagrange multiplier, and constructs a dual optimization problem, adopts a double-layer iteration structure of minimizing the control variables in the inner layer and maximizing the multipliers in the outer layer. The product purity hard constraint and energy consumption soft target can be met at the same time; through the structure of the dual problem, the constraint is relaxed into the objective function, so that the constrained optimization which is difficult to solve directly is transformed into a feasible iterative calculation problem. Compared with the penalty function or direct constraint projection commonly used in the prior art, the method is more precise in handling constraints, can achieve a better balance between control convergence speed and constraint severity, and ensures stability and convergence effect.

[0060] 5、For the update of the Lagrange multiplier, the present scheme designs an iteration rule combining the sub-gradient direction and the projection operation, which ensures that the multiplier gradually approaches the optimal value under the non-negative constraint. Through the sub-gradient increment mechanism, the multiplier increment can be adjusted adaptively, avoiding the oscillation caused by difficult step length setting or direct gradient descent; the non-negative multiplier is ensured by the projection function, which can strictly maintain the effectiveness of the purity constraint. The traditional multiplier update problem of slow or unstable convergence is effectively solved, making the online real-time calculation more reliable, and the method is more intelligent and efficient compared with the existing fixed step length or experience step length method.

[0061] 6、In the iteration, the present scheme performs a descent update on the control variable vector according to the Lagrange energy consumption optimization function, and sets multiple convergence conditions according to the function value and the multiplier change amount. Considering the dual information of the objective function and the constraint, the accuracy requirement can be met while avoiding premature termination or invalid long iteration; by introducing the lower limit value and the upper limit value truncation operation, the real-time protection of the control variable update is realized, avoiding the overrunning or instability of the equipment. Different from the single target or artificial experience convergence determination in the prior art, the convergence determination of the present scheme is more objective and quantifiable, which is conducive to the reliable implementation of the online control system.

[0062] 7、The scheme obtains a convergent solution, and directly issues the control variable vector to the reboiler and condensing system, and simultaneously encapsulates the key operating parameters, optimal variables and energy consumption indexes for storage, so as to provide data support for model updating and working condition evaluation in subsequent periods. The optimization result is combined with historical data in a closed loop, so that the system has a continuous self-learning ability; and through structured data storage, the operation record and fault tracking work are simplified. Compared with the prior art which mainly relies on manual recording or scattered storage, the scheme has a significant improvement in operation and reliability, and effectively supports the long-term application and optimization iteration of the intelligent rectification control system. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The flowchart of the present application is shown. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0065] Embodiment, refer to Figure 1 A data-driven rectification energy consumption optimization control method, comprising:

[0066] According to the control period, the rectification feed mass flow, the feed light component molar fraction, the column top reflux ratio, the reboiler heat load, and the column top light component molar fraction are collected, the control variable vector composed of the reboiler heat load and the column top reflux ratio is set, and the lower limit of the column top target product purity is obtained;

[0067] According to the feed light component molar fraction and the column top light component molar fraction, a separation efficiency function is constructed, and a unit separation efficiency heat load function and an invalid heat energy item are established in combination with the reboiler heat load;

[0068] Based on the unit separation efficiency heat load and the invalid heat energy item, an overall equivalent unit energy consumption index is constructed, which is taken as an optimization objective function, and the control variable vector is minimized;

[0069] A column top light component molar fraction prediction function is constructed, the reboiler heat load and the column top reflux ratio are input, the column top light component molar fraction is output, and the predicted value and the lower limit of the column top target product purity form a column top purity constraint function;

[0070] The Lagrange multiplier related to the column top purity constraint is introduced, the optimization objective function and the column top purity constraint function are combined into a Lagrange energy consumption optimization function, and a dual problem is constructed;

[0071] Setting Lagrange multipliers and control variable vector iteration initial value, constructing multiplier sub-gradient correction mechanism based on prediction function and purity lower limit, and updating multiplier in sub-gradient direction;

[0072] Within the constraint range of reboiler heat load and column top reflux ratio, the control variable vector is iteratively updated based on the Lagrange energy consumption optimization function, and the convergence criterion is set according to the function and the multiplier change to obtain a converged solution or a solution that reaches the preset maximum iteration number;

[0073] The control variable vector at the time of convergence is output to the reboiler heating adjustment device and the condensate reflux adjustment device as the optimal control solution of the current control period, and the working condition parameters, optimal control variable vector, column top light component molar fraction, column top target product purity lower limit, overall equivalent unit energy consumption index and Lagrange multiplier are packaged and stored.

[0074] By synchronously collecting five key working condition parameters, including feed mass flow rate, feed light component content, column top reflux state, reboiler heat load, and column top product light component content, in each control period, and combining the control variable vector composed of heat load and reflux ratio with the real-time obtained product purity lower limit, the scheme can timely establish a control model reflecting the actual operating conditions when the working conditions change; by constructing a unit efficiency heat load function coupled with separation efficiency and heat load and an invalid heat energy item, an overall equivalent unit energy consumption index is formed, so that the energy consumption optimization target truly reflects the heat energy input required for unit separation efficiency and the heat loss that does not form separation efficiency; then, using the purity prediction function trained by historical data, the column top light component content under the given combination of heat load and reflux ratio is predicted online, and a constraint function is formed in combination with the product purity lower limit; further, the equivalent energy consumption index and the purity constraint are unified to the same optimization framework through the Lagrange multiplier, and a dual optimization problem is constructed, so as to realize the synchronous optimization of energy consumption and purity constraint; finally, through the sub-gradient correction mechanism and dual iteration, the heat load and reflux ratio are iteratively updated online under the premise of ensuring that the purity hard constraint is not violated, until the convergence condition is met or the maximum iteration number is reached, and the converged solution is immediately issued to the adjustment device and packaged and stored. This method solves the problem that the traditional empirical regulation cannot balance product purity and energy consumption minimization, realizes efficient energy consumption utilization and product quality double protection, and has the beneficial effects of real-time adaptation to working condition changes, self-learning optimization, high precision, and reliable convergence.

[0075] The control variable vector composed of reboiler heat load and column top reflux ratio is set, and the column top target product purity lower limit is obtained, specifically including:

[0076] Set the system control cycle to collect data every five minutes, generate control cycle numbers that start from zero and increment sequentially according to the control cycle setting, and specify the start time corresponding to the control cycle number for each control cycle, and preset the upper limit of the control cycle number.

[0077] At the beginning of each control cycle, the mass flow rate of the distillation feed, the mole fraction of the light component in the distillation feed, the current reflux ratio at the top of the column, the current heat load of the reboiler, and the current mole fraction of the light component at the top of the column are collected synchronously to form quantitative physical quantities of the mass flow rate of the distillation column feed, feed composition, reflux status at the top of the column, reboiler heating status, and column composition at the top of the column under the current control cycle.

[0078] Within each control cycle, the reboiler heat load and the top reflux ratio are used as components to construct a control variable vector, which represents the combined state of heat input and reflux operation under the current control cycle.

[0079] Obtain the lower limit of the purity of the target product at the top of the tower, and use the lower limit of the purity of the target product at the top of the tower as the target value for subsequent purity constraints at the top of the tower.

[0080] Further specific implementation steps include:

[0081] Set the system control cycle to Collect data once, the first The start time of each cycle is ;in, The discrete sequence number of the control cycle; For the first The continuous time corresponding to the start time of each control cycle; The upper limit of the preset control cycle number;

[0082] During each control cycle, the following quantitative physical variables are collected synchronously on-site, including: distillation feed mass flow rate. Mole fraction of light components in distillation feed Current tower top reflux ratio Current reboiler heat load The current molar fraction of the light component at the top of the column ;in, , , , , They were respectively in the second The start time of each control cycle The corresponding distillation column feed mass flow rate, feed light component mole fraction, actual overhead reflux ratio, reboiler heat load, and overhead light component mole fraction;

[0083] Construct the first Control variable vector within each control cycle ;

[0084] Obtaining the lower limit of the purity of the target product at the column top, denoted as .

[0085] The separation efficiency function is constructed according to the molar fraction of the light component in the feed and the molar fraction of the light component at the column top, and the unit separation efficiency heat load function and the invalid heat energy item are established by combining the reboiler heat load, specifically including:

[0086] In each control period, the light component separation efficiency function is constructed according to the collected molar fraction of the light component in the feed and the molar fraction of the light component at the column top, so that the light component separation efficiency increases with the increase of the difference between the molar fraction of the light component at the column top and the molar fraction of the light component in the feed, and the effective separation efficiency is defined under the condition that the molar fraction of the light component in the feed and the molar fraction of the light component at the column top are both greater than or equal to zero and less than one, and the molar fraction of the light component at the column top is greater than or equal to the molar fraction of the light component in the feed;

[0087] In each control period, the unit separation efficiency heat load function is constructed according to the reboiler heat load and the separation efficiency, and the reboiler heat load in each control period is combined with the separation efficiency and the positive offset constant to generate an equivalent heat load index corresponding to each unit separation efficiency;

[0088] When there is at least one control period with separation efficiency greater than zero, the minimum positive separation efficiency is obtained in all control periods with separation efficiency greater than zero, and the positive offset constant is set to five percent of the minimum positive separation efficiency;

[0089] When all separation efficiencies in the control period number are equal to or less than the preset upper limit range, the positive offset constant is set to a fixed small positive value, and the fixed small positive value is one percent;

[0090] In each control period, the invalid heat energy item is obtained by taking the product of the reboiler heat load and one minus the separation efficiency, and the invalid heat energy item is regarded as the heat load part in the reboiler heat load which does not form effective separation efficiency.

[0091] Further specific implementation operations include:

[0092] The separation efficiency function is constructed, specifically: ; Wherein, is the separation efficiency of the light component at the column top relative to the feed composition in the i-th control period; the molar fraction of the feed and the column top satisfies: , , , ;

[0093] The unit efficiency heat load function is constructed, specifically: ; Wherein, is the i-th the equivalent heat duty corresponding to the unit separation efficiency in each control period; To avoid the positive offset constant introduced by the denominator being zero, specifically:

[0094] When there is satisfies : ; wherein, is the control period index; is the separation efficiency in the first control period;

[0095] When for all have : ;

[0096] The invalid heat energy term is constructed as: ; wherein, is the heat duty considered invalid in the first control period.

[0097] The overall equivalent unit energy consumption index is constructed based on the unit separation efficiency heat duty and the invalid heat energy term, and the index is used as an optimization objective function to perform minimization on the control variable vector, specifically including:

[0098] In each control period, the reboiler heat duty and the invalid heat energy term are added to obtain the equivalent total heat duty, and the overall equivalent unit energy consumption index is constructed in combination with the separation efficiency, so that the overall equivalent unit energy consumption index reflects the ratio relationship between the equivalent total heat duty and the separation efficiency;

[0099] In each control period, the overall equivalent unit energy consumption index is used as an optimization objective function to perform optimization search on the reboiler heat duty and the column top reflux ratio in the control variable vector, so that the overall equivalent unit energy consumption index obtains the minimum value.

[0100] Further specific implementation operations include:

[0101] The overall objective function is constructed as: ; wherein, is the equivalent unit energy consumption index under the first control period; ;

[0102] The optimization objective set in each control period is: .

[0103] The tower top light component mole fraction prediction function is constructed, the reboiler heat duty and the column top reflux ratio are input, and the tower top light component mole fraction is output, and the prediction value and the tower top target product purity lower limit form a tower top purity constraint function, specifically including:

[0104] In each control cycle, the mole fraction of light components of the rectification feed of the next control cycle is taken as the amplitude parameter, the reboiler heat load and the reflux ratio of the top of the current control cycle are taken as the independent variables, a prediction function of the mole fraction of light components of the top is constructed, the change rule of the mole fraction of light components of the top with the reboiler heat load is described in the form of exponential decay, the predicted value of the mole fraction of light components of the top changes between zero and the mole fraction of light components of the feed, and gradually approaches the upper limit of the mole fraction of light components of the feed with the increase of the reboiler heat load;

[0105] When the control cycle number is not less than two, the reboiler heat loads of the previous control cycle and the previous two control cycles are not equal, and the corresponding mole fraction of light components of the top is between zero and one, the reboiler heat loads and the mole fraction of light components of the top of the previous control cycle and the previous two control cycles are selected, the heat-driven sensitivity coefficient of the current control cycle is calculated based on the exponential decay form of the prediction function of the mole fraction of light components of the top by making the outputs of the prediction function at the two sets of working conditions consistent with the actual mole fraction of light components of the top at the same time;

[0106] When the control cycle number is less than two or the reboiler heat loads of the previous control cycle and the previous two control cycles are equal, two sets of steady-state calibration working conditions are selected before the system is put into operation, different reboiler heat load values and corresponding mole fractions of light components of the top are given, and the mole fraction of light components of the top is between zero and one, the heat-driven sensitivity coefficient is calculated by making the outputs of the prediction function of the mole fraction of light components of the top at the two sets of steady-state calibration working conditions consistent with the corresponding mole fraction of light components of the top;

[0107] In each control cycle, the reboiler heat load and the reflux ratio of the top of the current control cycle are input into the prediction function of the mole fraction of light components of the top, and the predicted value of the mole fraction of light components of the top is obtained in combination with the heat-driven sensitivity coefficient;

[0108] In each control cycle, the predicted value of the mole fraction of light components of the top is subtracted from the lower limit of the purity of the target product of the top to construct a purity constraint function of the top, and the constraint condition that the purity constraint function of the top is not less than zero is taken as the constraint condition for meeting the purity requirement of the product of the top.

[0109] Further specific implementation operations include:

[0110] The prediction function of the purity of the top is constructed as: ; wherein, is the prediction function of the purity of the top, the input is the reboiler heat load and the reflux ratio of the first control cycle, and the output is the predicted mole fraction of light components of the top; represents the decay characteristics of the purity improvement of the reboiler heat load under the heat-driven sensitivity. The heat-driven sensitivity coefficient estimated from the historical data in the first control cycle is specifically:

[0111] S401, when and and , , ;

[0112] S402, when or :

[0113] Before the system is put into operation, two groups of steady-state calibration working conditions are selected: , , meet , , , let: ; wherein, , respectively represent the reboiler heat load value under the first group and the second group of calibration working conditions; , respectively represent the mole fraction of light components at the top of the column under the first group of calibration working conditions and the second group of calibration working conditions;

[0114] The constraint function of the top purity is constructed, and specifically: ; wherein, is the constraint function of the top purity, and the input is the control variable of the first control cycle, and the output is the difference between the predicted purity at the top and the target lower limit. The Lagrange multiplier related to the constraint of the top purity is introduced, the objective function and the constraint function of the top purity are combined into a Lagrange energy consumption optimization function, and a dual problem is constructed, specifically including:

[0115] The Lagrange multiplier related to the constraint of the top purity is introduced, and the overall equivalent unit energy consumption index and the constraint function of the top purity are combined through the Lagrange multiplier and the penalty weight parameter, wherein the penalty weight parameter is a preset non-negative constant, forming a Lagrange energy consumption optimization function in each control cycle;

[0116] A dual optimization problem is constructed for the Lagrange energy consumption optimization function, and under the premise that the value of the Lagrange multiplier is not less than zero, the structure of the inner layer performing minimization operation on the control variable vector and the outer layer performing maximization operation on the Lagrange multiplier is adopted, to obtain the minimum value of the Lagrange energy consumption optimization function under the given penalty weight parameter.

[0117]

[0118] Further specific implementation operations include:​​

[0119] By introducing Lagrange multipliers, we construct the Lagrange function as follows: ;in, For the first Lagrange multipliers related to the purity constraint at the top of the tower in each control cycle; Let be a Lagrangian function, representing the state at the th... Lagrange energy consumption optimization function for each control cycle;

[0120] The dual optimization problem is constructed based on the Lagrange function, specifically as follows: ;in, For a given Under what conditions can we find Minimal combination of control variables; In order to constrain In the case of non-negative values, find the most unfavorable penalty weight.

[0121] The process of setting initial values ​​for the Lagrange multipliers and control variable vector iterations, constructing a multiplier subgradient correction mechanism based on the prediction function and purity lower bound, and iteratively updating the multipliers according to the subgradient direction specifically includes:

[0122] In each control cycle, the maximum number of iterations of the inner layer of the multiplier is set to fifty, the initial value of the Lagrange multiplier is set to zero, and the reboiler heat load and the top reflux ratio measured at the beginning of the current control cycle are used as the starting point of the control variable vector iteration.

[0123] In each iteration, the predicted value of the light component mole fraction at the top of the tower is calculated using the light component mole fraction prediction function at the top of the tower according to the current iteration control variable vector, thus forming the predicted value of the light component mole fraction at the top of the tower for the current iteration.

[0124] The first partial derivative of the predicted mole fraction of the light component at the top of the column with respect to the reboiler heat load is obtained based on the structure of the prediction function for the mole fraction of the light component at the top of the column, and the first partial derivative is used as an intermediate quantity for subsequent gradient calculation.

[0125] Construct a projection function that takes the non-negative part of any real number, and use the difference between the lower limit of the purity of the target product at the top of the tower and the predicted value of the mole fraction of the light component at the top of the tower in the current iteration as the input of the projection function to obtain the non-negative purity deviation;

[0126] The non-negative purity deviation is used as the Lagrange multiplier increment. At the end of each iteration, the Lagrange multiplier values ​​are updated according to the rule of adding the multiplier increment to the current Lagrange multiplier value, so as to obtain the initial value of the Lagrange multiplier for the next iteration.

[0127] Further specific implementation steps include:

[0128] Set the maximum number of iterations. , initialize the Lagrange multiplier as , and take the measured current operating value as the initial control variable iteration starting point, specifically: , ; wherein, is the maximum number of iterations allowed when solving the inner optimization problem in the first control cycle; is the value of the first control cycle multiplier at the initial iteration time; , are the initial values of the reboiler heat load and reflux ratio in the first control cycle at the iteration start;

[0129] In each round of iteration , the following steps are performed, specifically:

[0130] S601, according to the current iteration control variable, calculate the predicted tower top purity, specifically: ; wherein, is the iteration number of the Lagrange optimization in the first control cycle; , are the reboiler heat load and reflux ratio in the first iteration of the first control cycle; is the tower top light component mole fraction calculated by the prediction function in the first iteration of the first control cycle;

[0131] S602, calculate the partial derivative of to , specifically: ; wherein, is the partial derivative of the tower top purity prediction function to the heat load in the first iteration of the first control cycle;

[0132] S603, first construct a projection function for taking the non-negative part of a real number: ; wherein, is the projection function for taking the non-negative part of a real number ;

[0133] Then calculate the difference between the lower limit of the tower target purity and the predicted tower top purity in the first iteration of the first control cycle;

[0134] S604, calculate the multiplier increment, specifically: ; wherein, is the increment of the Lagrange multiplier at the first iteration of the control cycle;

[0135] S605, update the Lagrange multiplier: ; wherein, is the updated value of the Lagrange multiplier after the first iteration of the control cycle.

[0136] The control variable vector is iteratively updated based on the Lagrange energy consumption optimization function within the constraint range of the reboiler heat load and the overhead reflux ratio, and the convergence criterion is set according to the function and the multiplier change to obtain a converged solution or a solution that reaches the preset maximum iteration number, specifically including:

[0137] Obtain the safe lower limit value and the safe upper limit value of the reboiler heat load from the reboiler nameplate information, and obtain the allowed upper limit of the overhead reflux ratio;

[0138] In each iteration, according to the first-order partial derivative direction of the current Lagrange energy consumption optimization function with respect to the reboiler heat load and the overhead reflux ratio, perform a descent update on the reboiler heat load and the overhead reflux ratio to obtain the updated control variable vector candidate value;

[0139] According to the first-order partial derivative value of the Lagrange energy consumption optimization function with respect to the reboiler heat load and the overhead reflux ratio at the initial iteration, combined with the safe lower limit value and the safe upper limit value of the reboiler heat load and the allowed upper limit of the overhead reflux ratio, calculate the step length coefficient for the reboiler heat load and the overhead reflux ratio respectively, so that the step length coefficient is related to the allowed adjustment range and the initial gradient magnitude of each;

[0140] Perform truncation operation on the updated reboiler heat load and overhead reflux ratio according to their respective allowed ranges, correct the part of the reboiler heat load less than the safe lower limit value to the safe lower limit value, correct the part of the reboiler heat load greater than the safe upper limit value to the safe upper limit value, correct the part of the overhead reflux ratio less than zero to zero, and correct the part of the overhead reflux ratio greater than the allowed upper limit to the allowed upper limit;

[0141] After each iteration, calculate the current Lagrange energy consumption optimization function value according to the updated control variable vector and the Lagrange multiplier, and calculate the difference between the Lagrange energy consumption optimization function value and the Lagrange multiplier value between the current iteration and the previous iteration;

[0142] The convergence tolerance of the numerical difference of the Lagrange energy consumption optimization function and the convergence tolerance of the numerical difference of the Lagrange multiplier are set, wherein the convergence tolerance of the numerical difference of the Lagrange energy consumption optimization function is one percent of the lower limit value of the reboiler heat load safety, and the convergence tolerance of the numerical difference of the Lagrange multiplier is one thousandth, and when the numerical difference of the Lagrange energy consumption optimization function and the numerical difference of the Lagrange multiplier between two consecutive iterations are both less than the corresponding convergence tolerance, it is determined that the optimization calculation of the current control period converges.

[0143] When the convergence determination condition is met before the number of iterations reaches the preset maximum number of iterations, the control variable vector and the Lagrange multiplier corresponding to when the convergence determination condition is met are taken as the iteration result of the current control period, and when the convergence determination condition is not met when the number of iterations reaches the preset maximum number of iterations, the control variable vector and the Lagrange multiplier corresponding to the maximum number of iterations are taken as the iteration result of the current control period.

[0144] Further specific implementation operations include:

[0145] The lower limit value and the upper limit value of the reboiler heat load are obtained from the reboiler nameplate, and are denoted as and respectively.

[0146] The upper limit of the reflux ratio of the column top is obtained, and is denoted as .

[0147] In each iteration, the control variables and are updated, specifically: , ; wherein , are the updated heat load and reflux ratio values after the th control period and the th iteration; is the step coefficient when updating the heat load ; is the step coefficient when updating the reflux ratio .

[0148] In the th iteration, the specific expression of the first-order partial derivative is: , , .

[0149] Accordingly, we have: , .

[0150] The step coefficients and are obtained as follows: , .

[0151] For With The following truncation operations are performed respectively: , ;

[0152] After each iteration, the current Lagrange function value is calculated: ; where is the Lagrange function value calculated at the th iteration of the th control period;

[0153] The convergence tolerance parameters are set as: , ; where is the convergence tolerance of the Lagrange function value; is the convergence tolerance of the Lagrange multiplier change;

[0154] After the th iteration is completed, the following calculations are performed: , ; where is the absolute change in the Lagrange function value between the th iteration and the th iteration of the th control period; is the absolute change in the Lagrange multiplier between the th iteration and the th iteration of the th control period;

[0155] The convergence determination condition is set as: , where is the logical AND operation;

[0156] When the convergence determination condition is met, it is determined that the optimization calculation of the current period has converged;

[0157] If the convergence determination condition is not met when the number of iterations reaches , the result at the th iteration is taken as the calculation result of the current period.

[0158] The control variable vector at convergence is output to the reboiler heat supply adjustment device and the condensation reflux adjustment device as the optimal control solution of the current control period, and the working condition parameters, the optimal control variable vector, the tower top light component molar fraction, the tower top target product purity lower limit, the overall equivalent unit energy consumption index, and the Lagrange multiplier execute data encapsulation and storage, specifically including: when the multiplier iteration update and the control variable optimization iteration meet the convergence judgment condition, the convergence iteration number is obtained, and the reboiler heat load and the tower top reflux ratio corresponding to the convergence iteration number are taken as the optimal control solution of the current control period;

[0159] The optimal reboiler heat load of the current control period is output to the reboiler heat supply adjustment device, and the optimal tower top reflux ratio of the current control period is output to the condensation reflux adjustment device.

[0160] The distillation feed mass flow, the distillation feed light component molar fraction, the optimal reboiler heat load, the optimal tower top reflux ratio, the tower top light component molar fraction, the tower top target product purity lower limit, the overall equivalent unit energy consumption index, and the converged Lagrange multiplier value under the current control period are written into the data record module, and the working condition parameter data acquisition, function construction, optimization calculation and control output are continued based on the recorded data and new collected data in the next control period.

[0161] Further specific implementation operations include:

[0162] When the iteration process meets the convergence judgment condition, the final iteration step number is recorded as The corresponding control variable is taken as the optimal control solution of the current period: , ; wherein, is the final iteration serial number when the iteration converges in the th control period; is the optimal reboiler heat load value obtained in the th control period; is the optimal reboiler reflux ratio obtained in the th control period;

[0163] The optimal control variable obtained in the current period is output to the reboiler heat supply adjustment device and the condensation reflux adjustment device, and the data acquisition, function construction, optimization calculation and control output process are continued to be repeatedly executed in the subsequent control period;

[0164] At the same time, the following data set is written into the data record module: .

[0165] It is to be noted that, as used in this document, the term "indicia" is intended to encompass any type of data, information, or other content, whether in the form of text, graphics, images, video, audio, or otherwise. It is also to be noted that, as used in this document, the terms "first" and "second" are merely used to distinguish one entity or operation from another, and do not necessarily imply or suggest any actual relationship or order between the entities or operations. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0166] The above description is merely that of the preferred embodiments of the present application, and it is to be noted that various modifications and improvements made to the present application without departing from the technical principles of the present application shall fall within the scope of protection of the present application.

Claims

1. A data-driven based distillation energy consumption optimization control method, characterized in that, The method comprises the following steps: acquiring the rectification feed mass flow rate, the feed light component molar fraction, the column top reflux ratio, the reboiler heat load, and the column top light component molar fraction according to a control period, setting a control variable vector composed of the reboiler heat load and the column top reflux ratio, and obtaining a lower limit of the column top target product purity; constructing a separation efficiency function according to the feed light component molar fraction and the column top light component molar fraction, and combining the reboiler heat load to establish a unit separation efficiency heat load function and an invalid heat energy item; constructing an overall equivalent unit energy consumption index based on the unit separation efficiency heat load and the invalid heat energy item, taking the index as an optimization objective function, and performing minimization on the control variable vector; constructing a column top light component molar fraction prediction function, inputting the reboiler heat load and the column top reflux ratio, outputting the column top light component molar fraction, and forming a column top purity constraint function with the predicted value and the lower limit of the column top target product purity; introducing a Lagrange multiplier related to the column top purity constraint, combining the optimization objective function and the column top purity constraint function into a Lagrange energy consumption optimization function, and constructing a dual problem; setting the initial value of the Lagrange multiplier and the control variable vector, constructing a multiplier sub-gradient correction mechanism based on the prediction function and the lower limit of the purity, and iteratively updating the multiplier in the sub-gradient direction; iteratively updating the control variable vector based on the Lagrange energy consumption optimization function within the constraint range of the reboiler heat load and the column top reflux ratio, setting a convergence criterion according to the function and the multiplier change, and obtaining a converged solution or a solution reaching a preset maximum iteration number; taking the control variable vector at the convergence as the optimal control solution of the current control period, outputting to the reboiler heat supply adjustment device and the condensation reflux adjustment device, and performing data packaging and storage on the working condition parameters, the optimal control variable vector, the column top light component molar fraction, the lower limit of the column top target product purity, the overall equivalent unit energy consumption index, and the Lagrange multiplier.

2. The data-driven rectifying energy consumption optimization control method according to claim 1, wherein, The method comprises the following steps: setting the system control period to collect data every five minutes, generating control period numbers starting from zero and increasing sequentially according to the control period setting, and assigning a starting time corresponding to the control period number to each control period, and presetting an upper limit of the control period number; at the starting time of each control period, synchronously collecting the rectification feed mass flow rate, the rectification feed light component molar fraction, the current column top reflux ratio, the current reboiler heat load, and the current column top light component molar fraction, and correspondingly forming the quantified physical quantities of the rectification tower feed mass flow rate, the feed composition, the column top reflux state, the reboiler heat supply state, and the column top composition in the current control period; in each control period, constructing a control variable vector with the reboiler heat load and the column top reflux ratio as components, and the control variable vector representing the combined state of heat input and reflux operation in the current control period; obtaining the lower limit of the column top target product purity, and taking the lower limit of the column top target product purity as the target value of the subsequent column top purity constraint.

3. The data-driven based rectifying energy consumption optimization control method according to claim 2, wherein, The separation efficiency function is constructed according to the mole fraction of the light component of the feed and the mole fraction of the light component at the top of the column, and the unit separation efficiency heat load function and the invalid heat energy item are established by combining the reboiler heat load, and specifically include: In each control period, the light component separation efficiency function is constructed according to the collected mole fraction of the light component of the feed and the mole fraction of the light component at the top of the column, so that the light component separation efficiency increases with the increase of the difference between the mole fraction of the light component at the top of the column and the mole fraction of the light component of the feed, and the effective separation efficiency is defined under the condition that the mole fraction of the light component of the feed and the mole fraction of the light component at the top of the column are greater than or equal to zero and less than one, and the mole fraction of the light component at the top of the column is greater than or equal to the mole fraction of the light component of the feed; In each control period, the unit separation efficiency heat load function is constructed according to the reboiler heat load and the separation efficiency, and the reboiler heat load in each control period is combined with the separation efficiency and the positive offset constant to generate an equivalent heat load index corresponding to each unit separation efficiency; When there is at least one control period with a separation efficiency greater than zero, the minimum positive separation efficiency is obtained in all control periods with a separation efficiency greater than zero, and the positive offset constant is set to five percent of the minimum positive separation efficiency; When all separation efficiencies in the control period number are equal to or less than the preset upper limit, the positive offset constant is set to a fixed small positive value, and the fixed small positive value is one percent; In each control period, the invalid heat energy item is obtained by taking the product of the reboiler heat load and one minus the separation efficiency, and the invalid heat energy item is regarded as the heat load part of the reboiler heat load which does not form effective separation efficiency.

4. The data-driven rectifying energy consumption optimization control method according to claim 3, wherein, The overall equivalent unit energy consumption index is constructed based on the unit separation efficiency heat load and the invalid heat energy item, and the index is taken as an optimization objective function to perform minimization on the control variable vector, and specifically includes: In each control period, the equivalent total heat load is obtained by adding the reboiler heat load and the invalid heat energy item, and the overall equivalent unit energy consumption index is constructed in combination with the separation efficiency, so that the overall equivalent unit energy consumption index reflects the ratio relationship between the equivalent total heat load and the separation efficiency; In each control period, the overall equivalent unit energy consumption index is taken as an optimization objective function to perform optimization search on the reboiler heat load and the top reflux ratio in the control variable vector, so that the overall equivalent unit energy consumption index obtains a minimum value.

5. The data-driven rectifying energy consumption optimization control method according to claim 4, wherein, The top light component mole fraction prediction function is constructed, the reboiler heat load and the top reflux ratio are input, and the top light component mole fraction is output, and the top purity constraint function is formed by the predicted value and the lower limit of the top target product purity, and specifically includes: In each control period, the rectification feed light component mole fraction of the next control period is taken as an amplitude parameter, the reboiler heat load and the top reflux ratio of the current control period are taken as independent variables, the top light component mole fraction prediction function is constructed, the exponential decay form is used to describe the change law of the top light component mole fraction with the reboiler heat load, the top light component mole fraction prediction value is changed between zero and the feed light component mole fraction, and gradually approaches the upper limit of the feed light component mole fraction with the increase of the reboiler heat load; When the control cycle number is not less than two and the reboiler heat duty of the previous control cycle and the second previous control cycle is not equal and the corresponding overhead light component mole fraction is between zero and one, the reboiler heat duty and the overhead light component mole fraction of the previous control cycle and the second previous control cycle are selected, the heat-driven sensitivity coefficient of the current control cycle is calculated based on the exponential decay form of the overhead light component mole fraction prediction function by making the outputs of the prediction function at the two sets of working conditions consistent with the actual overhead light component mole fraction at the same time; When the control cycle number is less than two or the reboiler heat duty of the previous control cycle and the second previous control cycle is equal, two sets of steady-state calibration working conditions are selected before the system is put into operation, different reboiler heat duty values and corresponding overhead light component mole fractions are given, and the overhead light component mole fraction is between zero and one, the heat-driven sensitivity coefficient is calculated by making the outputs of the overhead light component mole fraction prediction function at the two sets of steady-state calibration working conditions consistent with the corresponding overhead light component mole fraction; In each control cycle, the reboiler heat duty and the overhead reflux ratio of the current control cycle are input into the overhead light component mole fraction prediction function, and the overhead light component mole fraction prediction value is obtained by combining the heat-driven sensitivity coefficient; In each control cycle, the overhead purity constraint function is constructed by subtracting the lower limit of the overhead target product purity from the overhead light component mole fraction prediction value, and the constraint condition that the overhead purity constraint function is not less than zero is taken as the constraint condition for meeting the overhead product purity requirement.

6. The data-driven rectifying energy consumption optimization control method according to claim 5, wherein, The Lagrange multiplier related to the overhead purity constraint is introduced, the optimization objective function and the overhead purity constraint function are combined into a Lagrange energy consumption optimization function, and a dual problem is constructed, specifically including: The Lagrange multiplier related to the overhead purity constraint is introduced, the overall equivalent unit energy consumption index and the overhead purity constraint function are combined through the Lagrange multiplier and the penalty weight parameter, where the penalty weight parameter is a preset non-negative constant, to form a Lagrange energy consumption optimization function in each control cycle; A dual optimization problem is constructed for the Lagrange energy consumption optimization function, and under the premise that the Lagrange multiplier is not less than zero, the inner layer performs a minimum operation on the control variable vector and the outer layer performs a maximum operation on the Lagrange multiplier to obtain the minimum value of the Lagrange energy consumption optimization function under the given penalty weight parameter.

7. The data-driven rectifying energy consumption optimization control method according to claim 6, wherein, The Lagrange multiplier and the control variable vector iteration initial value are set, a multiplier sub-gradient correction mechanism is constructed based on the prediction function and the purity lower limit, and the multiplier is iteratively updated in the sub-gradient direction, specifically including: In each control cycle, the maximum number of inner layer iterations of the multiplier is set to fifty, the initial value of the Lagrange multiplier is set to zero, and the reboiler heat duty and the overhead reflux ratio measured at the starting time of the current control cycle are taken as the iteration starting point of the control variable vector; In each iteration, the overhead light component mole fraction prediction value is calculated through the overhead light component mole fraction prediction function according to the current iteration control variable vector to form the overhead light component mole fraction prediction value of the current iteration; Obtaining a first-order partial derivative of the predicted value of the mole fraction of the overhead light components with respect to the heat load of the reboiler according to the structure of the prediction function of the mole fraction of the overhead light components, and taking the first-order partial derivative as an intermediate quantity for subsequent gradient calculation; Constructing a projection function for taking the non-negative part of any real number, and taking the difference between the lower limit of the target product purity of the column top and the predicted value of the mole fraction of the overhead light components in the current iteration as the input of the projection function to obtain a non-negative purity deviation; Taking the non-negative purity deviation as a Lagrange multiplier increment, updating the Lagrange multiplier value at the end of each iteration according to the rule of adding the multiplier increment to the current Lagrange multiplier value to obtain the initial value of the Lagrange multiplier corresponding to the next iteration.

8. The data-driven rectifying energy consumption optimization control method according to claim 7, wherein, The iteration and update of the control variable vector based on the Lagrange energy consumption optimization function within the constraint range of the heat load of the reboiler and the reflux ratio of the column top, and the convergence criterion is set according to the function and the multiplier change to obtain a converged solution or a solution that reaches a preset maximum number of iterations, which specifically includes: Obtaining the safe lower limit value and the safe upper limit value of the heat load of the reboiler from the reboiler nameplate information, and obtaining the allowed upper limit of the reflux ratio of the column top; In each iteration, performing a descent update on the heat load of the reboiler and the reflux ratio of the column top according to the first-order partial derivative direction of the current Lagrange energy consumption optimization function with respect to the heat load of the reboiler and the reflux ratio of the column top, to obtain the updated control variable vector candidate value; According to the first-order partial derivative value of the Lagrange energy consumption optimization function with respect to the heat load of the reboiler and the reflux ratio of the column top at the initial iteration, combined with the safe lower limit value and the safe upper limit value of the heat load of the reboiler and the allowed upper limit of the reflux ratio of the column top, calculate the step length coefficient for the heat load of the reboiler and the reflux ratio of the column top respectively, so that the step length coefficient is related to the allowed adjustment range and the initial gradient magnitude of each other; Performing a truncation operation on the updated heat load of the reboiler and the reflux ratio of the column top according to the allowed range of each, correcting the part of the heat load of the reboiler less than the safe lower limit value to the safe lower limit value, correcting the part of the heat load of the reboiler greater than the safe upper limit value to the safe upper limit value, correcting the part of the reflux ratio of the column top less than zero to zero, and correcting the part of the reflux ratio of the column top greater than the allowed upper limit to the allowed upper limit; After each iteration, calculate the current Lagrange energy consumption optimization function value according to the updated control variable vector and the Lagrange multiplier, and calculate the difference between the Lagrange energy consumption optimization function value and the Lagrange multiplier value between this iteration and the last iteration; Set the convergence tolerance of the Lagrange energy consumption optimization function value difference and the convergence tolerance of the Lagrange multiplier value difference, wherein the convergence tolerance of the Lagrange energy consumption optimization function value difference is one percent of the safe lower limit value of the heat load of the reboiler, and the convergence tolerance of the Lagrange multiplier value difference is one thousandth, and when the Lagrange energy consumption optimization function value difference and the Lagrange multiplier value difference between two consecutive iterations are both less than the corresponding convergence tolerance, it is determined that the optimization calculation of the current control period converges. When the convergence determination condition is met before the iteration number reaches the preset maximum iteration number, the control variable vector and the Lagrange multiplier corresponding to when the convergence determination condition is met are taken as the iteration result of the current control period; when the convergence determination condition is not met when the iteration number reaches the preset maximum iteration number, the control variable vector and the Lagrange multiplier corresponding to the maximum iteration number are taken as the iteration result of the current control period.

9. The data-driven rectifying energy consumption optimization control method according to claim 8, wherein, The control variable vector at the time of convergence is taken as the optimal control solution of the current control period, and is output to the reboiler heat supply adjustment device and the condensation reflux adjustment device. The working condition parameters, the optimal control variable vector, the tower top light component mole fraction, the tower top target product purity lower limit, the overall equivalent unit energy consumption index, and the Lagrange multiplier are subjected to data packaging and storage. Specifically, the method comprises the following steps: When the multiplier iteration update and the control variable optimization iteration satisfy the convergence determination condition, the convergence iteration number is obtained, and the reboiler heat load and the tower top reflux ratio corresponding to the convergence iteration number are taken as the optimal control solution of the current control period; The optimal reboiler heat load of the current control period is output to the reboiler heat supply adjustment device, and the optimal tower top reflux ratio of the current control period is output to the condensation reflux adjustment device; The distillation feed mass flow, the distillation feed light component mole fraction, the optimal reboiler heat load, the optimal tower top reflux ratio, the tower top light component mole fraction, the tower top target product purity lower limit, the overall equivalent unit energy consumption index, and the converged Lagrange multiplier value of the current control period are written into the data record module, and the working condition parameter data acquisition, function construction, optimization calculation, and control output are continued based on the recorded data and newly collected data in the next control period.

Images