Multi-parameter feedback-based precise automatic control system and method for NID desulfurization device

Abstract

The present application relates to the technical field of flue gas desulfurization, and particularly relates to a multi-parameter feedback NID desulfurization device precise automatic control system and method, through real-time collection of multi-source parameters, the multi-source parameters are input to a pre-trained machine learning model to output the set values of new ash feeding amount, circulating ash feeding amount and water spraying amount, and then the new ash feeder, the circulating ash feeder and the water spraying adjusting valve are adjusted according to the set values, and the set values are feedback corrected according to the deviation between the measured value and the target value of the outlet emission parameter; at the same time, the ash discharging time and the ash discharging amount are automatically decided according to the real-time acquired effective Ca(OH)2 content of the mixed ash and the fluidized bottom bin level. Through the collaborative control of multi-parameter fusion perception, machine learning prediction decision and feedback correction, the problems of control lag, parameter mismatch and extensive ash discharging in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of flue gas desulfurization technology, and in particular to a precise automatic control system and method for a multi-parameter feedback NID desulfurization device. Background Technology

[0002] NID (New Integrated Desulfurization) is a mature semi-dry flue gas desulfurization process. It utilizes quicklime (CaO) or hydrated lime (Ca(OH)2) as a desulfurizing agent, reacting with SO2 in the flue gas to produce calcium sulfite or calcium sulfate, thereby achieving desulfurization. This process offers advantages such as low investment, small footprint, and no wastewater discharge, and is widely used in small and medium-sized coal-fired boilers, steel sintering machines, and waste incineration. A typical NID desulfurization system usually includes a reactor, mixer, fluidized bed, dust collector, and corresponding material conveying and water spraying devices. Its core control parameters include the dosage of new ash (desulfurizing agent), the amount of recycled ash added, and the water spraying rate. The degree of matching of these parameters directly determines the desulfurization efficiency and operational stability.

[0003] However, existing NID desulfurization systems generally suffer from insufficient control precision and lag in actual operation. Currently, most NID desulfurization units employ a traditional control method based on SO2 concentration feedback from a chimney CEMS (Continuous Emission Monitoring System). Operators manually adjust the desulfurizing agent feed rate based on the outlet SO2 concentration, or use a simple PID controller for automatic adjustment. Since the residence time of flue gas in the reactor is typically only a few seconds, while the CEMS analysis cycle is 1–5 minutes, this results in a significant delay in control response, making it difficult to handle sudden changes in flue gas parameters (such as SO2 concentration and flue gas volume). Furthermore, the adjustment of desulfurizing agent dosage, circulating ash volume, and water spray volume relies heavily on operator experience and judgment, lacking real-time monitoring and quantitative data on intermediate process parameters. This easily leads to problems such as fluctuating desulfurization efficiency, desulfurizing agent waste, system caking, or fluidized bed silo overflow. Specifically, the moisture content and effective calcium content of circulating ash are key intermediate parameters affecting the efficiency of desulfurization reaction, but the existing system lacks online monitoring methods, and operators cannot grasp this information in real time; the timing and amount of ash discharge from the fluidized bed silo are usually controlled based on fixed time intervals or simple level switches, failing to combine with the effective calcium content in the circulating ash for real-time judgment, resulting in waste of effective calcium hydroxide or loss of control over the silo level.

[0004] The long-term existence of the above problems not only leads to increased operating costs and energy and material consumption of the desulfurization system, but also affects the stable compliance of flue gas emissions, and may even cause equipment failures and unplanned shutdowns. Summary of the Invention

[0005] The purpose of this invention is to provide a precise automatic control system and method for a multi-parameter feedback NID desulfurization device, which solves the problems of control lag, parameter mismatch and crude ash discharge in the prior art.

[0006] To achieve the above objectives, the present invention provides a precise automatic control method for a multi-parameter feedback NID desulfurization unit, comprising the following steps: Real-time acquisition of multi-source parameters, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters; The multi-source parameters are input into a pre-trained machine learning model, and the machine learning model outputs set values ​​for the new ash feed rate, the circulating ash feed rate, and the water spray rate. Adjust the new ash feeder, circulating ash feeder, and water spray regulating valve according to the set values; The setpoint is corrected based on the deviation between the measured value and the target value of the export emission parameters.

[0007] Among them, real-time acquisition of multi-source parameters, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters, specifically includes: The inlet flue gas parameters include inlet SO2 concentration, flue gas flow rate, flue gas temperature, and flue gas humidity; the mixed ash state parameters include mixed ash moisture content and effective Ca(OH)2 content; the material level parameter is the material level height of the fluidized bed; and the outlet emission parameter is the outlet SO2 concentration.

[0008] Specifically, the multi-source parameters are input into a pre-trained machine learning model, and the machine learning model outputs set values ​​for the new ash feed rate, the circulating ash feed rate, and the water spray rate, including: The input features of the machine learning model also include composite features constructed based on the multi-source parameters. The composite features include at least one of SO2 load, calcium-sulfur ratio estimate, and moisture content deviation. The SO2 load is obtained by multiplying the inlet SO2 concentration and the flue gas flow rate. The calcium-sulfur ratio estimate is calculated based on the new ash feed rate, the effective Ca(OH)2 content, and the SO2 load. The moisture content deviation is the difference between the current moisture content and the target moisture content.

[0009] Specifically, the feedback correction of the setpoint based on the deviation between the measured value and the target value of the outlet emission parameters includes: The outlet SO2 concentration is acquired every first preset period, and the current desulfurization efficiency is calculated. If the current desulfurization efficiency is lower than the preset threshold, the set value is incrementally compensated according to the proportional-integral-derivative control algorithm, or the machine learning model is updated online using historical data from the most recent period, until the desulfurization efficiency recovers to above the preset threshold.

[0010] The method also includes intelligent ash removal control: The system can acquire the effective Ca(OH)2 content of the mixed ash and the material level in the fluidized bed silo in real time, and automatically decide the timing and amount of ash discharge based on the combination of effective Ca(OH)2 content and material level.

[0011] The intelligent ash removal control specifically includes: When the effective Ca(OH)2 content is lower than the first threshold and the material level is higher than the second threshold, the ash removal device is started and runs for the first preset time. After the ash removal is completed, retest. If the effective Ca(OH)2 content is still lower than the first threshold and the material level is still higher than the second threshold, continue ash removal until it returns to normal. When the effective Ca(OH)2 content is lower than the first threshold but the material level is not higher than the second threshold, an alarm for insufficient effective calcium is issued and the amount of new ash feed is increased first. When the material level is higher than the third threshold, the ash removal device is forcibly activated and a silo burst warning is issued; wherein the third threshold is higher than the second threshold.

[0012] The machine learning model is constructed using a gradient boosting decision tree algorithm and is incrementally trained using newly added running data according to a second preset period to maintain the model's prediction accuracy.

[0013] Before inputting the multi-source parameters into the pre-trained machine learning model, the method further includes: The raw data collected in real time is subjected to outlier removal, data smoothing and normalization, and composite features are constructed for model input.

[0014] A multi-parameter feedback NID desulfurization device precision automatic control system includes a detection system, an execution system, and a control system, wherein the control system is connected to the detection system and the execution system respectively. The detection system is used to collect multi-source parameters in real time, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters. The execution system includes a new ash feeder, a circulating ash feeder, a water spray regulating valve, and an ash discharge device, which are used to adjust their respective actions according to control commands; The control system includes a data acquisition layer, a model decision-making layer, and a control execution layer; The data acquisition layer is connected to the detection system and is used to receive and preprocess the multi-source parameters acquired by the detection system. The model decision layer is connected to the data acquisition layer and is used to output the set values ​​of new ash feed rate, circulating ash feed rate and water spray rate through a machine learning model based on the preprocessed multi-source parameters. The control execution layer is connected to the model decision layer and the execution system respectively, and is used to convert the set value into control commands and send them to the execution system, and to provide feedback correction to the set value based on the deviation between the measured value and the target value of the outlet emission parameters.

[0015] The detection system includes: The flue gas pre-monitoring module is installed in the reactor inlet flue and includes an SO2 concentration monitor, a flue gas flow meter, a temperature sensor, and a humidity sensor. The output of each instrument is communicatively connected to the input of the data acquisition layer. The mixer monitoring module, installed at the mixer, includes a near-infrared moisture meter and an online calcium content analyzer, with the output of each instrument communicating with the input of the data acquisition layer; The silo monitoring module, installed on the top of the fluidized bed silo, includes a radar level gauge, the output of which is communicatively connected to the input of the data acquisition layer. The emission monitoring module is communicatively connected to the chimney CEMS system, and its output is communicatively connected to the input of the data acquisition layer.

[0016] This invention discloses a multi-parameter feedback-based precise automatic control system and method for a NID desulfurization device. It collects multi-source parameters in real time, including inlet flue gas parameters, mixed ash state parameters, fluidized bed silo level parameters, and outlet emission parameters. These parameters are input into a pre-trained machine learning model to output setpoints for the fresh ash feed rate, circulating ash feed rate, and water spray rate. The model then adjusts the fresh ash feeder, circulating ash feeder, and water spray regulating valve according to these setpoints, and provides feedback correction based on the deviation between the measured and target values ​​of the outlet emission parameters. Simultaneously, it includes an intelligent ash discharge control step, automatically determining the timing and amount of ash discharge based on the real-time acquired effective Ca(OH)2 content of the mixed ash and the fluidized bed silo level. Through the coordinated control of multi-parameter fusion sensing, machine learning predictive decision-making, and feedback correction, it solves the problems of control lag, parameter mismatch, and coarse ash discharge in existing technologies, achieving precise automatic control of the desulfurization process, significantly improving desulfurization efficiency and system operational stability, and reducing material consumption and operating costs. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a flowchart of the steps of the precise automatic control method for a multi-parameter feedback NID desulfurization device according to the first embodiment of the present invention.

[0019] Figure 2This is a schematic diagram of the principle of the multi-parameter feedback NID desulfurization device precision automatic control system according to the second embodiment of the present invention.

[0020] In the diagram: 201-Detection system, 202-Execution system, 203-Control system, 204-Data acquisition layer, 205-Model decision layer, 206-Control execution layer, 207-Flue gas pre-monitoring module, 208-Mixer monitoring module, 209-Bus monitoring module, 210-Emission monitoring module. Detailed Implementation

[0021] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0022] The first embodiment of this application is as follows: Please see Figure 1 This invention provides a precise automatic control method for a multi-parameter feedback NID desulfurization device, comprising the following steps: S101: Real-time acquisition of multi-source parameters, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters; Specifically, various online monitoring instruments deployed at key locations in the NID desulfurization system synchronously collect multi-source parameters according to a fixed sampling period. The sampling period can be set according to the dynamic characteristics of the system and control requirements; in this embodiment, it is set to 10 seconds to ensure timely capture of changes in flue gas parameters.

[0023] The acquisition of inlet flue gas parameters is achieved through the pre-flue gas monitoring module 207. This module is installed in the straight pipe section before the reactor inlet, approximately five times the pipe diameter from the mixer inlet, to ensure flow field stability and measurement accuracy. Specifically, this module includes an SO2 concentration monitor using the principle of ultraviolet differential absorption spectroscopy, used to measure the SO2 concentration in the inlet flue gas in real time, in mg / m³. 3 A thermal gas mass flow meter is used to measure the total volumetric flow rate of flue gas, with units of m³ / s. 3 / h; A PT100 platinum resistance temperature sensor is used to measure the flue gas temperature, in °C; A resistive-capacitive humidity transmitter is used to measure the absolute or relative humidity of the flue gas. The analog output signals (4-20mA) or digital communication signals of the above instruments are accessed to the data acquisition layer 204 of the control system 203 via an industrial Ethernet.

[0024] The acquisition of mixed ash state parameters is achieved through the mixer monitoring module 208, which is installed at the mixer outlet chute or on the mixer sidewall. This module is used to monitor the physicochemical properties of the circulating mixed ash entering the reactor in real time. Specifically, it includes: a near-infrared moisture meter, installed above the mixer outlet chute, with an automatic cleaning device to prevent dust adhesion, used to measure the moisture content of the mixed ash in real time, typically within a range of 0–20% with an accuracy of ±0.2%; and an online calcium content analyzer, which can be based on X-ray fluorescence or titration principles, installed on the mixer sidewall, with a sampling probe that is automatically cleaned periodically, used to measure the percentage of available Ca(OH)2 in the mixed ash in real time. The output signals of these instruments are also connected to the data acquisition layer 204.

[0025] The fluidized bed silo level parameters are acquired through the silo monitoring module 209, which is installed at the top of the silo. This module uses a guided wave radar level gauge or a non-contact radar level gauge, with the antenna extending downwards into the silo. It measures the level of circulating ash in the silo in real time, in meters or as a percentage of the silo's range. This measurement is used to determine the ash accumulation status of the silo and is a crucial basis for ash discharge decisions.

[0026] The acquisition of outlet emission parameters is achieved through the emission monitoring module 210. This module communicates with the existing CEMS system on the chimney, and reads the net flue gas SO2 concentration obtained from the CEMS system in real time via protocols such as RS485, Modbus TCP, or OPCUA, in units of mg / m³. 3 This parameter serves as the final control target and is used in the feedback correction process.

[0027] All collected raw multi-source parameters, including inlet SO2 concentration, flue gas flow rate, flue gas temperature, flue gas humidity, mixed ash moisture content, effective Ca(OH)2 content of mixed ash, fluidized bed material level, and outlet SO2 concentration, are aggregated to the data acquisition layer 204 of the control system 203 via industrial Ethernet at a 10-second cycle and stored in a real-time database for subsequent data preprocessing and model decision-making.

[0028] S102: Input the multi-source parameters into a pre-trained machine learning model, and the machine learning model outputs the set values ​​of the new ash feed rate, the circulating ash feed rate and the water spray rate; Specifically, the data acquisition layer 204 first reads the latest set of multi-source parameters collected and stored in step S101, including inlet SO2 concentration, flue gas flow rate, flue gas temperature, flue gas humidity, mixed ash moisture content, effective Ca(OH)2 content in mixed ash, fluidized bed silo level, and outlet SO2 concentration. To improve the model's prediction accuracy and generalization ability, this step also performs data preprocessing operations before model inference, including outlier removal, data smoothing, normalization, and construction of composite features.

[0029] The specific data preprocessing process is as follows: First, the 3σ criterion is used to identify outliers in the real-time measurements of each parameter. This involves calculating the mean and standard deviation of the most recent 100 historical data for each parameter. If the current value exceeds the range of mean ± 3 times the standard deviation, it is considered an outlier and replaced with the valid value from the previous moment. Second, a moving average filtering algorithm is used to smooth the continuously collected data. The sliding window size is set to 5 to eliminate interference from sensor noise and instantaneous fluctuations. Then, all input parameters are normalized, linearly mapping the values ​​of each parameter to the interval between 0 and 1. The normalized maximum and minimum values ​​are automatically updated every 24 hours based on the historical data distribution. Finally, after completing the preprocessing of the basic parameters, multiple composite features with physical meaning are further constructed to enhance the input information of the model. The composite features include at least one or more of the following three: SO2 load, calculated by multiplying the inlet SO2 concentration and flue gas flow rate, in mg / s, used to characterize the total amount of SO2 entering the system per unit time; calcium-sulfur ratio estimate, calculated based on the new ash feed rate, effective Ca(OH)2 content, and SO2 load, used to roughly reflect the current desulfurizer dosage margin; and moisture content deviation, which is the difference between the current mixed ash moisture content and the target moisture content, with the target moisture content preset to 5% based on operating conditions such as flue gas temperature or dynamically calculated using empirical formulas. After the above preprocessing, a complete model input feature vector is formed, which contains 7 directly measured parameters and at least 3 constructed composite features, totaling 10 dimensions.

[0030] The machine learning model is deployed in the model decision layer 205 of the control system 203. In this embodiment, the machine learning model is constructed using the gradient boosting decision tree algorithm, which has good nonlinear fitting ability and robustness, and is suitable for handling complex multivariate mapping relationships in industrial processes. The model training process adopts a combination of offline training and online incremental updates. In the initial stage of system commissioning, 30 consecutive days of historical operating data are collected as the initial training set. The data covers different load conditions and operating states. After removing data from abnormal periods such as downtime and failures, approximately 200,000 valid samples are finally obtained. During training, the input feature vector X of the model consists of the aforementioned 10-dimensional parameters, and the output vector Y of the model consists of the setpoints of three control signals, namely the feeder frequency corresponding to the new ash feed rate, the feeder frequency corresponding to the circulating ash feed rate, and the regulating valve opening corresponding to the water spray rate. The model training uses mean squared error as the loss function, and the optimal hyperparameters are determined through grid search, including 300 trees, a learning rate of 0.05, a maximum tree depth of 6, and a subsampling ratio of 0.8. After training, the model is exported as PMML format and deployed to the real-time control program of the industrial control computer.

[0031] During real-time operation, within each 10-second control cycle, the model's decision layer 205 receives the latest pre-processed input feature vector, invokes the deployed gradient boosting decision tree model for forward inference calculation, and the model immediately outputs three control setpoints. Simultaneously, to maintain the model's adaptability to changes in operating conditions, the system automatically performs incremental training every 24 hours, retraining or fine-tuning the model using newly added operating data from the past 24 hours and automatically replacing the old model. The training process is scheduled during low-load periods at night, without affecting normal daytime control. Through this method, the model achieves a precise mapping from multi-source input parameters to key control setpoints, providing a decision-making basis for subsequent actuator adjustments.

[0032] S103: Adjust the new ash feeder, circulating ash feeder and water spray regulating valve respectively according to the set values; Specifically, the control execution layer 206 is deployed in an industrial control computer or PLC, and communicates with the field actuators in real time via PROFINET or OPCUA protocols. First, the control execution layer 206 receives three setpoints from the model decision layer 205: the frequency setpoint for the new ash feeder, the frequency setpoint for the circulating ash feeder, and the opening settingpoint for the water spray regulating valve. To ensure the reliability of the control commands, the control execution layer 206 performs a rationality check on the setpoints before execution, including whether they are within the preset safety range, whether they exceed the physical adjustment limits of the actuators, and whether the rate of change of the setpoints in adjacent cycles is too large. If the check passes, the setpoints are written to the corresponding registers of the PLC according to the control cycle; if the check fails, the valid value of the previous cycle is maintained, and a data anomaly alarm is triggered.

[0033] The feed rate of quicklime is adjusted via a quicklime feeder. This feeder is a variable frequency screw feeder, whose control terminal is connected to the analog output module of the PLC, receiving 4–20mA current signals or digital communication commands. The control execution layer 206 converts the setpoint for the quicklime feed frequency into a corresponding analog output value. The PLC adjusts the output frequency of the frequency converter based on this output value, thereby changing the speed of the screw feeder and achieving precise control over the feed rate of quicklime or hydrated lime. The frequency adjustment range is typically 5–50Hz, corresponding to a feed rate of 0–5t / h. Simultaneously, the PLC reads the actual operating frequency feedback from the frequency converter and compares it with the setpoint, forming a closed-loop verification to ensure accurate feed rate.

[0034] The amount of circulating ash fed is adjusted via a circulating ash feeder. This feeder also uses a variable frequency speed-regulating screw feeder, and its control method is similar to that of the new ash feeder. The control execution layer 206 converts the circulating ash feeding frequency setpoint into an analog signal and sends it to the PLC. The PLC adjusts the frequency of the frequency converter to control the amount of circulating ash added. The frequency adjustment range is typically 5–50Hz, corresponding to a feed rate of 0–20 t / h. The amount of circulating ash added directly affects the solid-to-gas ratio and reaction contact area of ​​the mixed ash, therefore precise adjustment is required. The PLC monitors the operating status and feedback frequency of the frequency converter in real time to ensure proper execution.

[0035] The water spray volume is adjusted via a water spray regulating valve and atomizing nozzles. The water spray regulating valve is an electrically operated ball valve that receives 4–20mA analog control signals, and its opening degree can be continuously adjusted within the range of 0–100%. The control execution layer 206 converts the water spray regulating valve opening setpoint into an analog output to the PLC. The PLC drives the electric actuator to change the valve opening degree, thereby regulating the water flow into the atomizing nozzles. The atomizing nozzles are dual-fluid atomizing nozzles that utilize compressed air to atomize water into fine particles, which are then thoroughly mixed with circulating ash in the mixer to adjust the moisture content of the mixed ash. To ensure atomization effect, the system is also equipped with an air supply pressure monitoring and adjustment device to ensure stable atomizing air pressure.

[0036] During the aforementioned actuator adjustment process, the PLC is also responsible for executing the underlying safety interlock logic. For example, when the new ash feeder is running, if a downstream equipment malfunction or abnormal material level is detected, the PLC can immediately cut off the feeder's power supply; if the water spray regulating valve opening is too large, causing the mixed ash moisture content to exceed 15%, the PLC can forcibly close the valve and issue an alarm. Furthermore, the actual feedback values ​​from all actuators are transmitted back to the industrial control computer via the data acquisition layer 204 for real-time monitoring and subsequent analysis and optimization. Through this method, the model's output settings can be accurately translated into physical actions, achieving coordinated and precise adjustment of new ash, circulating ash, and water spray volume.

[0037] Based on the completion of the main control loop (steps S101 to S103) described above, this invention also executes a parallel intelligent ash removal control step to automatically decide the timing and amount of ash removal from the fluidized bed silo, in order to prevent effective calcium waste or silo overflow. This step operates independently of the main control cycle and is typically executed at a low frequency, such as once per minute, to ensure the stability of the ash removal action.

[0038] In practical implementation, the control system 203 first acquires the latest effective Ca(OH)2 content data of the mixed ash measured by the mixer monitoring module 208 and the fluidized bed silo level data measured by the silo monitoring module 209 in real time through the data acquisition layer 204. These two data sets have already been acquired and preprocessed in step S101 and can be directly used for subsequent logical judgments. The decision logic for intelligent ash discharge control is based on the combination of effective calcium content and ash level, specifically including the following four scenarios: When the effective Ca(OH)2 content is lower than a preset first threshold and the fluidized bed silo level is higher than a preset second threshold, it indicates that the effective components in the accumulated circulating ash in the silo are insufficient, but the ash level is too high. At this point, some inefficient ash needs to be discharged to make room for fresh desulfurizer. The control system 203 sends a start command to the ash discharge device, the pneumatic gate valve opens, and the rotary unloader operates at a preset frequency for a first preset duration, such as 20 minutes. After ash discharge, the system waits a short time for the ash level to stabilize, and then re-acquires the effective calcium content and ash level data. If the detection results still meet the condition that the effective calcium content is lower than the first threshold and the ash level is higher than the second threshold, then ash discharge continues, and this cycle repeats until the state returns to normal.

[0039] When the effective Ca(OH)2 content is below the first threshold but the fluidized bed silo level is not above the second threshold, it indicates that the overall effective calcium in the system is insufficient, but the silo level is still within a safe range. If ash discharge is initiated at this time, it will further waste effective components. Therefore, the control system 203 does not initiate ash discharge but instead issues an audible and visual alarm for "insufficient effective calcium," reminding operators to check the quality of the new ash or increase the amount of new ash fed. Simultaneously, the system can automatically prioritize increasing the amount of new ash fed to replenish effective calcium, and then resume normal control once the effective calcium content recovers.

[0040] When the fluidized bed silo level exceeds the preset third threshold, regardless of the effective calcium content, it indicates a risk of silo overflow. The third threshold is higher than the second threshold, for example, set to 95% of full capacity. At this time, the control system 203 immediately forces the ash removal device to start and issues a "silo overflow warning" alarm until the material level drops below the safe range to ensure equipment safety.

[0041] Through the intelligent ash removal logic based on the combination of multiple parameters, the present invention achieves effective management of materials in the fluidized bed silo, which not only avoids the waste of effective calcium, but also eliminates the risk of silo overflow, further improving the system's operational stability and economy.

[0042] S104: Based on the deviation between the measured value and the target value of the outlet emission parameters, the set value is corrected by feedback.

[0043] Specifically, the control system 203 triggers a feedback correction process every first preset cycle. In this embodiment, the first preset cycle is set to 5 minutes, matching the regular analysis cycle of the CEMS system. At the end of each correction cycle, the control execution layer 206 reads the latest outlet SO2 concentration value measured by the chimney CEMS system through the emission monitoring module 210. Simultaneously, it obtains the corresponding inlet SO2 concentration from the real-time database and calculates the current desulfurization efficiency using the following formula: Desulfurization efficiency = (Inlet SO2 concentration - Outlet SO2 concentration) / Inlet SO2 concentration × 100%. The calculated desulfurization efficiency is compared with a preset target threshold. In this embodiment, the preset threshold is set to 95%, meaning the desulfurization efficiency is required to be no less than 95%.

[0044] If the current desulfurization efficiency is higher than or equal to the preset threshold, it indicates that the system is operating normally and no correction is needed. The control execution layer 206 maintains the original model output settings and will reassess in the next correction cycle. If the current desulfurization efficiency is lower than the preset threshold, a feedback correction procedure is initiated to dynamically adjust the set values ​​of the new ash feed rate, circulating ash feed rate, and water spray rate output by the model in step S102 until the desulfurization efficiency recovers to above the preset threshold. The specific implementation methods of feedback correction include the following two, which can be selected or combined according to the operating conditions and the degree of deviation.

[0045] The first correction method is incremental compensation based on a PID control algorithm. The control execution layer 206 uses the desulfurization efficiency deviation as input to the PID controller, performs proportional-integral-derivative (PI-DI) calculations, and outputs a compensation increment. This compensation increment acts on the setpoint of the new ash feed rate, adding a correction amount to the original setpoint of the new ash feed frequency to increase the dosage of desulfurizing agent. The magnitude of the compensation amount is related to the degree of deviation; the proportional coefficient Kp is empirically set to 0.2, meaning that for every 1% decrease in desulfurization efficiency below the target value, the new ash feed frequency increases by 0.2 Hz. Simultaneously, to maintain stable moisture content in the mixed ash, the water spray volume is also adjusted proportionally to prevent the mixed ash from drying out due to the increase in new ash. The PID-compensated setpoint is sent from the control execution layer 206 to the execution system 202 and remains effective in subsequent control cycles until the desulfurization efficiency reaches the target. This method is suitable for situations with small deviations or short-term fluctuations, offering fast response and simple calculation.

[0046] The second correction method is online updating of the machine learning model. When the desulfurization efficiency is lower than a preset threshold and fails to improve after several correction cycles, or when the deviation is large, the control system 203 triggers the online model update mechanism. The control execution layer 206 extracts the operating data from the real-time database for the most recent period. In this embodiment, historical data from the most recent hour is selected as training samples. These samples cover the operating status near the current operating conditions. The original machine learning model is rapidly incrementally trained using this new data to generate a temporary model, which temporarily replaces the currently deployed model for subsequent control decisions. The temporary model is used until the desulfurization efficiency returns to normal, or until it is overwritten during the next regular daily model update. This method is suitable for situations where the operating conditions undergo structural changes or the model predictions show systematic deviations. It can fundamentally correct the model's output and has stronger adaptability.

[0047] In practical applications, the two correction methods can be used in combination. For example, PID compensation can be used for fast response first. If the desulfurization efficiency still fails to meet the standard after three consecutive correction cycles, the model will be automatically updated online. At the same time, regardless of the correction method used, all feedback correction instructions are adjusted by the execution system 202 described in step S103, including adjusting the frequency of the new ash feeder, the frequency of the circulating ash feeder, and the opening of the water spray regulating valve, forming a complete closed-loop control circuit.

[0048] Through the above-mentioned feedback correction mechanism, the present invention can effectively overcome the control lag problem, correct the system operation deviation in a timely manner, and ensure that the desulfurization efficiency remains stable above the target value under the circumstances of sudden changes in flue gas parameters or operating condition drift, so as to achieve long-term stable compliance of SO2 emission concentration.

[0049] The second embodiment of this application is as follows: Based on the first embodiment, please refer to Figure 2 The multi-parameter feedback NID desulfurization device precision automatic control system of this embodiment includes a detection system 201, an execution system 202 and a control system 203. The control system 203 includes a data acquisition layer 204, a model decision layer 205 and a control execution layer 206. The detection system 201 includes a flue gas pre-monitoring module 207, a mixer monitoring module 208, a silo monitoring module 209 and an emission monitoring module 210.

[0050] In this specific embodiment, the control system 203 is connected to the detection system 201 and the execution system 202 respectively; The detection system 201 is used to collect multi-source parameters in real time, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters. The execution system 202 includes a new ash feeder, a circulating ash feeder, a water spray regulating valve, and an ash discharge device, which are used to adjust their respective actions according to control commands. The data acquisition layer 204 is connected to the detection system 201 and is used to receive and preprocess the multi-source parameters acquired by the detection system 201. The model decision layer 205 is connected to the data acquisition layer 204 and is used to output the set values ​​of new ash feed rate, circulating ash feed rate and water spray rate through machine learning model based on the preprocessed multi-source parameters. The control execution layer 206 is connected to the model decision layer 205 and the execution system 202 respectively, and is used to convert the set value into a control command and send it to the execution system 202, and to provide feedback correction to the set value based on the deviation between the measured value and the target value of the outlet emission parameter.

[0051] Among them, the flue gas pre-monitoring module 207 is installed in the reactor inlet flue and includes an SO2 concentration monitor, a flue gas flow meter, a temperature sensor and a humidity sensor. The output of each instrument is communicatively connected to the input of the data acquisition layer 204. The mixer monitoring module 208, installed at the mixer, includes a near-infrared moisture meter and an online calcium content analyzer. The output of each instrument is communicatively connected to the input of the data acquisition layer 204. The silo monitoring module 209 is installed on the top of the fluidized bed silo and includes a radar level gauge. Its output is communicatively connected to the input of the data acquisition layer 204. The emission monitoring module 210 is communicatively connected to the chimney CEMS system, and its output is communicatively connected to the input of the data acquisition layer 204.

[0052] Using a multi-parameter feedback NID desulfurization device precision automatic control system according to this embodiment, the system hardware deployment is first completed, including installing an SO2 concentration monitor, flue gas flow meter, temperature sensor, and humidity sensor at the reactor inlet; installing a near-infrared moisture meter and an online calcium content analyzer at the mixer; installing a radar level gauge at the top of the fluidized bed; and connecting the chimney CEMS system to the industrial control computer via a communication interface. At the same time, a new ash feeder, a circulating ash feeder, a water spray regulating valve, and an ash discharge device are configured as actuators, with a Siemens S7-1500 series PLC as the bottom controller and the industrial control computer as the upper-level decision core, and data interaction is realized through industrial Ethernet.

[0053] After the system is put into operation, the control program executes the main control flow according to a fixed 10-second cycle. The data acquisition layer 204 first reads the raw parameters uploaded in real time by each monitoring instrument, such as inlet SO2 concentration, flue gas flow rate, temperature, humidity, mixed ash moisture content, effective Ca(OH)2 content, fluidized bed material level, and outlet SO2 concentration. It then performs outlier removal, data smoothing, normalization, and composite feature construction to form a complete input feature vector. This feature vector is fed into the gradient boosting decision tree model deployed in the model decision layer 205. After forward inference calculation, the model immediately outputs the set values ​​for the new ash feeding frequency, the circulating ash feeding frequency, and the water spray regulating valve opening. After receiving these set values, the control execution layer 206 sends them to the PLC via the PROFINET protocol. The PLC then adjusts the operating frequencies of the new ash variable frequency screw feeder and the circulating ash variable frequency feeder, as well as the opening of the water spray electric regulating valve, to achieve real-time control of the desulfurization process.

[0054] While the main control cycle is running, the system executes two auxiliary control logics in parallel. The first is a feedback correction mechanism triggered every 5 minutes: the control execution layer 206 reads the measured SO2 concentration at the CEMS outlet, calculates the current desulfurization efficiency, and if it is lower than the preset target of 95%, it initiates PID incremental compensation or triggers online model updates to dynamically correct the setpoint of the main control cycle until emissions meet standards. The second is intelligent ash discharge control executed every minute: the control system 203 performs a combined state judgment based on the real-time acquired effective Ca(OH)2 content and fluidized bed silo level. When the effective calcium content is lower than 10% and the level is higher than 80%, it automatically discharges ash for 20 minutes; when the level is higher than 95%, it forces ash discharge and triggers an alarm; when the effective calcium is insufficient but the level is not high, it prioritizes replenishing new ash and issues an alarm, thereby achieving precise management of the silo material.

[0055] Through the coordinated operation of the above-mentioned multi-cycle and multi-level control logic, this system realizes full-process automatic closed-loop control from inlet detection, model decision-making, execution adjustment to feedback correction, effectively solving the problems of lagging control, parameter mismatch and rough ash discharge in traditional NID desulfurization, and ultimately achieving the goal of improving desulfurization efficiency, reducing material consumption and enhancing operational stability.

[0056] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A precise automatic control method for NID desulfurization units with multi-parameter feedback, characterized in that, Includes the following steps: Real-time acquisition of multi-source parameters, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters; The multi-source parameters are input into a pre-trained machine learning model, and the machine learning model outputs set values ​​for the new ash feed rate, the circulating ash feed rate, and the water spray rate. Adjust the new ash feeder, circulating ash feeder, and water spray regulating valve according to the set values ​​respectively; The setpoint is corrected based on the deviation between the measured value and the target value of the export emission parameters.

2. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, Real-time acquisition of multi-source parameters, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters, specifically including: The inlet flue gas parameters include inlet SO2 concentration, flue gas flow rate, flue gas temperature, and flue gas humidity; the mixed ash state parameters include mixed ash moisture content and effective Ca(OH)2 content; the material level parameter is the material level height of the fluidized bed; and the outlet emission parameter is the outlet SO2 concentration.

3. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, The multi-source parameters are input into a pre-trained machine learning model, and the machine learning model outputs set values ​​for the new ash feed rate, the circulating ash feed rate, and the water spray rate, specifically including: The input features of the machine learning model also include composite features constructed based on the multi-source parameters. The composite features include at least one of SO2 load, calcium-sulfur ratio estimate, and moisture content deviation. The SO2 load is obtained by multiplying the inlet SO2 concentration and the flue gas flow rate. The calcium-sulfur ratio estimate is calculated based on the new ash feed rate, the effective Ca(OH)2 content, and the SO2 load. The moisture content deviation is the difference between the current moisture content and the target moisture content.

4. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, The feedback correction of the setpoint based on the deviation between the measured value and the target value of the export emission parameters specifically includes: The outlet SO2 concentration is acquired every first preset period, and the current desulfurization efficiency is calculated. If the current desulfurization efficiency is lower than the preset threshold, the set value is incrementally compensated according to the proportional-integral-derivative control algorithm, or the machine learning model is updated online using historical data from the most recent period, until the desulfurization efficiency recovers to above the preset threshold.

5. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, The method also includes intelligent ash removal control: The system can acquire the effective Ca(OH)2 content of the mixed ash and the material level in the fluidized bed silo in real time, and automatically decide the timing and amount of ash discharge based on the combination of effective Ca(OH)2 content and material level.

6. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 5, characterized in that, Intelligent ash removal control specifically includes: When the effective Ca(OH)2 content is lower than the first threshold and the material level is higher than the second threshold, the ash removal device is started and runs for the first preset time. After the ash removal is completed, retest. If the effective Ca(OH)2 content is still lower than the first threshold and the material level is still higher than the second threshold, continue ash removal until it returns to normal. When the effective Ca(OH)2 content is lower than the first threshold but the material level is not higher than the second threshold, an alarm for insufficient effective calcium is issued and the amount of new ash feed is increased first. When the material level is higher than the third threshold, the ash removal device is forcibly activated and a silo burst warning is issued; wherein the third threshold is higher than the second threshold.

7. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, The machine learning model is constructed using the gradient boosting decision tree algorithm and is incrementally trained using newly added running data according to the second preset period to maintain the model's prediction accuracy.

8. The precise automatic control method for a multi-parameter feedback NID desulfurization unit as described in claim 1, characterized in that, Before inputting the multi-source parameters into the pre-trained machine learning model, the method further includes: The raw data collected in real time is subjected to outlier removal, data smoothing and normalization, and composite features are constructed for model input.

9. A multi-parameter feedback NID desulfurization unit precision automatic control system, used to implement the method as described in claim 1, characterized in that, It includes a detection system, an execution system, and a control system, wherein the control system is connected to the detection system and the execution system, respectively; The detection system is used to collect multi-source parameters in real time, including inlet flue gas parameters, mixed ash state parameters, fluidized bed material level parameters, and outlet emission parameters. The execution system includes a new ash feeder, a circulating ash feeder, a water spray regulating valve, and an ash discharge device, which are used to adjust their respective actions according to control commands; The control system includes a data acquisition layer, a model decision-making layer, and a control execution layer; The data acquisition layer is connected to the detection system and is used to receive and preprocess the multi-source parameters acquired by the detection system. The model decision layer is connected to the data acquisition layer and is used to output the set values ​​of new ash feed rate, circulating ash feed rate and water spray rate through a machine learning model based on the preprocessed multi-source parameters. The control execution layer is connected to the model decision layer and the execution system respectively, and is used to convert the set value into control commands and send them to the execution system, and to provide feedback correction to the set value based on the deviation between the measured value and the target value of the outlet emission parameters.

10. The multi-parameter feedback precision automatic control system for NID desulfurization devices as described in claim 9, characterized in that, The detection system includes: The flue gas pre-monitoring module is installed in the reactor inlet flue and includes an SO2 concentration monitor, a flue gas flow meter, a temperature sensor, and a humidity sensor. The output of each instrument is communicatively connected to the input of the data acquisition layer. The mixer monitoring module, installed at the mixer, includes a near-infrared moisture meter and an online calcium content analyzer, with the output of each instrument communicating with the input of the data acquisition layer; The silo monitoring module, installed on the top of the fluidized bed silo, includes a radar level gauge, the output of which is communicatively connected to the input of the data acquisition layer. The emission monitoring module is communicatively connected to the chimney CEMS system, and its output is communicatively connected to the input of the data acquisition layer.

Images