Wet desulphurization intelligent slurry supply control method

Abstract

The invention relates to the technical field of flue gas desulfurization control of coal-fired power plants, in particular to an intelligent slurry supply control method for wet desulfurization. Comprising the following steps: acquiring a system working condition based on a preset full working condition intelligent identification model, and accurately dividing a system operation state into three working conditions of stability, expansion and fall; calling an SO2 inlet flow prediction model, a circulating slurry pH value prediction model and an SO2 outlet concentration prediction model corresponding to system working conditions; a model prediction result is input into an advanced process control system, and closed-loop optimization control over the pH value and the SO2 outlet concentration is achieved by adjusting the frequency of a slurry supply pump and the operation combination of a slurry circulating pump in real time. According to the invention, the problem that the pH value of the slurry and the SO2 concentration at the outlet of the desulfurization tower cannot be stably and accurately controlled due to frequent fluctuation of unit load and fuel quality is solved, the automation level of a desulfurization system is effectively improved, the material consumption and the energy consumption are obviously reduced, and the economic and efficient operation of an environment-friendly island is realized.

Description

Technical Field

[0001] This invention relates to the field of flue gas desulfurization control technology in coal-fired power plants, and in particular to a control method for intelligent slurry supply in wet desulfurization. Background Technology

[0002] Limestone-gypsum wet desulfurization technology is currently the most widely used flue gas desulfurization technology in coal-fired power plants. Its core principle is to spray limestone slurry into the absorption tower, where it reacts chemically with SO2 in the flue gas to achieve desulfurization. The slurry supply system is a key subsystem of the desulfurization island, its task being to stably and accurately supply the absorption tower with slurry of appropriate concentration to maintain the stability of the circulating slurry pH and the outlet SO2 concentration.

[0003] Currently, most wet desulfurization systems still rely on traditional PID control or manual operation by operators for slurry supply control. However, desulfurization systems face two major challenges in actual operation: 1) frequent fluctuations in unit load lead to drastic changes in flue gas volume and SO2 concentration; 2) variable coal quality, especially sulfur content, directly causes significant changes in inlet SO2 load. These external disturbances give the system characteristics of large lag, strong coupling, and nonlinearity, making it difficult for traditional PID control to adapt, resulting in frequent fluctuations in circulating slurry pH and outlet SO2 concentration.

[0004] For example, the pH value of the No. 1 absorber in a power plant fluctuates between 5.2 and 5.8 (with a difference of about 4 times in hydrogen ion concentration), and the outlet SO2 concentration is between 15 and 35 mg / m³. 3 The pH level fluctuates and there is a risk of instantaneous exceedance. In order to mask the fluctuations and prevent exceedance, operators often adopt a conservative strategy, maintaining a low pH setpoint and a high slurry circulation volume for a long time. This leads to excessive limestone consumption and high power consumption of the slurry circulation pump, resulting in huge waste of energy and materials.

[0005] Therefore, developing a slurry supply method that can intelligently identify operating conditions and make accurate predictions and advanced controls based on these conditions is of great significance for improving the automation level of desulfurization systems, achieving energy conservation and emission reduction, and ensuring stable compliance with emission standards. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of existing wet desulfurization slurry supply systems being unable to adapt to load and coal quality fluctuations, having poor control accuracy, and high energy and material consumption. This invention provides a control method for intelligent slurry supply in wet desulfurization, which can intelligently identify operating conditions, accurately predict key parameters, and achieve optimized control.

[0007] The present invention provides a control method for intelligent slurry supply in wet desulfurization, comprising the following steps: S1. Collect historical operating data in real time from the power plant's DCS system database or SIS system database, and perform data cleaning, noise reduction and normalization processing; S2. The system operating conditions are obtained based on the preset full-condition intelligent identification model. The full-condition intelligent identification model takes the unit load change rate and fuel sulfur content change rate as core input features and automatically identifies the system operating conditions into three types: stable operating conditions, rising operating conditions, and falling operating conditions. S3. Call the SO2 inlet flow prediction model, circulating slurry pH prediction model and SO2 outlet concentration prediction model corresponding to the system operating conditions; S4. The output of the SO2 inlet flow prediction model is used as a feedforward signal, and the deviations between the outputs of the circulating slurry pH prediction model and the SO2 outlet concentration prediction model and the set values ​​are used as feedback signals, and both are input to the APC controller. S5. The APC controller calculates the optimized control command based on the input prediction and deviation signals and sends it to the actuator to adjust the frequency of the slurry pump and the start-stop combination of the slurry circulation pump, and finally completes the control of the pH value of the circulating slurry and the SO2 outlet concentration.

[0008] Furthermore, the historical operating data in step S1 includes unit load signal, fuel sulfur content data, flue gas flow rate at the absorber inlet, SO2 concentration at the inlet, pH value of circulating slurry, slurry flow rate, slurry density, absorber level, and SO2 concentration at the outlet.

[0009] Furthermore, in step S2, the full-condition intelligent identification model uses a machine learning classification algorithm to automatically identify the system operating conditions into three types: stable operating condition, rising operating condition, and falling operating condition, as detailed below: Define the unit load change rate: ; Fuel sulfur content change rate: ,in L(t) Let be the unit load at time t. S(t) Let Δ be the sulfur content of the fuel at time t. t The sampling interval; With thresholds α and β set, the discrimination criteria for classifying operating conditions are as follows: when R load (t) <α and R sulfur (t) When β < 0, it is determined to be a steady-state condition; when R load (t) ≥α and R sulfur(t) When ≥β, it is determined to be an expansion condition; when R load (t) ≤-α and R sulfur (t) ≤ － When β is reached, it is determined to be a drop condition; Among them, the threshold α is 5% of the rated load / minute, and β is 0.1% of the sulfur content / minute.

[0010] Furthermore, the mathematical expression for the SO2 inlet flow prediction model in step S3 is: ; Where t is the current time, For the future k Predicted SO2 inflow rate at time [time]. L For unit load, S For fuel sulfur content, The inlet flue gas flow rate, The inlet SO2 concentration, It is a machine learning regression function obtained by training on historical data. k The step size is set to 5-10 minutes.

[0011] Furthermore, the pH prediction model for the circulating slurry in step S3 is constructed using LSTM, and its hidden layer state update formula is as follows: The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in, x t The input vector at time t includes These represent the SO2 inlet flow rate, the pH value at the previous moment, the slurry flow rate, the slurry density, and the absorber level, respectively. h t The output state at time t is used to finally obtain the pH prediction value through mapping via a fully connected layer: The model's prediction error is no greater than ±0.1. Among them, the forget gate determines how many historical cell states are retained; the input gate determines how much new information is updated; the candidate cell state is the carrier of new information; the cell state update is the combination of historical and new information; the output gate determines how many cell states are output; and the hidden layer output is used for subsequent prediction.

[0012] Furthermore, in step S3, the SO2 outlet concentration prediction model and the circulating slurry pH prediction model use the same LSTM framework, and their hidden layer state update formula is as follows: The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in, x t The input vector at time t includes , pH pred (t+1) , F slurry (t) , D slurry (t) , H level (t) , Label(t) These represent the predicted SO2 inlet flow rate, predicted circulating slurry pH value, slurry supply flow rate, slurry supply density, absorber level, and operating condition label, respectively. The hidden layer output h t By mapping the fully connected layer to the predicted SO2 outlet concentration, an adaptive weight based on operating conditions is introduced. W Label(t) To adapt to all operating conditions, the formula is as follows: ; in: This is the predicted SO2 outlet concentration at time k, in mg / Nm³. 3 ; W Label(t) For adaptive weighting based on operating conditions; W h This is the weight matrix of the fully connected layer, with dimensions of and . h t match;b h For bias terms of fully connected layers; This is the model systematic error correction term, with a value range of [-3, 3] mg / Nm 3 This is used to compensate for prediction bias.

[0013] Furthermore, in step S5, the APC controller employs a model predictive control algorithm, the optimization problem of which is expressed as:

[0014] in, N p To predict the time domain, N c To control the time domain, pH set and C set These are the set values ​​for pH and SO2 outlet concentration, respectively. This is the weight matrix. u The control variables include the slurry pump frequency command and the start / stop status of the slurry circulation pump, Δ u The rate of change of the control variable is used to solve the optimization problem, obtain the optimal control sequence, and send it to the DCS system for execution.

[0015] Furthermore, it also includes configuring bumpless handover and safety interlocking logic in the DCS system, the logic including: Communication heartbeat detection logic: If the communication between the APC controller and the DCS system is interrupted for more than a set time, the control will be automatically switched back to the DCS manual mode. Output value anti-jump logic: The output of the APC controller must pass through a deviation comparator, and the write operation will not be performed if the difference between the output and the current value exceeds the safety limit.

[0016] The present invention provides a storage medium storing a computer program that executes the intelligent slurry supply control method for wet desulfurization.

[0017] The present invention provides an electronic device, including a processor, a display, and a storage medium, wherein the storage medium is the aforementioned storage medium, and the processor is used to run a computer program on the storage medium.

[0018] Compared with the prior art, the present invention has the following significant advantages: High control precision: This invention stabilizes the pH fluctuation range of the circulating slurry from ±0.3 to ±0.1 through multi-step prediction and optimization control, and controls the SO2 outlet concentration at 28±3 mg / Nm³. 3 Within this range, the risk of exceeding emission standards is effectively avoided.

[0019] Significant energy-saving and consumption-reducing effects: Under the premise of ensuring desulfurization efficiency, this invention can achieve an energy saving rate of over 17.54% by optimizing the operation combination and frequency of the slurry circulation pump; and can reduce limestone consumption by precisely controlling the slurry supply.

[0020] Strong adaptability: The unique full-condition identification model of this invention enables the system to automatically sense external disturbances, such as load and coal quality changes, and switch to the corresponding prediction and control strategies, thereby improving the robustness of the system.

[0021] Improved automation level: This invention realizes the transformation from manual intervention to intelligent automation, increasing the basic circuit self-control rate to over 95%, and significantly reducing the labor intensity of operators. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the control method in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, the singular form includes the plural form unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this description, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The overall architecture of this invention's system comprises three layers: a digital software platform, an APC intelligent controller, and intelligent instruments. The software platform is deployed on a plant-level server and is responsible for data storage, model calculation, and visualization. The APC controller, using an industrial host computer, is deployed in the electronics room and interacts with the DCS system in real time. Intelligent instruments (such as densitometers and future expandable slurry flow meters) are used to enhance the system's sensing capabilities.

[0028] Example A method for controlling intelligent slurry supply in wet desulfurization includes the following steps: Data acquisition and preprocessing: Historical operating data is collected in real time from the power plant's DCS, SIS system and coal quality testing database, and data cleaning, noise reduction and normalization are performed; Intelligent identification of all operating conditions: Construct a classification model with the unit load change rate and fuel sulfur content change rate as the core features to accurately classify the system operating status into three types: stable operating condition, rising operating condition, and falling operating condition; Multivariate prediction model construction: For different operating conditions, short-term prediction models for SO2 inlet flow rate, circulating slurry pH value, and SO2 outlet concentration were established respectively. The circulating slurry pH value prediction model adopts an LSTM network, and its inputs include SO2 inlet flow rate, historical pH value, and slurry supply flow rate, with a prediction accuracy within ±0.1. APC Advanced Control: The output of the predictive model is connected to the APC controller. This controller uses a model predictive control (MPC) algorithm to minimize the deviation between the predicted and set values ​​of pH and SO2 concentrations, while constraining the variation of control variables, and solves for the optimal slurry pump frequency and slurry circulation pump start / stop commands. System integration and safety deployment: The APC controller is integrated with the power plant's DCS system via the OPC protocol, and complete seamless switching and safety interlocking logic is configured to ensure that the control system can be put into automatic operation safely and reliably.

[0029] like Figure 1 As shown, the specific content is as follows: S1. Collect historical operating data in real time from the power plant's DCS system database or SIS system database, and perform data cleaning, noise reduction and normalization processing. The historical operating data includes unit load signal, fuel sulfur content data, flue gas flow rate at the absorber inlet, SO2 concentration at the inlet, pH value of circulating slurry, slurry flow rate, slurry density, absorber level and SO2 concentration at the outlet. S2. The system operating conditions are obtained based on a preset intelligent identification model for all operating conditions. The intelligent identification model for all operating conditions uses the unit load change rate and fuel sulfur content change rate as core input features. It uses a machine learning classification algorithm to automatically identify the system operating conditions into three types: stable operating condition, rising operating condition, and falling operating condition, as detailed below: Define the unit load change rate: ; Fuel sulfur content change rate: ,in L(t) Let be the unit load at time t. S(t) Let Δ be the sulfur content of the fuel at time t. t The sampling interval; Set thresholds α and β, where threshold α is 5% of rated load / minute and β is 0.1% of sulfur content / minute. The system calculates the rate of change of load and sulfur content every minute. If the calculation yields R load (t) =3% / min, R sulfur (t) When the percentage is 0.05% / min, all values ​​are less than the threshold, so the model determines that the current condition is a steady-state condition. If the load command suddenly increases, the calculation will yield the following results: R load (t) If 8% / min > α, the model immediately determines that it has entered a rising working condition and activates the corresponding prediction and control sequences.

[0030] S3. Call the SO2 inlet flow prediction model, circulating slurry pH prediction model, and SO2 outlet concentration prediction model corresponding to the system operating conditions; the mathematical expression of the SO2 inlet flow prediction model is:

[0031] Where t is the current time, For the future k Predicted SO2 inflow rate at time [time]. L For unit load, S For fuel sulfur content, The inlet flue gas flow rate, The inlet SO2 concentration, It is a machine learning regression function obtained by training on historical data. k To predict the step size, a value of 5-10 minutes is used; The pH prediction model for circulating slurry is constructed using an LSTM (Long Short-Term Memory) network, and its hidden layer state update formula is as follows: The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in, x t The input vector at time t includes , respectively represent SO2 inlet flow rate, pH value at the previous moment, slurry flow rate, slurry density, and absorber level; h t The output state at time t is used to finally obtain the pH prediction value through mapping via a fully connected layer: The model's prediction error is no greater than ±0.1. Historical data from six consecutive months of continuous operation were collected from absorber #1, and the data segment under stable operating conditions was selected as the training set. A two-layer LSTM network was constructed, with the input variables being [SO2 inlet flow rate at time t, pH value at time t-1, slurry flow rate at time t, slurry density at time t, and absorber level at time t], and the output being the predicted pH value at time t+1. After training, the model's prediction error RMSE on the test set was less than 0.05, and the maximum absolute error did not exceed 0.1, meeting the control requirements.

[0032] The SO2 export concentration prediction model, taking into account the dynamic differences between stable, rising, and falling operating conditions, employs a condition-specific LSTM construction to predict the next 5-10 minutes (prediction step size). k SO2 outlet concentration (denoted as 5-10 minutes) C SO2-out Accurate prediction provides feedback signals to support the subsequent APC controller, as detailed below: I. Input Feature Design for SO2 Outlet Concentration Prediction Model The input characteristics of the SO2 outlet concentration prediction model are based on the operational data already collected in the original scheme and the output of other prediction models, without adding any additional collected parameters, ensuring consistency with the parameter system of the overall technical scheme. Specifically, this includes: 1. Predicted SO2 inflow rate: denoted as Q SO2-in (t + k) The SO2 inlet flow prediction model constructed in step S3 reflects the lag effect of future SO2 input load on outlet concentration. 2. Predicted pH value of circulating slurry: denoted as pH pred (t+1) The pH value prediction model for circulating slurry, which was constructed in step S3, is derived from the pH value prediction model. The pH value directly determines the desulfurization reaction efficiency and is the core influencing factor of the outlet concentration. 3. Grout supply flow rate: denoted as F slurry (t)The historical operating data collected in step S1 reflects the regulatory effect of slurry supply on desulfurization capacity. 4. Grout density: denoted as D slurry (t) The historical operating data collected from step S1 reflects the limestone content in the slurry, which affects the sufficiency of reactants. 5. Absorber tower liquid level: denoted as H level (t) Historical operational data collected from step S1 shows that stable liquid level is the foundation for uniform slurry spraying and avoids concentration fluctuations. 6. Operating Condition Label: Recorded as Label(t) The full-condition intelligent identification model from step S2 marks the current condition type (0 = stable condition, 1 = rising condition, 2 = falling condition) for sub-model switching and parameter adaptation.

[0033] The input features need to undergo the same data preprocessing as in step S1: use moving average filtering (window size = 3 sampling intervals) to remove high-frequency noise, use linear interpolation to complete missing data (missing rate ≤ 0.1%), and perform normalization in the [0,1] interval to ensure model training efficiency and prediction accuracy.

[0034] Structure and mathematical expression of the SO2 outlet concentration prediction model: The SO2 outlet concentration prediction model adopts a condition-specific LSTM architecture, consistent with the LSTM framework of the circulating slurry pH prediction model. Independent sub-models are trained for three different operating conditions, using operating condition labels. Label(t) Automatic switching, the specific structure and mathematical expression are as follows: 1. LSTM Hidden Layer State Update Formula LSTM captures temporal features through forgetting gates, input gates, candidate cell states, cell state updates, and output gates. Its formulas are completely consistent with the notation system of circulating slurry pH prediction models. The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in: x tLet t be the input vector at time t, with a dimension of 6 (corresponding to the 6 input features mentioned above); h t-1 The hidden layer state of the Long Short-Term Memory network at time t-1; the dimension is 64 under stable conditions, and 128 under rising and falling conditions; W f 、W i 、W c 、W o : These are the weight matrices for the forget gate, input gate, candidate cell state, and output gate, respectively, which are trained and iteratively updated using historical data; b f 、b i 、b c 、b o : These are the bias terms of the above gating, initially set to 0, and adaptively adjusted during training; σ It is a sigmoid activation function with an output range of [0,1], used to control the gate switch; tanh is the hyperbolic tangent activation function with an output range of [-1, 1], used to alleviate gradient vanishing. This is an element product operation, consistent with the operation rules of the circulating slurry pH prediction model; C t-1 、C t These are the cell states at time t-1 and time t, respectively, used to store temporal features.

[0035] 2. Output layer mapping formula Hidden layer output of LSTM h t By mapping the fully connected layer to the predicted SO2 outlet concentration, an adaptive weight based on operating conditions is introduced. W Label(t) To adapt to all operating conditions, the formula is as follows: ; in: This is the predicted SO2 outlet concentration at time k, in mg / Nm³. 3 ; W Label(t) For adaptive weighting under stable operating conditions W Label(t)=1.0, under increased operating conditions W Label(t) =1.3, under drop conditions W Label(t) =1.1, used to enhance the predicted response under extreme conditions; W h This is the weight matrix of the fully connected layer, with dimensions of and . h t Match (64×1 or 128×1); b h For bias terms of fully connected layers; This is the model systematic error correction term, with a value range of [-3, 3] mg / Nm 3 This is used to compensate for prediction bias.

[0036] II. Parameter Optimization of the Working Condition Model To adapt to the dynamic characteristics of different operating conditions, the sub-model parameters of the SO2 outlet concentration prediction model are optimized according to the operating condition type, consistent with the parameter design logic of the circulating slurry pH prediction model. The specific parameters are shown in the table below:

[0037] S4. The output of the SO2 inlet flow prediction model is used as a feedforward signal, and the deviations between the outputs of the circulating slurry pH prediction model and the SO2 outlet concentration prediction model and the set values ​​are used as feedback signals, and both are input to the APC controller. The S5.APC controller calculates optimized control commands based on the input prediction and deviation signals and sends them to the actuators to regulate the frequency of the slurry supply pump and the start-stop combination of the slurry circulation pump, ultimately achieving stable control of the pH value and SO2 outlet concentration of the circulating slurry. The APC controller uses a model predictive control algorithm, and its optimization problem is expressed as:

[0038] in, N p To predict the time domain, N c To control the time domain, pH set and C set These are the set values ​​for pH and SO2 outlet concentration, respectively. This is the weight matrix. u The control variables include the slurry pump frequency command and the start / stop status of the slurry circulation pump, Δ u The rate of change of the control variable is used to solve the optimization problem, obtain the optimal control sequence, and send it to the DCS system for execution.

[0039] Set control targets: pH setpoint = 5.5, SO2 concentration setpoint = 28 mg / Nm³ 3 Predicting the time domain N p =10, control time domain N c =2. At a certain point during the expansion phase, the circulating slurry pH prediction model predicted that the pH would drop to 5.35 after 120 seconds, while the SO2 outlet concentration prediction model predicted that it would rise to 32 mg / Nm³. 3 The MPC algorithm within the APC controller solved an optimization problem and determined the optimal control action: to increase the slurry pump frequency by 3Hz and start a standby slurry circulation pump within the next 30 seconds. This instruction, after a safety logic check, was sent to the DCS for execution, successfully stabilizing the pH and SO2 concentrations within the target range.

[0040] Practical applications: This invention is implemented using a wet desulfurization system (absorber tower #1) for a 1000MW unit in a power plant as an example.

[0041] Hardware deployment: Two new data acquisition servers and two APC servers were added and connected to the existing DCS / SIS network via switches.

[0042] Software deployment: Deploy the intelligent operating condition recognition, pH prediction, SO2 prediction, and other models developed in this invention on the digital intelligence platform.

[0043] Commissioning and operation: First, historical data collection and model training are carried out. Then, the pH-APC controller is put into operation under stable conditions. After stabilization, the SO2-APC controller is put into operation.

[0044] Operational Results: After the system was put into operation, one month's worth of operational data was collected, and the results showed: The standard deviation of pH value decreased from 0.15 to 0.05; The standard deviation of SO2 outlet concentration was 5.2 mg / Nm³. 3 Reduced to 1.8 mg / Nm 3 And there are no records of exceeding the standard; The overall power consumption of the slurry circulation pump decreased by 17.6%; The consumption of limestone slurry decreased by approximately 3.5%.

[0045] The project saves approximately 2.6 million yuan in costs annually, achieving significant economic and environmental benefits.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A control method for intelligent slurry supply in wet desulfurization, characterized in that, Includes the following steps: S1. Collect historical operating data in real time from the power plant's DCS system database or SIS system database, and perform data cleaning, noise reduction and normalization processing; S2. The system operating conditions are obtained based on the preset full-condition intelligent identification model. The full-condition intelligent identification model takes the unit load change rate and fuel sulfur content change rate as core input features and automatically identifies the system operating conditions into three types: stable operating conditions, rising operating conditions, and falling operating conditions. S3. Call the SO2 inlet flow prediction model, circulating slurry pH prediction model and SO2 outlet concentration prediction model corresponding to the system operating conditions; S4. The output of the SO2 inlet flow prediction model is used as a feedforward signal, and the deviations between the outputs of the circulating slurry pH prediction model and the SO2 outlet concentration prediction model and the set values ​​are used as feedback signals, and both are input to the APC controller. S5. The APC controller calculates the optimized control command based on the input prediction and deviation signals and sends it to the actuator to adjust the frequency of the slurry pump and the start-stop combination of the slurry circulation pump, and finally completes the control of the pH value of the circulating slurry and the SO2 outlet concentration.

2. The control method according to claim 1, characterized in that, The historical operating data in step S1 includes unit load signal, fuel sulfur content data, flue gas flow rate at the absorber inlet, SO2 concentration at the inlet, pH value of circulating slurry, slurry flow rate, slurry density, absorber level, and SO2 concentration at the outlet.

3. The control method according to claim 2, characterized in that, The intelligent identification model for all operating conditions in step S2 uses a machine learning classification algorithm to automatically identify the system operating conditions into three types: stable operating condition, rising operating condition, and falling operating condition, as detailed below: Define the unit load change rate: ； Fuel sulfur content change rate: ,in L(t) Let be the unit load at time t. S(t) Let Δ be the sulfur content of the fuel at time t. t The sampling interval; With thresholds α and β set, the discrimination criteria for classifying operating conditions are as follows: when R load (t) <α and R sulfur (t) When β < 0, it is determined to be a steady-state condition; when R load (t) ≥α and R sulfur (t) When ≥β, it is determined to be an expansion condition; when R load (t) ≤-α and R sulfur (t) ≤ － When β is reached, it is determined to be a drop condition.

4. The control method according to claim 2, characterized in that, The mathematical expression for the SO2 inflow prediction model in step S3 is: ； Where t is the current time, For the future k Predicted SO2 inflow rate at time [time]. L For unit load, S For fuel sulfur content, The inlet flue gas flow rate, The inlet SO2 concentration, It is a machine learning regression function obtained by training on historical data. k The step size is set to 5-10 minutes.

5. The control method according to claim 4, characterized in that, The pH prediction model for the circulating slurry in step S3 is constructed using LSTM, and its hidden layer state update formula is as follows: The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in, x t The input vector at time t includes These represent the SO2 inlet flow rate, the pH value at the previous moment, the slurry flow rate, the slurry density, and the absorber level, respectively. h t The output state at time t is used to finally obtain the pH prediction value through mapping via a fully connected layer: .

6. The control method according to claim 5, characterized in that, In step S3, the SO2 outlet concentration prediction model uses the same LSTM framework as the circulating slurry pH prediction model, and its hidden layer state update formula is as follows: The formula for the forgetting gate is: ; The formula for the input gate is: ; Formula for candidate cell state: ; The formula for cell state renewal: ; The formula for the output gate is: ; The formula for the output of the hidden layer is: ; in, x t The input vector at time t includes , pH pred (t+1) , F slurry (t) , D slurry (t) , H level (t) , Label(t) These represent the predicted SO2 inlet flow rate, predicted circulating slurry pH value, slurry supply flow rate, slurry supply density, absorber level, and operating condition label, respectively. The hidden layer output h t By mapping the fully connected layer to the predicted SO2 outlet concentration, an adaptive weight based on operating conditions is introduced. W Label(t) To adapt to all operating conditions, the formula is as follows: ； in: This is the predicted SO2 outlet concentration at time k, in mg / Nm³. 3 ; W Label(t) For adaptive weighting based on operating conditions; W h This is the weight matrix of the fully connected layer, with dimensions of and . h t match; b h For bias terms of fully connected layers; This is the model systematic error correction term, with a value range of [-3, 3] mg / Nm 3 This is used to compensate for prediction bias.

7. The control method according to claim 6, characterized in that, In step S5, the APC controller employs a model predictive control algorithm, and its optimization problem is expressed as follows: in, N p To predict the time domain, N c To control the time domain, pH set and C set These are the set values ​​for pH and SO2 outlet concentration, respectively. This is the weight matrix; u The control variables include the slurry pump frequency command and the start / stop status of the slurry circulation pump, Δ u The rate of change of the control variable is used to solve the optimization problem, obtain the optimal control sequence, and send it to the DCS system for execution.

8. The control method according to claim 1, characterized in that, It also includes configuring bumpless handover and safety interlocking logic in the DCS system, the logic including: Communication heartbeat detection logic: If the communication between the APC controller and the DCS system is interrupted for more than a set time, the control will be automatically switched back to the DCS manual mode. Output value anti-jump logic: The output of the APC controller must pass through a deviation comparator, and the write operation will not be performed if the difference between the output and the current value exceeds the safety limit.

9. A storage medium, characterized in that, The computer program stores the control method for intelligent slurry supply in wet desulfurization as described in any one of claims 1-8.

10. An electronic device, characterized in that, It includes a processor, a display, and a storage medium, wherein the storage medium is the storage medium of claim 9, and the processor is used to run a computer program on the storage medium.

Images