Method and system for dynamic regulation of acetylene hydrochlorination catalytic reaction temperature based on multi-source parameter fusion

Abstract

The application discloses a method and system for dynamically regulating the temperature of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, and relates to the technical field of industrial automation control. The method comprises the following steps: acquiring multi-source parameters of a reactor in real time, inputting the multi-source parameters into a mechanism sub-model and a sparse Gaussian process regression model respectively, and generating a mixed prediction benchmark according to the obtained mechanism prediction value and error compensation value; taking the deviation between the mixed prediction benchmark and the measured hot spot temperature as feedback, estimating the lumped disturbance through an extended state observer; extrapolating the lumped disturbance estimation value at the current time to the entire prediction time domain, directly superimposing the feedforward compensation term in the prediction model of the model predictive control, constructing a prediction equation after disturbance compensation, and solving a quadratic programming problem with constraints to obtain the optimal control sequence of the cooling water valve opening. In this way, high-precision, high-robustness and full-life-cycle optimization control of the reactor temperature is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to a method and system for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion. Background Technology

[0002] The acetylene hydrochlorination reaction is the core step in the production of vinyl chloride monomer (VCM). This reaction is strongly exothermic, and the precision of its hot spot temperature control directly affects the activity, selectivity, and service life of the catalyst. The low-mercury catalyst (mercuric chloride content ≤1%) widely used in the calcium carbide PVC industry is particularly sensitive to temperature. Long-term operation above 145℃ will accelerate mercury sublimation, leading to catalyst deactivation and environmental pollution risks.

[0003] Currently, PID control or predictive control based on single-variable models are widely used in industry. Existing technologies have the following main drawbacks:

[0004] Insufficient model accuracy: Traditional mechanistic models (based on material balance and Arrhenius kinetics) make simplifications and struggle to accurately obtain all kinetic parameters (such as the decay curve of catalyst activity over time). This leads to systematic deviations between model predictions and actual reactor conditions, especially at the end of catalyst life or when feed gas composition fluctuates, resulting in severe model mismatch issues.

[0005] Non-affine and coupled characteristics are difficult to handle: there is strong coupling between reactor temperature, cooling medium flow rate, reactant concentration, and catalyst activity state. Existing control methods have failed to effectively decouple these factors, resulting in lag in control response and a tendency to cause temperature overshoot or oscillation.

[0006] Weak constraint handling capability: Industrial production involves explicit physical constraints (such as cooling water valve opening limits and valve action rate limits) and safety constraints (hot spot temperatures must not exceed the catalyst tolerance limit). While traditional MPC (Model Predictive Control) can handle these constraints, predictions based on nominal models may violate these constraints under disturbances, leading to safety hazards or control failure.

[0007] Lack of adaptability to catalyst aging: As catalyst operating time increases, its activity gradually decreases, and the optimal reaction temperature window also shifts. Existing control strategies have fixed setpoints and cannot adapt to this change, leading to reduced catalyst utilization and increased energy consumption. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion. It aims to solve the problems of insufficient model accuracy, difficulty in coupling processing, easy violation of constraints, and inability to adapt to catalyst aging in the prior art, so as to achieve high-precision, high-robustness and full life cycle optimization control of reactor temperature.

[0009] In a first aspect, embodiments of the present invention provide a method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, comprising:

[0010] The reactor's multi-source parameters are acquired in real time, including inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperatures and flow rates, regulating valve opening, and catalyst cumulative operating time.

[0011] The multi-source parameters are input into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and a hybrid prediction benchmark is generated based on the mechanism prediction value and the error compensation value.

[0012] Using the deviation between the hybrid prediction benchmark and the measured hotspot temperature as feedback, the lumped disturbance is estimated through the extended state observer to obtain the lumped disturbance estimate.

[0013] The lumped disturbance estimate at the current moment is extrapolated to the entire prediction time domain and directly superimposed as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation.

[0014] Based on the prediction equation after disturbance compensation, a constrained quadratic programming problem is solved to obtain the optimal control sequence for the opening degree of the cooling water valve, and the first control variable in the sequence is executed.

[0015] In a preferred embodiment, the multi-source parameters are input into the mechanistic sub-model and a pre-trained sparse Gaussian process regression model respectively to obtain mechanistic prediction values ​​and error compensation values. A hybrid prediction benchmark is then generated based on the mechanistic prediction values ​​and error compensation values, including:

[0016] The multi-source parameters are simultaneously input into the mechanism sub-model and the sparse Gaussian process regression model. The mechanism sub-model outputs the mechanism prediction value, and the sparse Gaussian process regression model outputs the error compensation value.

[0017] The mechanistic prediction value is added to the error compensation value to obtain the hybrid prediction benchmark.

[0018] In a preferred embodiment, the deviation between the hybrid prediction benchmark and the measured hotspot temperature is used as feedback to estimate the lumped disturbance through an extended state observer, resulting in an estimated value of the lumped disturbance, including:

[0019] The difference between the hybrid prediction baseline and the measured hotspot temperature is calculated as the observer feedback error;

[0020] The observer feedback error is input into the extended state observer, which outputs a lumped disturbance estimate; wherein the extended state observer adopts a second-order state equation, and the gain parameter is tuned according to the time constant of the unmodeled dynamics of the system.

[0021] In a preferred embodiment, the extended state observer is discretized using the forward Euler method, and the discretized recursive equations include:

[0022] The difference between the hybrid prediction baseline and the measured hotspot temperature is calculated as the feedback error at the current moment;

[0023] Obtain the observer state and control input from the previous time step;

[0024] The observer state at the current time step is updated using the observer state at the previous time step, the control input at the previous time step, and the feedback error at the current time step, according to the forward Euler recursion.

[0025] The feedback error, multiplied by the observer gain after forward Euler discretization, is used as a correction term in the observer state update.

[0026] In a preferred embodiment, the estimated lumped disturbance value at the current moment is extrapolated to the entire prediction time domain and directly superimposed as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation, including:

[0027] Divide the current lumped disturbance estimate by the control channel gain to obtain the disturbance compensation term;

[0028] The disturbance compensation term is superimposed on the sum of the mechanism prediction value output by the mechanism sub-model and the error compensation value output by the sparse Gaussian process regression model to form the disturbance-compensated prediction equation.

[0029] In a preferred embodiment, based on the prediction equation after disturbance compensation, a constrained quadratic programming problem is solved to obtain the optimal control sequence for the cooling water valve opening, and the first control variable in the sequence is executed, including:

[0030] At each sampling time, based on the prediction equation after disturbance compensation, and with the objective of minimizing the sum of squared deviations between the predicted hotspot temperature and the reference temperature trajectory in the prediction time domain, a control increment penalty term is added to construct a quadratic programming problem.

[0031] The quadratic programming problem is set with hard constraints, including upper and lower limits of cooling water valve opening, limit of single-step change rate of valve opening, and upper limit of hot spot temperature prediction value.

[0032] Solving the quadratic programming problem yields the optimal control sequence, and the first element of the sequence is extracted as the valve opening command output at the current moment.

[0033] In a preferred embodiment, the method further includes:

[0034] Based on the cumulative operating time of the catalyst and the catalyst activity state parameters, the reference temperature trajectory in the model predictive control is periodically and adaptively updated, and a safety upper limit is truncated.

[0035] In a preferred embodiment, the method for obtaining the catalyst activity state parameters specifically includes:

[0036] Dynamic parameters were obtained by identifying the step response experiment;

[0037] Based on the measured acetylene conversion rate, inlet acetylene flow rate, measured hot spot temperature and the aforementioned kinetic parameters, the activity factor is calculated using a simplified kinetic equation.

[0038] The calculated activity factor was filtered using an exponentially weighted moving average and periodically cross-validated with the offline catalyst analysis results.

[0039] In a preferred embodiment, the reference temperature trajectory in the model predictive control is periodically and adaptively updated based on the cumulative operating time of the catalyst and the catalyst activity state parameters, including:

[0040] The first reference temperature is calculated based on the sum of the products of the catalyst's initial optimal reaction temperature, the activity compensation coefficient, and the deviation of the activity factor.

[0041] The second reference temperature is calculated by subtracting the control safety margin from the nominal safety upper limit temperature of the catalyst.

[0042] The smaller value between the first reference temperature and the second reference temperature is selected as the updated reference temperature trajectory.

[0043] The above updates are performed periodically based on the cumulative operating time of the catalyst.

[0044] Secondly, embodiments of the present invention also provide a dynamic temperature control system for the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, comprising:

[0045] The multi-source parameter acquisition unit is used to acquire the multi-source parameters of the reactor in real time. The multi-source parameters include inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperature and flow rate, regulating valve opening degree, and catalyst cumulative running time.

[0046] The hybrid prediction benchmark generation unit is used to input the multi-source parameters into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and generate the hybrid prediction benchmark based on the mechanism prediction value and the error compensation value.

[0047] The disturbance estimation unit is used to estimate the lumped disturbance by using the deviation between the hybrid prediction benchmark and the measured hot spot temperature as feedback, and obtain the lumped disturbance estimate value through the extended state observer;

[0048] The prediction equation construction unit is used to extrapolate the lumped disturbance estimate at the current time to the entire prediction time domain and directly superimpose it as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation.

[0049] The rolling optimization and execution unit is used to solve the constrained quadratic programming problem based on the prediction equation after disturbance compensation, obtain the optimal control sequence of cooling water valve opening, and execute the first control variable in the sequence.

[0050] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0051] One or more processors;

[0052] Storage device for storing one or more programs;

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion as described in any embodiment of the present invention.

[0054] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion as described in any embodiment of the present invention.

[0055] Compared to existing technologies, the following beneficial effects have been achieved:

[0056] (1) Compared with traditional PID control, the root mean square error of the mechanism-data hybrid prediction model adopted in this invention is significantly reduced; at the end of the catalyst life, compared with the model prediction control of the pure mechanism model, the root mean square error of the prediction is further reduced, which verifies the accuracy advantage of the hybrid model throughout the entire life cycle.

[0057] (2) Under the perturbation caused by batch differences in kinetic parameters, the system can still control the hot spot temperature fluctuation within a small range. Under the external disturbance of a step change in the feed gas flow rate, the maximum deviation of the hot spot temperature is limited, and it can quickly recover to near the target value within several sampling cycles, demonstrating excellent anti-disturbance capability.

[0058] (3) By feeding the disturbance estimate of the extended state observer forward into the prediction model instead of retroactively correcting the execution instructions, this invention ensures that the control instructions solved in the presence of disturbances naturally satisfy the constraint conditions. In a large number of random disturbance simulation tests, no constraint violations occurred due to disturbance feedforward compensation, verifying the constraint reliability of this architecture.

[0059] (4) By adaptively adjusting the reference temperature trajectory and cutting off the safety upper limit, the present invention can maintain a high acetylene conversion rate in the later stage of catalyst operation, and the effective service life of the catalyst is significantly extended compared with the fixed set value control, thus realizing the optimized utilization of the entire life cycle of the catalyst. Attached Figure Description

[0060] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0061] Figure 1 This is a flowchart of a method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion provided in an embodiment of the present invention;

[0062] Figure 2 This is an overall architecture diagram of the acetylene hydrochlorination reaction temperature dynamic control system provided in this embodiment of the invention;

[0063] Figure 3 This is a schematic diagram of the structure and information flow of the mechanism-data hybrid prediction model provided in the embodiments of the present invention;

[0064] Figure 4 This is an internal optimization and feedback logic diagram of the hierarchical MPC controller provided in this embodiment of the invention;

[0065] Figure 5This is a simplified schematic diagram of an industrial acetylene hydrochlorination reactor provided in an embodiment of the present invention;

[0066] Figure 6 This is a schematic diagram of a dynamic temperature control system for acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion provided in an embodiment of the present invention;

[0067] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0068] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0069] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0070] Example 1

[0071] To address the problems of insufficient model accuracy, difficulty in coupling processing, easy violation of constraints, and inability to adapt to catalyst aging in existing technologies, this invention proposes a dynamic temperature control method for the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion. The core of this method is as follows: First, a hybrid prediction model integrating a fusion mechanism model and data-driven compensation terms is constructed. The data-driven part utilizes the slow time-varying systematic bias of the sparse Gaussian process regression learning mechanism model. Second, a hierarchical control strategy is adopted. The upper layer performs multi-step prediction and rolling optimization based on the hybrid model and online catalyst activity identification. The lower layer uses an extended state observer to estimate unmodeled high-frequency disturbances in real time, and these disturbances are superimposed into the MPC prediction model through a disturbance feedforward compensation term to strictly ensure constraint satisfaction. Finally, structured parameter optimization ensures the real-time performance and stability of the controller.

[0072] like Figure 1 As shown, Embodiment 1 of the present invention provides a flowchart of a method 100 for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion. The method 100 specifically includes the following steps:

[0073] Step S110: Real-time acquisition of multiple source parameters of the reactor, including inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperatures and flow rates, regulating valve opening degree, and catalyst cumulative operating time.

[0074] In a preferred embodiment, the multi-source parameters include:

[0075] Reaction inlet parameters: acetylene flow rate, hydrogen chloride flow rate, feed gas temperature;

[0076] Reactor status parameters: multiple bed temperatures distributed along the reactor axis (with a focus on monitoring the measured hot spot temperature), and reactor inlet and outlet pressure difference.

[0077] Cooling system parameters: inlet temperature of cooling water, outlet temperature, current flow rate, and regulating valve opening.

[0078] Catalyst state parameters: cumulative operating time, and catalyst activity state parameters obtained online or offline, such as activity decay factor.

[0079] As a preferred embodiment, the method for obtaining the catalyst active factor is as follows:

[0080] Based on reactor inlet acetylene flow rate and measured export conversion rate The activity factor was calculated by back-calculating the simplified kinetic equation. :

[0081] , ;

[0082] in: Pre-exponential factor Activation energy; The low-energy catalyst used in this device has its kinetic parameters identified through step response experiments during the initial operation of the device, and the specific values ​​are as follows: , If catalysts from other suppliers are used, they need to be re-identified. Catalyst bed volume; the loading volume of a single reactor in this unit is [missing information]. ; Measured hot spot temperature (median of axial temperature measurement points); Ideal gas constant; Space and time.

[0083] The activity factor is calculated once at a preset period, such as every 6 hours. Online gas chromatography (GC) completes an analysis every 15 minutes, and the median of all analytical results within the 6-hour window is used as the input for conversion calculation to eliminate the influence of single abnormal readings. If online GC temporarily fails, a material balance-based estimation method can be switched to: the calculated activity factor is processed using an exponentially weighted moving average filter and periodically cross-validated with offline catalyst analysis results (such as mercury content analysis) to ensure the reliability of the activity factor estimate.

[0084] Step S120: Input the multi-source parameters into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and generate a hybrid prediction benchmark based on the mechanism prediction value and the error compensation value.

[0085] As a preferred embodiment, the above-mentioned mechanistic sub-model is constructed based on the one-dimensional pseudo-homogeneous heat balance equation and Langmuir-Hinshelwood kinetics of the acetylene hydrochlorination reaction, and is used to describe the heat generation and removal process within the reactor. This model outputs a predicted nominal hotspot temperature. .

[0086] The data-driven compensation sub-model described above employs Sparse Gaussian Process Regression (Sparse GPR), and its specific implementation process is as follows:

[0087] Training set construction: Approximately 5000 representative points were uniformly sampled over time from the catalyst's entire life cycle operation data to form the training set. ,in The input feature vector includes current acetylene flow rate, hydrogen chloride flow rate, hotspot temperature, and cumulative running time. For the prediction error of the mechanism model , For the first The actual hotspot temperature measurement value corresponding to each training sample. For the first The mechanistic model outputs the temperature corresponding to each training sample.

[0088] Sparse Approximation: Employing the Sparse Variational Gaussian Process (SVGP) framework, 200 inducing points are used to approximate the complete kernel matrix, reducing the prediction complexity to... train / Prediction, among which much smaller .

[0089] Kernel function and training: The squared exponential kernel is selected as the kernel function, and the hyperparameters are optimized by the maximum likelihood method. Training is completed offline, and the kernel matrix decomposition results are pre-computed and stored.

[0090] Online prediction: When running online, input the current state. Output error compensation value (posterior mean) The error compensation value is equal to the posterior mean. , Let be the mean of the posterior distribution of the Gaussian process regression.

[0091] As a preferred embodiment, combined with Figure 3 As shown, the multi-source parameters are simultaneously input into the mechanistic sub-model and the sparse Gaussian process regression model. The mechanistic sub-model outputs the mechanistic prediction value, and the sparse Gaussian process regression model outputs the error compensation value. The mechanistic prediction value and the error compensation value are added together to obtain the hybrid prediction benchmark.

[0092] Step S130: Using the deviation between the hybrid prediction benchmark and the measured hot spot temperature as feedback, the lumped disturbance is estimated through the extended state observer to obtain the lumped disturbance estimate.

[0093] In a preferred embodiment, the above extended state observer uses a second-order extended state observer (ESO) to define the system state:

[0094] First state variable: Hybrid prediction baseline temperature ;

[0095] Second state variable: lumped disturbance This includes unmodeled dynamics and external disturbances;

[0096] Control input Cooling water valve opening;

[0097] Control channel gain It is obtained through open-loop step response identification. In this example, we take... ;

[0098] Given nonlinear terms A state-dependent function determined by the mechanistic model;

[0099] Equations of state:

[0100] ;

[0101] in, The state variables of the Extended State Observer (ESO) are defined as the mixed predictive reference temperature. ; The state variables of an ESO are defined as lumped disturbances. ; : Represents a state variable The first derivative with respect to time, i.e., the rate of change; : Represents a state variable The first derivative of the lumped disturbance with respect to time, i.e., the rate of change of the disturbance; : Rate of change of disturbance.

[0102] Continuous-time observer equations ( (Actual measured value)

[0103] ;

[0104] Among them, the observer gain For the observer bandwidth, the tuning method is as follows: , The time constant for the unmodeled dynamics is obtained by fitting higher-order terms to the open-loop step response. In this embodiment, it is approximately 18 seconds. ; These are actual measured values. ; For observer error, ;

[0105] For ESO state The estimated value; For ESO state The estimated value; for The estimated value The first derivative with respect to time, i.e., the rate of change of the estimated value; for The estimated value The first derivative with respect to time, i.e., the rate of change of the estimated value.

[0106] The discretization of the extended state observer is achieved by discretizing the continuous-time observer using the forward Euler method, with a sampling period of... Seconds, continuous gain The discretized recurrence equation is:

[0107] ;

[0108] in, , , ESO states At discrete time The estimated value; ESO states At the previous discrete time The estimated value; ESO states At the previous discrete time The estimated value; : Control input (cooling water valve opening) at the previous discrete time step The value; Given a nonlinear term, the state-dependent function is determined by the mechanistic model; Observer error ; ESO states At discrete time The estimated value; : The observer gain coefficient after discretization ; : The observer gain coefficient after discretization .

[0109] It should be noted that forward Euler transformation may introduce numerical instability; in practical engineering implementations, bilinear transformation (Tustin) or zero-order preservation can be used instead. Discretization requires the corresponding gain to be recalculated according to the transformation formula, which can be selected by those skilled in the art based on digital control theory.

[0110] Step S140: Extrapolate the lumped disturbance estimate at the current moment to the entire prediction time domain, and directly superimpose it as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation.

[0111] As a preferred embodiment, combined with Figure 4 As shown, the prediction model employs hierarchical model predictive control (MPC), where the upper layer is a hybrid model predictive control path planner.

[0112] The prediction model uses the assumption of constant perturbation extrapolation:

[0113] ;

[0114] in, The predicted hotspot temperature output by the prediction model, at discrete time intervals. ; The nominal hotspot temperature prediction value output by the mechanistic model at discrete time intervals. ; Sparse Gaussian process regression (GPR) models based on input features The error compensation value is as follows.

[0115] This means that the lumped disturbance estimate remains unchanged in the prediction time domain. Therefore, the prediction equation after disturbance compensation is constructed as follows: the hot spot temperature prediction value at each moment in the prediction time domain is equal to the output value of the mechanism sub-model at the corresponding moment plus the error compensation value output by the sparse Gaussian process regression model, plus the disturbance compensation term obtained by dividing the lumped disturbance estimate at the current moment by the control channel gain.

[0116] Step S150: Based on the prediction equation after disturbance compensation, solve the constrained quadratic programming problem to obtain the optimal control sequence of cooling water valve opening, and execute the first control variable in the sequence.

[0117] In a preferred embodiment, based on the disturbance-compensated prediction equation, in each control cycle Solve the following quadratic programming problem in seconds:

[0118] ;

[0119] in, The number of prediction time-domain steps in Model Predictive Control (MPC); : The step index in the prediction time domain, with a range of values. ; : Control input increment (sequence), i.e., the change in valve opening; : Reference temperature trajectory at discrete times The value (generated by the catalyst adaptive module); The square of the weighted Euclidean norm, with the weight matrix being... ; Output (temperature error) weight matrix, in this embodiment the numerical values ; : Control the incremental weight matrix; in this embodiment, the values ​​are... ; The number of time-domain control steps for MPC, in this embodiment .

[0120] Constraint handling: All constraints are treated as hard constraints and handled in the quadratic programming solver using inequality constraints.

[0121] ;

[0122] in, : Control input (cooling water valve opening) at discrete time The value; : Control input increment at discrete time The value, ; : Lower limit of valve opening, in this embodiment ; : Upper limit of valve opening, in this embodiment ; The upper limit of the valve's single-step rate of change (hard constraint) is shown in this embodiment. Each step; Hotspot temperature upper limit (hard constraint), in this embodiment ;

[0123] In a preferred embodiment, the lower layer of hierarchical model predictive control (MPC) is disturbance feedforward compensation, and the disturbance estimated by ESO is... Without directly modifying the solution obtained by MPC Instead, the data is fed into the MPC prediction model via a superposition of feedforward terms. Since the prediction model already contains perturbation information, the MPC solution... It naturally satisfies all constraints, requiring no subsequent modifications. The optimal control sequence (first value) obtained by MPC (For execution).

[0124] As a preferred embodiment, combined with Figure 2 As shown, the above method also includes calculating based on the cumulative operating time of the catalyst. and active factors obtained online or offline. The reference temperature trajectory in model predictive control is dynamically adjusted, and a safety upper limit cutoff is set.

[0125] ;

[0126] in: (Optimal temperature for the new catalyst); / Unit activity factor, its tuning is based on the activity-temperature compensation curve provided by the catalyst supplier: when the activity factor decreases by 0.1, the reaction temperature needs to be increased by approximately to maintain the same conversion rate. ; (Nominal safety limit for catalysts); (Controlling safety margin) Therefore, the upper limit of the reference temperature is limited to 140℃. Even if the activity factor approaches 0, it will not exceed the safety limit, ensuring that the MPC will not remain in a constrained activation state. The update frequency is once every 24 hours.

[0127] According to the above embodiments, the present invention achieves the following beneficial effects:

[0128] (1) Compared with traditional PID control, the root mean square error of the mechanism-data hybrid prediction model adopted in this invention is significantly reduced; at the end of the catalyst life, compared with the model prediction control of the pure mechanism model, the root mean square error of the prediction is further reduced, which verifies the accuracy advantage of the hybrid model throughout the entire life cycle.

[0129] (2) Under the perturbation caused by batch differences in kinetic parameters, the system can still control the hot spot temperature fluctuation within a small range. Under the external disturbance of a step change in the feed gas flow rate, the maximum deviation of the hot spot temperature is limited, and it can quickly recover to near the target value within several sampling cycles, demonstrating excellent anti-disturbance capability.

[0130] (3) By feeding the disturbance estimate of the extended state observer forward into the prediction model instead of retroactively correcting the execution instructions, this invention ensures that the control instructions solved in the presence of disturbances naturally satisfy the constraint conditions. In a large number of random disturbance simulation tests, no constraint violations occurred due to disturbance feedforward compensation, verifying the constraint reliability of this architecture.

[0131] (4) By adaptively adjusting the reference temperature trajectory and cutting off the safety upper limit, the present invention can maintain a high acetylene conversion rate in the later stage of catalyst operation, and the effective service life of the catalyst is significantly extended compared with the fixed set value control, thus realizing the optimized utilization of the entire life cycle of the catalyst.

[0132] Example 2

[0133] Combination Figure 5 As shown, this embodiment uses a single acetylene hydrochlorination reactor as an example for illustration. The temperature control system of the present invention can be independently deployed in each reactor, and the control between each reactor is independent of each other. The following is a detailed description using a single reactor as an example.

[0134] 1.1 Reaction Apparatus and Reactor Parameters: Tubular fixed-bed reactor, inner diameter 2.5m, tube length 4.5m, catalyst bed volume... The tubes are filled with a low-mercury catalyst, and heat is transferred between the tubes via circulating cooling water. The recommended operating temperature window is 125-140℃, with a safe upper limit of 145℃.

[0135] 1.2 Hardware Deployment: A mass flow meter and a thermometer are installed at the inlet; three K-type thermocouples are installed at each of the five different height sections along the reactor axis, for a total of 15 temperature measurement points; an electromagnetic flow meter and an electric regulating valve are installed on the cooling water return pipeline; all signals are connected to the DCS system, and an industrial control computer is configured to run the MPC algorithm.

[0136] 1.3 Sparse GPR Training:

[0137] Data acquisition: From the previous batch of catalysts' 8000 hours of operation data, 4000 sample points were obtained by sampling every 2 hours. 3500 samples were randomly selected as the training set and 500 samples as the validation set.

[0138] Sparse approximation: The SVGP framework is used with 200 induced points. Training is performed using the TensorFlow Probability library, with a maximum of 500 iterations. After convergence, the induced point positions and kernel hyperparameters are fixed.

[0139] Model validation: On the validation set, the root mean square error of the prediction was 0.42°C, and the mean absolute percentage error was 3.1%. The average time for a single prediction on the target hardware was measured to be 7.8 milliseconds.

[0140] 1.4 ESO Parameter Tuning and Implementation:

[0141] Open-loop step response testing was conducted, with the cooling water valve stepping from 40% to 50%. The hot spot temperature response curves were recorded, and the system's dominant time constant was fitted to approximately 150 seconds. The time constant for identifying the unmodeled dynamics was approximately 18 seconds; therefore, it was set... .

[0142] Continuous gain: , .

[0143] Forward Euler discretization is used, with a sampling period of seconds, the recurrence relation is:

[0144] ;

[0145] Right now , Note: When When the size is large, as in this embodiment For a given second, forward Euler may introduce numerical instability. In practical engineering implementation, bilinear transformation (Tustin) or zero-order hold (ZOH) discretization can be used. The corresponding gain needs to be recalculated according to the transformation formula. Those skilled in the art can adjust it themselves according to digital control theory.

[0146] 1.5 MPC Structured Parameter Configuration:

[0147] Sampling time Seconds; prediction time domain Step; control time domain Step; weight Q=10 (temperature error), R=1 (control increment);

[0148] Constraints: Valve opening single-step rate of change Hotspot temperature limit (Hard constraints).

[0149] 1.6 Online Operation Process:

[0150] Step S1 (Initialization): Before the system is put into use, the training of the sparse GPR model is completed in advance, including: collecting historical operation data of the catalyst, using the difference between the predicted value and the measured value of the mechanism model as the training label, determining the hyperparameters of the kernel function and the location of the induction point, and pre-calculating and storing the kernel matrix decomposition results. This initialization step needs to be re-executed after the catalyst batch is replaced.

[0151] Step S2 (online, executed every 15 seconds): The DCS reads real-time data such as 15 temperature measurement points, acetylene flow rate, hydrogen chloride flow rate, cooling water inlet and outlet temperatures, valve opening degree, and cumulative running time.

[0152] Step S3: Simultaneously input the multi-source parameters into the mechanistic sub-model and the sparse Gaussian process regression model. The mechanistic sub-model outputs the mechanistic prediction value, and the sparse Gaussian process regression model outputs the error compensation value. Add the mechanistic prediction value and the error compensation value to obtain the hybrid prediction benchmark.

[0153] Step S4: ESO according to Update internal status Output lumped disturbance estimate It is important to note that here... Not included This ensures the observer's independence.

[0154] Step S5: Using the perturbation constant assumption, Extrapolating to the entire prediction time domain, the MPC prediction equation is constructed:

[0155] ;

[0156] Step S6: The upper-level MPC solves the quadratic programming problem to obtain the optimal control sequence. Take the first value. The current command is output to the DCS to drive the electric regulating valve.

[0157] Step S7 (every 24 hours): Based on the cumulative running time and the filtered activity factor (Obtained via online GC median method or offline analysis), updated reference temperature: (Unit: °C) is used for the target trajectory in step S6.

[0158] 1.7 Simulation Verification Results:

[0159] Before the system described in this invention was put into use, the performance of three control strategies was compared using a Matlab / Simulink and industrial DCS co-simulation platform and real historical disturbance data playback. The test results of the embodiment are shown in Table 1 below:

[0160] Table 1

[0161]

[0162] In a demonstration application involving continuous operation for 6 months (approximately 4320 hours) on a single reactor of a 30,000-ton / year PVC plant, the following engineering statistics were obtained compared to the same period of the previous year (using fixed-parameter PID control):

[0163] The average conversion rate of acetylene has been significantly improved (statistics within a typical production month).

[0164] The number of unplanned shutdowns due to overheating has decreased significantly;

[0165] The catalyst replacement cycle has been extended;

[0166] The energy consumption of the cooling water circulation pump has been reduced (due to smoother valve operation).

[0167] Example 3

[0168] Figure 6 This is a schematic diagram of a dynamic temperature control system 600 for the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, as provided in Embodiment 3 of the present invention. Figure 6 As shown, the system includes:

[0169] The multi-source parameter acquisition unit 610 is used to acquire multi-source parameters of the reactor in real time. The multi-source parameters include inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperature and flow rate, regulating valve opening degree, and catalyst cumulative running time.

[0170] The hybrid prediction benchmark generation unit 620 is used to input the multi-source parameters into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and generate the hybrid prediction benchmark based on the mechanism prediction value and the error compensation value.

[0171] The disturbance estimation unit 630 is used to estimate the lumped disturbance by using the deviation between the hybrid prediction benchmark and the measured hot spot temperature as feedback, and obtain the lumped disturbance estimate value through the extended state observer;

[0172] The prediction equation construction unit 640 is used to extrapolate the lumped disturbance estimate at the current time to the entire prediction time domain and directly superimpose it as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation.

[0173] The rolling optimization and execution unit 650 is used to solve the constrained quadratic programming problem based on the prediction equation after disturbance compensation, obtain the optimal control sequence of cooling water valve opening, and execute the first control variable in the sequence.

[0174] The acetylene hydrochlorination catalytic reaction temperature dynamic control system provided in this embodiment of the invention can execute the acetylene hydrochlorination catalytic reaction temperature dynamic control method based on multi-source parameter fusion provided in any of the embodiments of the invention above. It has the corresponding functions and beneficial effects of executing the acetylene hydrochlorination catalytic reaction temperature dynamic control method based on multi-source parameter fusion. For detailed process, please refer to the relevant operation of the acetylene hydrochlorination catalytic reaction temperature dynamic control method based on multi-source parameter fusion in the foregoing embodiments.

[0175] Example 4

[0176] Figure 7 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0177] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0178] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0179] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 executes the method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion described above.

[0180] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0181] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, characterized in that, Includes the following steps: The reactor's multi-source parameters are acquired in real time, including inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperatures and flow rates, regulating valve opening, and catalyst cumulative operating time. The multi-source parameters are input into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and a hybrid prediction benchmark is generated based on the mechanism prediction value and the error compensation value. Using the deviation between the hybrid prediction benchmark and the measured hotspot temperature as feedback, the lumped disturbance is estimated through the extended state observer to obtain the lumped disturbance estimate. The lumped disturbance estimate at the current moment is extrapolated to the entire prediction time domain and directly superimposed as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation. Based on the prediction equation after disturbance compensation, a constrained quadratic programming problem is solved to obtain the optimal control sequence for the opening of the cooling water valve, and the first control variable in the sequence is executed.

2. The method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 1, characterized in that, The multi-source parameters are input into the mechanistic sub-model and the pre-trained sparse Gaussian process regression model, respectively, to obtain mechanistic prediction values ​​and error compensation values. A hybrid prediction benchmark is then generated based on the mechanistic prediction values ​​and error compensation values, including: The multi-source parameters are simultaneously input into the mechanism sub-model and the sparse Gaussian process regression model. The mechanism sub-model outputs the mechanism prediction value, and the sparse Gaussian process regression model outputs the error compensation value. The mechanistic prediction value is added to the error compensation value to obtain the hybrid prediction benchmark.

3. The method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 1, characterized in that, Using the deviation between the hybrid prediction benchmark and the measured hotspot temperature as feedback, the lumped disturbance is estimated through an extended state observer to obtain an estimated value of the lumped disturbance, including: The difference between the hybrid prediction baseline and the measured hotspot temperature is calculated as the observer feedback error; The observer feedback error is input into the extended state observer, which outputs a lumped disturbance estimate; wherein the extended state observer adopts a second-order state equation, and the gain parameter is tuned according to the time constant of the unmodeled dynamics of the system.

4. The method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 3, characterized in that, The extended state observer is discretized using the forward Euler method, and the discretized recursive equations include: The difference between the hybrid prediction baseline and the measured hotspot temperature is calculated as the feedback error at the current moment; Obtain the observer state and control input from the previous time step; The observer state at the current time step is updated using the observer state at the previous time step, the control input at the previous time step, and the feedback error at the current time step, according to the forward Euler recursive relation; wherein, the feedback error is multiplied by the observer gain after forward Euler discretization and used as a correction term in the observer state update.

5. The method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 1, characterized in that, The estimated lumped disturbance value at the current moment is extrapolated to the entire prediction time domain and directly superimposed as a feedforward compensation term into the prediction model of model predictive control to construct the prediction equation after disturbance compensation, including: Divide the current lumped disturbance estimate by the control channel gain to obtain the disturbance compensation term; The disturbance compensation term is superimposed on the sum of the mechanism prediction value output by the mechanism sub-model and the error compensation value output by the sparse Gaussian process regression model to form the disturbance-compensated prediction equation.

6. The method for dynamic temperature control of the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 5, characterized in that, Based on the prediction equation after disturbance compensation, a constrained quadratic programming problem is solved to obtain the optimal control sequence for the cooling water valve opening. The first control variable in this sequence is then executed, including: At each sampling time, based on the prediction equation after disturbance compensation, and with the objective of minimizing the sum of squared deviations between the predicted hotspot temperature and the reference temperature trajectory in the prediction time domain, a control increment penalty term is added to construct a quadratic programming problem. The quadratic programming problem is set with hard constraints, including upper and lower limits of cooling water valve opening, limit of single-step change rate of valve opening, and upper limit of hot spot temperature prediction value. Solving the quadratic programming problem yields the optimal control sequence, and the first element of the sequence is extracted as the valve opening command output at the current moment.

7. The method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 1, characterized in that, The method further includes: Based on the cumulative operating time of the catalyst and the catalyst activity state parameters, the reference temperature trajectory in the model predictive control is periodically and adaptively updated, and a safety upper limit is truncated.

8. The method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 7, characterized in that, The methods for obtaining the catalyst activity state parameters specifically include: Dynamic parameters were obtained by identifying the step response experiment; Based on the measured acetylene conversion rate, inlet acetylene flow rate, measured hot spot temperature and the aforementioned kinetic parameters, the activity factor is calculated using a simplified kinetic equation. The calculated activity factor was filtered using an exponentially weighted moving average and periodically cross-validated with the offline catalyst analysis results.

9. The method for dynamic temperature control of acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion according to claim 8, characterized in that, Based on the cumulative operating time of the catalyst and the catalyst activity state parameters, the reference temperature trajectory in the model predictive control is periodically and adaptively updated, including: The first reference temperature is calculated based on the sum of the products of the catalyst's initial optimal reaction temperature, the activity compensation coefficient, and the deviation of the activity factor. The second reference temperature is calculated by subtracting the control safety margin from the nominal safety upper limit temperature of the catalyst. The smaller value between the first reference temperature and the second reference temperature is selected as the updated reference temperature trajectory. The above updates are performed periodically based on the cumulative operating time of the catalyst.

10. A dynamic temperature control system for the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion, characterized in that, The system is used to perform the dynamic temperature control method for the acetylene hydrochlorination catalytic reaction based on multi-source parameter fusion as described in any one of claims 1 to 9, the system comprising: The multi-source parameter acquisition unit is used to acquire the multi-source parameters of the reactor in real time. The multi-source parameters include inlet acetylene flow rate, hydrogen chloride flow rate, feed gas temperature, measured hot spot temperature, cooling water inlet and outlet temperature and flow rate, regulating valve opening degree, and catalyst cumulative running time. The hybrid prediction benchmark generation unit is used to input the multi-source parameters into the mechanism sub-model and the pre-trained sparse Gaussian process regression model respectively to obtain the mechanism prediction value and the error compensation value, and generate the hybrid prediction benchmark based on the mechanism prediction value and the error compensation value. The disturbance estimation unit is used to estimate the lumped disturbance by using the deviation between the hybrid prediction benchmark and the measured hot spot temperature as feedback, and obtain the lumped disturbance estimate value through the extended state observer; The prediction equation construction unit is used to extrapolate the lumped disturbance estimate at the current time to the entire prediction time domain and directly superimpose it as a feedforward compensation term into the prediction model of the model predictive control to construct the prediction equation after disturbance compensation. The rolling optimization and execution unit is used to solve the constrained quadratic programming problem based on the prediction equation after disturbance compensation, obtain the optimal control sequence of cooling water valve opening, and execute the first control variable in the sequence.

Images