Predictive control method for wet flue gas desulfurization system considering uncertainty compensation
By combining Gaussian process regression model and residual model, a predictive control method for wet desulfurization system considering uncertainty was constructed, which solved the problems of model uncertainty and nonlinear effects, and achieved more efficient control and energy saving effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-05
- Publication Date
- 2026-03-24
AI Technical Summary
In wet desulfurization systems of thermal power plants that operate flexibly under wide loads, traditional control methods struggle to cope with the nonlinear effects caused by model uncertainties and load changes, resulting in poor control performance. Furthermore, existing MPC methods lack effective uncertainty compensation mechanisms during nonlinear processes.
A discrete-time model of the wet desulfurization system is constructed using a Gaussian process regression model. Unmodeled dynamics and process noise are predicted through the residual model. Uncertainty propagation and constraints are set in the prediction time domain, and a predictive controller considering uncertainty is constructed to compensate for time delay and unmeasurable disturbances.
It enables the prediction and assessment of uncertainties in wet desulfurization systems, improves control efficiency, provides flexible control, reduces the need for the system to maintain a high power level for extended periods, and has energy-saving effects.
Smart Images
Figure CN116679556B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of model predictive control technology, and in particular to a predictive control method for a wet desulfurization system that considers uncertainty compensation. Background Technology
[0002] As a major source of sulfur dioxide emissions, thermal power plants are subject to strict regulations under the "Emission Standard for Air Pollutants from Thermal Power Plants." In actual operation, to absorb a high proportion of renewable energy generation, coal-fired units currently generally adopt a wide-load, flexible operation mode, with varying coal types, leading to complex operation of flue gas desulfurization systems. Under such wide-load operation, current control methods are mainly based on dynamic characteristic modeling under typical loads, which introduces uncertainties at the model level. Furthermore, when the load changes significantly, the large time delay and nonlinearity of the desulfurization object itself cause uncertainties to have a stronger impact on control. In this situation, traditional PID control faces parameter tuning difficulties, and some advanced control strategies fail to achieve the desired results.
[0003] To meet the increasingly stringent control requirements in the industrial sector, Model Predictive Control (MPC) has been widely adopted in industrial process control due to its powerful ability to handle complex constraints. It is well known that most industrial process control systems are nonlinear systems, but currently, the application of MPC is largely limited to linear or quasi-linear processes. This is mainly because accurate models of nonlinear processes are difficult to obtain, and efficient algorithms are lacking for optimizing nonlinear problems. Consequently, MPC cannot support the predictive control needs of various advanced industries for nonlinear processes. The control community and industry have long recognized the importance of Nonlinear Model Predictive Control (NMPC), but even now, although NMPC has become a research hotspot in academia, its application in industrial practice is still in its early stages.
[0004] For NMPC, methods such as multi-model, robust design, disturbance-resistant design, and artificial intelligence have been incorporated into the predictive control framework to achieve better control results. However, in the aforementioned studies, most improvements focus on compensating for model uncertainties in the rolling time domain, paying insufficient attention to the control safety issues caused by inaccurate modeling in a single predictive time domain, and rarely mentioning the uncertainties introduced by random disturbances. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a predictive control method for wet desulfurization systems that considers uncertainty compensation. This method achieves predictive control of time-delay processes that consider uncertainty compensation and has good uncertainty prediction and evaluation capabilities.
[0006] The technical solution adopted in this invention is as follows:
[0007] This application provides a predictive control method for a wet desulfurization system that considers uncertainty compensation, including:
[0008] Establish a discrete-time model for the wet desulfurization system:
[0009]
[0010] In the formula, , , Represent The concentration of sulfur dioxide at the outlet of the wet desulfurization system, the frequency of the desulfurization tower circulating pump, and the process noise at any given time; It is the nominal model of a wet desulfurization system. These are unmodeled dynamic characteristics used to characterize uncertainty. It is Transform into and Transformation matrix of results with the same dimension and All are differentiable functions, process noise It follows a Gaussian distribution and is spatially uncorrelated;
[0011] Using residual models For the unmodeled dynamics and the process noise Predictions are made to construct a prediction model for wet desulfurization processes:
[0012]
[0013] In the formula, the residual model This is a Gaussian process regression model constructed based on Gaussian process theory. Representative at The predicted value at any given time;
[0014] Based on the prediction model for wet desulfurization, the uncertainty propagation in the prediction time domain is improved: a prediction time domain longer than the desulfurization process delay is set, and the uncertainty propagation of the wet desulfurization process is carried out in the delay stage and the stage after the delay in the prediction time domain, respectively, and the corresponding uncertainty propagation formula is obtained.
[0015] According to the desulfurization control requirements, a set of constraints is set for the outlet sulfur dioxide concentration and the frequency of the circulating pump. Based on the set of constraints, a probabilistic constraint for the desulfurization control process is constructed.
[0016] A predictive controller is constructed based on the prediction model, the uncertainty propagation formula, and the probability constraints, and control is implemented based on the predictive controller.
[0017] Further technical features are as follows:
[0018] The method for constructing the nominal model includes:
[0019] Based on on-site dynamic characteristic experiments, a transfer function model of the circulating pump frequency on the sulfur dioxide concentration at the outlet of the wet desulfurization system was established. Based on the transfer function model, a discrete state-space model was established as the nominal model as follows:
[0020]
[0021] In the formula, This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump. For a moment, For the time delay of the process, , The coefficient matrices are as follows:
[0022]
[0023] in, , This refers to the sampling time of the on-site dynamic characteristic experiment. , These are the gain and time constant of the transmission process, respectively.
[0024] The residual model The construction methods include:
[0025] definition Input data for the residual model:
[0026] Based on the nominal model, the residuals are obtained. , This is the actual sulfur dioxide concentration value at the export site. This is the outlet sulfur dioxide concentration value output by the nominal model. yes The false reversal;
[0027] Based on Gaussian process theory, a kernel function is used to calculate the vector between training input data and test input data after kernel function computation. The matrix calculated by the kernel function between the training data Test data and its own constant calculated by the kernel function Construct a Gaussian process regression model And this regression model is used as a residual model. ,in, These are the prior variance of the training dataset, the identity matrix, and the data vector of the training output, respectively.
[0028] In the time delay phase of the prediction time domain, the uncertainty of the wet desulfurization process is propagated, and the propagation formula of the uncertainty of the wet desulfurization process in the time delay phase is obtained, including:
[0029] Configure a separate control law with state feedback. , It is feedback gain. This is the quantity to be optimized. This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump;
[0030] Based on the prediction model for wet desulfurization, the prediction output representation for the time delay stage is given:
[0031]
[0032] superscript represent The prediction time domain of the first time step, Represents the concentration of sulfur dioxide in exports exist Predicted output at time step Representing the known, in The actual circulating pump frequency value at any given time; For a moment, For the time delay of the process, To calculate the number of steps;
[0033] Constructing the export sulfur dioxide concentration during the time delay phase in the prediction time domain With residual Joint probability distribution:
[0034]
[0035] In the formula, , These are the sulfur dioxide concentrations at the export sites. residual mean , These are the sulfur dioxide concentrations at the export sites. and residuals The covariance of itself It is the concentration of sulfur dioxide at the export site. With residual Covariance between; subscript delay Represents the delay phase. , Represents the mean and covariance of the joint probability distribution; superscript Represents the time delay phase time;
[0036] Based on the joint probability distribution in the above equation, the time-delay propagation formula for uncertainty in the wet desulfurization process is obtained:
[0037]
[0038] In the prediction time domain, the uncertainty of the wet desulfurization process is propagated after the time delay, and the propagation formula of the uncertainty of the wet desulfurization process after the time delay is obtained, including:
[0039] Based on the prediction model of wet desulfurization, a random variable after the time delay stage in the prediction time domain is constructed. With residual Joint probability distribution:
[0040]
[0041] In the formula, Represents random variables , Represents random variables The mean, Represents random variables The covariance of itself Represents random variables The covariance between the residual d and the residual d; , These are the mean and covariance of the joint probability distribution, indicated by the superscript. Represents the period after the time delay. time, For prediction in the time domain; where:
[0042]
[0043] Based on the joint probability distribution in the above equation, the propagation formula for the time-delayed stage of uncertainty in the wet desulfurization process is obtained:
[0044] .
[0045] Set the constraint set for the outlet sulfur dioxide concentration. The set of constraints for the frequency of the circulating pump They are respectively:
[0046]
[0047] In the formula, , and , These are the corresponding coefficient matrices for the outlet sulfur dioxide concentration constraint and the circulating pump frequency constraint, respectively.
[0048] Based on constraint sets and Based on the reachable set, construct probabilistic constraints for the desulfurization control process:
[0049]
[0050] In the formula, The quantile function representing the normal distribution. and These represent the significance levels of the violation of constraints by the outlet sulfur dioxide concentration and the circulation pump frequency, respectively.
[0051] A predictive controller is constructed using the aforementioned predictive model, uncertainty propagation formula, and probability constraints, including:
[0052] Using trace operation, the mean value of the control objective function is predicted for models containing random variables;
[0053] The predictive controller is established, and its mathematical expression is as follows:
[0054]
[0055] In the formula, the objective function In , , The three terms represent the prediction time domain endpoint, the prediction time domain, and the prediction time domain, respectively. The objective function calculation values of the prediction results in the internal delay stage and the stage after the delay has passed; It is a norm 2. It is the trace operator; , These are weight matrices set based on the control requirements for outlet sulfur dioxide concentration and the requirements for the frequency of circulating pumps in the wet desulfurization control task. , These are the nominal model coefficient matrices.
[0056] The beneficial effects of this invention are as follows:
[0057] The controller of this invention employs a Gaussian process regression model to model the uncertainties of the wet desulfurization process, achieving time-delay process predictive control that considers uncertainty compensation. This ensures good uncertainty prediction and assessment capabilities. Time delays can be compensated during the prediction process, improving the control efficiency of time-delayed objects in the desulfurization process. Furthermore, it can provide safety compensation for uncertainties caused by inaccurate modeling and unmeasurable disturbances during the control process, offering more flexible control and allowing the wet desulfurization system to avoid maintaining high power levels for extended periods, thus achieving energy-saving effects.
[0058] The parameters required in the modeling and control algorithms can be directly read from the main control DCS system, and there is usually no need to add expensive auxiliary equipment such as analysis or measurement instruments on site. Attached Figure Description
[0059] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention.
[0060] Figure 2 This is a schematic diagram illustrating the control effect of the wet desulfurization process obtained in the verification example of this invention.
[0061] Figure 3 This is a schematic diagram of the uncertainty prediction results obtained from the verification example in the embodiments of the present invention. Detailed Implementation
[0062] The specific embodiments of the present invention are described below with reference to the accompanying drawings.
[0063] See Figure 1 This embodiment of a predictive control method for a wet desulfurization system considering uncertainty compensation includes:
[0064] S1. Establish a prediction model for wet desulfurization, which is constructed by using a residual model based on Gaussian process theory to predict the unmodeled dynamics and process noise with marked uncertainties in the discrete-time model of the wet desulfurization system.
[0065] Specifically, a discrete-time model of the wet desulfurization system is established:
[0066] (1)
[0067] In the formula, , , Represent The concentration of sulfur dioxide at the outlet of the wet desulfurization system, the frequency of the desulfurization tower circulating pump, and the process noise at any given time; It is the nominal model of a wet desulfurization system. These are unmodeled dynamic characteristics used to characterize uncertainty. It is Transform into and Transformation matrix of results with the same dimension and All are differentiable functions, process noise It follows a Gaussian distribution and is spatially uncorrelated;
[0068] Specifically, using residual models For the unmodeled dynamics and the process noise Predictions are made to construct a prediction model for wet desulfurization processes:
[0069] (2)
[0070] In the formula, Representative at The predicted value at time can be understood. These are the state variables and control variables of the prediction model, respectively.
[0071] Specifically, the methods for constructing the nominal model include:
[0072] Based on on-site dynamic characteristic experiments, a transfer function model of the circulating pump frequency on the sulfur dioxide concentration at the outlet of the wet desulfurization system was established. Based on the transfer function model, a discrete state-space model was established as the nominal model as follows:
[0073] (3)
[0074] In the formula, This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump. For a moment, For the time delay of the process, , The coefficient matrices are as follows:
[0075]
[0076] in, , This refers to the sampling time of the on-site dynamic characteristic experiment. , These are the gain and time constant of the transmission process, respectively.
[0077] Specifically, residual model The construction methods include:
[0078] definition Input data for the residual model:
[0079] Based on the nominal model, the residuals are obtained. , This is the actual sulfur dioxide concentration value at the export site. This is the outlet sulfur dioxide concentration value output by the nominal model. yes The false reversal;
[0080] Based on Gaussian process theory, a kernel function is used to calculate the vector between training input data and test input data after kernel function computation. The matrix calculated by the kernel function between the training data Test data and its own constant calculated by the kernel function Construct a Gaussian process regression model And this regression model is used as a residual model. , among which, among which, These are the prior variance of the training dataset, the identity matrix, and the data vector of the training output, respectively.
[0081] S2. Improve the uncertainty propagation in the prediction time domain based on the wet desulfurization target prediction model: set a value greater than the desulfurization process delay. Prediction time domain In the prediction time domain The time delay phase within, and the time delay after passing. In the subsequent stages, the propagation of uncertainties in the wet desulfurization process is carried out to obtain the uncertainty propagation formula;
[0082] Specifically, the formula for the propagation of time-delay stages of uncertainty in the wet desulfurization process includes:
[0083] Configure a separate control law with state feedback. , It is feedback gain. This is the quantity to be optimized. This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump;
[0084] Based on the prediction model for wet desulfurization, the prediction output representation for the time delay stage is given:
[0085] (4)
[0086] superscript represent The prediction time domain of the first time step, Represents the concentration of sulfur dioxide in exports exist Predicted output at time step Representing the known, in The actual circulating pump frequency value at any given time; For a moment, For the time delay of the process, To calculate the number of steps;
[0087] In the time-delay stage of prediction, the outlet sulfur dioxide concentration is constructed. With residual Joint probability distribution:
[0088] (5)
[0089] In the formula, , These are the sulfur dioxide concentrations at the export sites. residual mean , These are the sulfur dioxide concentrations at the export sites. and residuals The covariance of itself It is the concentration of sulfur dioxide at the export site. With residual Covariance between; subscript delay Represents the delay phase. , Represents the mean and covariance of the joint probability distribution; superscript Represents the time delay phase time;
[0090] Based on the joint probability distribution of equation (5), the time-delay propagation formula for the uncertainty of the wet desulfurization process is obtained:
[0091] (6)
[0092] in,
[0093]
[0094] Specifically, the formula for the propagation of time-delayed uncertainties in the wet desulfurization process includes:
[0095] Based on the prediction model for wet desulfurization, random variables are constructed in the prediction time domain after the time delay phase. With residual Joint probability distribution:
[0096] (7)
[0097] In the formula, Represents random variables , Represents random variables The mean, Represents random variables The covariance of itself Represents random variables The covariance between the residual d and the residual d; , These are the mean and covariance of the joint probability distribution, indicated by the superscript. Represents the period after the time delay. time, For prediction in the time domain; where:
[0098]
[0099] Based on the joint probability distribution of equation (7), the propagation formula for the time-delayed stage of uncertainty in the wet desulfurization process is obtained:
[0100] (8)
[0101] in,
[0102]
[0103] S3. According to the desulfurization control requirements, set a set of constraints for the outlet sulfur dioxide concentration and the circulation pump frequency, and construct a probabilistic constraint for the desulfurization control process based on the set of constraints.
[0104] Specifically, set the constraint set for the outlet sulfur dioxide concentration. The set of constraints for the frequency of the circulating pump They are respectively:
[0105] (9)
[0106] In the formula, , and , These are the corresponding coefficient matrices for the outlet sulfur dioxide concentration constraint and the circulating pump frequency constraint, respectively.
[0107] Based on constraint sets and Based on the reachable set, construct probabilistic constraints for the desulfurization control process:
[0108] (10)
[0109] In the formula, The quantile function representing the normal distribution. and These represent the significance levels of the violation of constraints by the outlet sulfur dioxide concentration and the circulation pump frequency, respectively.
[0110] S4. Construct a predictive controller based on the prediction model established in S1, the uncertainty propagation formula constructed in S2, and the probability constraints constructed in S3, specifically including:
[0111] Based on the control requirements for outlet sulfur dioxide concentration and the required frequency of circulating pumps in the wet desulfurization control task, corresponding weight matrices are set. , ;
[0112] Using trace calculation, the mean of the control objective function is calculated for models containing random variables:
[0113] (11)
[0114] In the formula, It is a norm 2. It is the trace operator;
[0115] The predictive controller is established, and its mathematical expression is as follows:
[0116] (12)
[0117] In equation (12), the objective function From the transformation of (11), we obtain, where , , The three terms represent the prediction time domain endpoint, the prediction time domain, and the prediction time domain, respectively. The objective function values for the prediction results during the internal delay stage and the stage after the delay have been completed are calculated. The meanings of the parameters and symbols in the formula have been explained above and will not be repeated here.
[0118] This embodiment enables safety compensation for uncertainties in wet desulfurization predictive control, exhibiting good uncertainty prediction and assessment capabilities. Specifically, this embodiment uses a Gaussian process regression model to model uncertainties, achieving time-delay process predictive control that considers uncertainty compensation. Time delays can be compensated during the prediction process, improving the control efficiency of time-delayed objects in the wet desulfurization process and demonstrating good control performance. Furthermore, it can provide safety compensation for uncertainties caused by inaccurate modeling and unmeasurable disturbances during desulfurization control, offering more flexible control and allowing the wet desulfurization system to avoid maintaining high power levels for extended periods, thus achieving energy-saving effects.
[0119] The parameters required in the modeling and control algorithms can be directly read from the main control DCS system, and there is usually no need to add expensive auxiliary equipment such as analysis or measurement instruments on site.
[0120] The following verification examples further illustrate the effectiveness of the predictive control method in this embodiment.
[0121] This verification example selects the desulfurization system of a 1000MW coal-fired unit as the controlled object, and considers a predictive control method for the wet desulfurization system with uncertainty compensation, including:
[0122] S1. Based on on-site dynamic characteristic experiments, a first-order inertial plus time-delay transfer function model of the circulating pump frequency versus the outlet sulfur dioxide concentration of the wet desulfurization system is constructed:
[0123]
[0124] in, It is process gain. It is the time constant of the process. It is the process delay, and s is the Laplace operator;
[0125] The transfer function model is transformed to obtain the discrete state-space equations, which serve as the nominal model as follows:
[0126]
[0127] ,
[0128] Based on this, time constant perturbation is considered. As a model of a real wet desulfurization system.
[0129] For residual models, the size of the training dataset Based on the Gaussian process regression model, a residual model d is constructed.
[0130] Combining the nominal model and the residual model, a prediction model for wet desulfurization is constructed:
[0131]
[0132] S2. Based on the nominal model, the feedback gain K in the separation control law is obtained through LQR calculation. In the prediction time domain, during the time delay stage, the outlet sulfur dioxide concentration is constructed. With residual The joint probability distribution is shown in Equation (5). Based on the conditional distribution properties of Equation (5), the time-delay propagation formula for the uncertainty of the wet desulfurization process shown in Equation (6) is obtained. In the prediction time domain, after passing the time-delay stage, random variables are constructed. With residual The joint probability distribution is shown in Equation (7). Based on the conditional distribution property of Equation (7), the propagation formula of the uncertainty of the wet desulfurization process after the time delay stage is obtained as shown in Equation (8).
[0133] S3. According to the desulfurization control requirements, the outlet sulfur dioxide concentration must meet the range [0,30], and the circulation pump frequency must meet the range [40,50]. Therefore, a constraint set for the outlet sulfur dioxide concentration and the circulation pump frequency is set. and As shown in Equation (9), based on the probabilistic reachability set, the probabilistic constraints of the desulfurization control process are constructed as shown in Equation (10);
[0134] S4. Combine S1-S3 to construct the controller as shown in equation (12). Controller parameter settings: weight matrix parameters , Predicting the time domain .
[0135] Figure 2 and Figure 3 These are schematic diagrams of the control effect of the wet desulfurization process and the uncertainty prediction results in this verification example.
[0136] Depend on Figure 2 The control results show that the predictive control method for wet desulfurization systems that considers uncertainty compensation proposed in this invention can compensate for time delays during the prediction process, thereby improving the control efficiency of the desulfurization process. It can also provide safe compensation for uncertainties caused by inaccurate modeling and unmeasurable disturbances during the control process, providing more flexible control and enabling the wet desulfurization system to avoid maintaining a high power level for a long time, thus achieving energy-saving effects.
[0137] Depend on Figure 3 The uncertainty prediction results show that using a Gaussian process regression model to model the uncertainty of the wet desulfurization process can ensure good uncertainty prediction and assessment capabilities, and achieve time-delay process prediction and control that takes uncertainty compensation into account.
[0138] It will be understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A predictive control method for a wet desulfurization system considering uncertainty compensation, characterized in that, include: Establish a discrete-time model for the wet desulfurization system: ， In the formula, , , Represent The concentration of sulfur dioxide at the outlet of the wet desulfurization system, the frequency of the desulfurization tower circulating pump, and the process noise at any given time; It is the nominal model of a wet desulfurization system. These are unmodeled dynamic characteristics used to characterize uncertainty. It is Transform into and Transformation matrix of results with the same dimension and All are differentiable functions, process noise It follows a Gaussian distribution and is spatially uncorrelated; Using residual models For the unmodeled dynamics and the process noise Predictions are made to construct a prediction model for wet desulfurization processes: ， In the formula, the residual model This is a Gaussian process regression model constructed based on Gaussian process theory. Representative at The predicted value at any given time; Based on the prediction model for wet desulfurization, the uncertainty propagation in the prediction time domain is improved: a prediction time domain longer than the desulfurization process delay is set, and the uncertainty propagation of the wet desulfurization process is carried out in the delay stage and the stage after the delay in the prediction time domain, respectively, and the corresponding uncertainty propagation formula is obtained. According to the desulfurization control requirements, a set of constraints is set for the outlet sulfur dioxide concentration and the frequency of the circulating pump. Based on the set of constraints, a probabilistic constraint for the desulfurization control process is constructed. A predictive controller is constructed based on the aforementioned predictive model, uncertainty propagation formula, and probability constraints, and control is performed based on the predictive controller. The residual model The construction methods include: definition Input data for the residual model: Based on the nominal model, the residuals are obtained. , This is the actual sulfur dioxide concentration value at the export site. This is the outlet sulfur dioxide concentration value output by the nominal model. yes The false reversal; Based on Gaussian process theory, a kernel function is used to calculate the vector between training input data and test input data after kernel function computation. The matrix calculated by the kernel function between the training data Test data and its own constant calculated by the kernel function Construct a Gaussian process regression model And this regression model is used as a residual model. ,in, These are the prior variance of the training dataset, the identity matrix, and the data vector of the training output, respectively.
2. The predictive control method for a wet desulfurization system considering uncertainty compensation according to claim 1, characterized in that, The method for constructing the nominal model includes: Based on on-site dynamic characteristic experiments, a transfer function model of the circulating pump frequency on the sulfur dioxide concentration at the outlet of the wet desulfurization system was established. Based on the transfer function model, a discrete state-space model was established as the nominal model as follows: ， In the formula, This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump. For a moment, For the time delay of the process, , The coefficient matrices are as follows: ， in, , This refers to the sampling time of the on-site dynamic characteristic experiment. , These are the gain and time constant of the transmission process, respectively.
3. The predictive control method for a wet desulfurization system considering uncertainty compensation according to claim 1, characterized in that, In the time delay phase of the prediction time domain, the uncertainty of the wet desulfurization process is propagated, and the propagation formula of the uncertainty of the wet desulfurization process in the time delay phase is obtained, including: Configure a separate control law with state feedback. , It is feedback gain. This is the quantity to be optimized. This refers to the sulfur dioxide concentration at the outlet of the wet desulfurization system. It is the frequency of the desulfurization tower circulating pump; Based on the prediction model for wet desulfurization, the prediction output representation for the time delay stage is given: ， superscript represent The prediction time domain of the first time step, Represents the concentration of sulfur dioxide in exports exist Predicted output at time step Representing the known, in The actual circulating pump frequency value at any given time; For a moment, For the time delay of the process, To calculate the number of steps; Constructing the export sulfur dioxide concentration during the time delay phase in the prediction time domain With residual Joint probability distribution: ， In the formula, , These are the sulfur dioxide concentrations at the export sites. residual mean , These are the sulfur dioxide concentrations at the export sites. and residuals The covariance of itself It is the concentration of sulfur dioxide at the export site. With residual Covariance between; subscript delay Represents the delay phase. , Represents the mean and covariance of the joint probability distribution; superscript Represents the time delay phase time; Based on the joint probability distribution in the above equation, the time-delay propagation formula for uncertainty in the wet desulfurization process is obtained: 。 4. The predictive control method for a wet desulfurization system considering uncertainty compensation according to claim 3, characterized in that, In the prediction time domain, the uncertainty of the wet desulfurization process is propagated after the time delay, and the propagation formula of the uncertainty of the wet desulfurization process after the time delay is obtained, including: Based on the prediction model of wet desulfurization, a random variable after the time delay stage in the prediction time domain is constructed. With residual Joint probability distribution: ， In the formula, Represents random variables , Represents random variables The mean, Represents random variables The covariance of itself Represents random variables The covariance between the residual d and the residual d; , These are the mean and covariance of the joint probability distribution, indicated by the superscript. Represents the period after the time delay. time, For prediction in the time domain; combined with the aforementioned separation control law We can obtain: ， Based on the joint probability distribution in the above equation, the propagation formula for the time-delayed stage of uncertainty in the wet desulfurization process is obtained: 。 5. The predictive control method for a wet desulfurization system considering uncertainty compensation according to claim 4, characterized in that, Set the constraint set for the outlet sulfur dioxide concentration. The set of constraints for the frequency of the circulating pump They are respectively: ， In the formula, , and , These are the corresponding coefficient matrices for the outlet sulfur dioxide concentration constraint and the circulating pump frequency constraint, respectively. Based on constraint sets and Based on the reachable set, construct probabilistic constraints for the desulfurization control process: ， In the formula, The quantile function representing the normal distribution. and These represent the significance levels of the violation of constraints by the outlet sulfur dioxide concentration and the circulation pump frequency, respectively.
6. The predictive control method for a wet desulfurization system considering uncertainty compensation according to claim 5, characterized in that, A predictive controller is constructed using the aforementioned predictive model, uncertainty propagation formula, and probability constraints, including: Using trace operation, the mean value of the control objective function is predicted for models containing random variables; The predictive controller is established, and its mathematical expression is as follows: ， In the formula, the objective function In , , The three terms represent the prediction time domain endpoint, the prediction time domain, and the prediction time domain, respectively. The objective function calculation values of the prediction results in the internal delay stage and the stage after the delay has passed; It is a norm 2. It is the trace operator; , These are weight matrices set based on the control requirements for outlet sulfur dioxide concentration and the requirements for the frequency of circulating pumps in the wet desulfurization control task. , These are the nominal model coefficient matrices.