Automatic control system for removing sulfur in flue gas

Abstract

The invention belongs to the field of artificial intelligence and automatic control. The system specifically comprises an in-furnace SO2 prediction model, an out-furnace SO2 prediction model, an in-furnace SO2 removal agent limestone control model, an out-furnace SO2 removal agent slaked lime control model, an in-furnace and out-furnace cooperation model and an early warning and alarm model for monitoring and analyzing the health state of a desulfurization system. The method comprises the following steps: firstly, acquiring real-time data of a desulfurization system associated control system (DCS); transmitting the collected data to an artificial intelligence algorithm system, and analyzing and calculating the data by using artificial intelligence and big data technologies; the calculated strategy is output to a DCS distributed control system, and field equipment is controlled to operate safely and efficiently; and finally, continuing to import the data of the last time period into the algorithm model unit, optimizing the removal process of the sulfur content in the flue gas by using a PID neural network and an operation optimization technology, and training through the model to realize self-optimization and self-adaptive adjustment.

Description

1. Technical Field

[0001] The present invention belongs to the field of artificial intelligence and automatic control, and specifically relates to the use of artificial intelligence and big data technology to study an automatic control system for removing sulfur from flue gas. 2. Background Technology

[0002] As a key link in environmental protection, industrial desulfurization plays a vital role in reducing air pollution and improving environmental quality. With the acceleration of global industrialization, the amount of sulfur oxides and dust emitted by industry continues to increase, posing a serious threat to the environment and human health. Therefore, the research and development and application of desulfurization technology is particularly important.

[0003] Desulfurization technologies are mainly divided into three categories: wet, semi-dry and dry. Among them, semi-dry desulfurization is a technology that combines the advantages of wet and dry desulfurization. Its principle is mainly to use alkaline desulfurizers, such as CaO or Ca(OH)2, to react with acidic gases such as SO2 in flue gas to generate solid desulfurization products.

[0004] With the continuous improvement of environmental protection requirements, semi-dry desulfurization technology is also constantly innovating and developing. However, the research and development and application of desulfurization technology still face many challenges. On the one hand, the research and development of desulfurization technology is difficult and costly, and it requires continuous investment in research and development for technological innovation; on the other hand, some companies do not pay enough attention to the desulfurization system, which also brings certain difficulties to technological innovation research. Therefore, in order to improve the efficiency and safety of industrial production desulfurization, the present invention constructs a model through artificial intelligence and big data analysis and calculation, combines the industrial automation of enterprise industrial production, and realizes intelligent analysis and control of the production process. III. Summary of the invention

[0005] The present invention provides an automatic regulation system for the desulfurization process. The energy consumption characteristics and optimization potential of the desulfurization process are studied through big data analysis and artificial intelligence technologies. The energy consumption distribution and energy-saving space conclusions of the desulfurization system are obtained through data analysis and accurate prediction. The intelligent optimization control system is developed with emphasis, and the intelligent regulation algorithm of the desulfurization process is studied (see the research route for details). Figure 1 ), and obtain the optimal energy-saving control key technology of the desulfurization system. Through the research of this project, a reference is provided for the application of industrial energy conservation and emission reduction and green production technology, which is of great significance to promoting the efficient utilization of industrial energy and technological innovation of industrial production. The present invention includes the following contents:

[0006] 1. The sulfur dioxide prediction model inside the furnace and the sulfur dioxide prediction model outside the furnace (see Figure 2 )

[0007] The sulfur dioxide prediction model in the furnace is the core component of the automatic control system. The model uses historical data and real-time data to predict the amount of sulfur dioxide generated during boiler combustion through artificial intelligence algorithms. The accuracy of the prediction directly determines the rationality of the subsequent desulfurization agent dosage, which in turn affects the desulfurization efficiency and operating costs.

[0008] The prediction model for sulfur dioxide outside the furnace is similar to the prediction model for sulfur dioxide inside the furnace, but the input variables and prediction targets are different. The prediction model for sulfur dioxide outside the furnace mainly focuses on the removal process of flue gas in the desulfurization tower. The input variables include sulfur dioxide concentration at the inlet of the desulfurization tower, sulfur dioxide concentration at the outlet of the desulfurization tower, desulfurization agent dosage, desulfurization tower temperature, etc. The prediction target is the sulfur dioxide concentration at the outlet of the desulfurization tower.

[0009] (1) Data collection and preprocessing

[0010] Data collection includes boiler operating parameters (such as coal quality, coal quantity, air volume, furnace temperature, etc.), flue gas composition (such as oxygen content, carbon dioxide content, etc.) and desulfurization system operating status (such as desulfurizer dosage, desulfurization efficiency, etc.). These data are monitored in real time by sensors and transmitted to the data center.

[0011] In the data preprocessing stage, the data is first cleaned to remove outliers and missing values; then normalization is performed to convert data of different dimensions to the same scale; finally, feature selection is performed to screen out characteristic variables that have a significant impact on sulfur dioxide generation.

[0012] (2) Model construction

[0013] A sulfur dioxide prediction model was constructed using a long short-term memory network (LSTM) based on deep learning (see Figure 3 LSTM architecture). LSTM is a special recurrent neural network (RNN) that can handle long-term dependencies in time series data and is suitable for predicting the amount of sulfur dioxide generated in flue gas.

[0014] The model input is the preprocessed feature variable time series, and the output is the predicted sulfur dioxide generation. The model is trained by training the data set and adjusting the model parameters so that the model can accurately predict the sulfur dioxide generation.

[0015] (3) Algorithm optimization

[0016] In order to improve the prediction accuracy and generalization ability of the model, the following algorithm optimization strategy is adopted:

[0017] 1) Regularization: Add a regularization term to the loss function to prevent the model from overfitting.

[0018] 2) Dropout: Randomly discard some neurons during training to improve the robustness of the model.

[0019] 3) Batch normalization: Normalize each batch of data to speed up the model training process and improve model stability.

[0020] 4) Learning rate adjustment: Adopt an adaptive learning rate adjustment strategy, such as the Adam optimizer, to dynamically adjust the learning rate according to the gradient changes during model training.

[0021] (4) Evaluation of prediction results

[0022] The prediction results are evaluated using indicators such as mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). By comparing the difference between the actual value and the predicted value, the model parameters are continuously optimized to improve the prediction accuracy.

[0023] 2. Limestone control model for sulfur dioxide removal agent inside the furnace and slaked lime control model for sulfur dioxide removal agent outside the furnace

[0024] The control model of limestone as sulfur dioxide removal agent in the furnace dynamically adjusts the amount of limestone according to the results of the sulfur dioxide prediction model in the furnace to achieve efficient removal of sulfur dioxide. The model comprehensively considers factors such as desulfurization efficiency, boiler combustion efficiency, operating costs, and the physical and chemical properties of limestone.

[0025] The slaked lime control model for off-furnace sulfur dioxide removal dynamically adjusts the amount of slaked lime added according to the results of the off-furnace sulfur dioxide prediction model. The model also adopts a control strategy based on fuzzy logic, taking the sulfur dioxide concentration at the desulfurization tower inlet, the predicted value of the sulfur dioxide concentration at the desulfurization tower outlet, the slaked lime inventory, the slaked lime price, etc. as input variables, and outputs the amount of slaked lime added.

[0026] (1) Control strategy (see Figure 4 )

[0027] A control strategy based on fuzzy logic is adopted, with the predicted value of sulfur dioxide in the furnace, limestone inventory, limestone price, and the impact of limestone addition on boiler combustion efficiency as input variables, and the output is the amount of limestone added. The fuzzy logic controller can obtain the fuzzy set of output variables through fuzzy reasoning based on the fuzzy set of input variables, and finally obtain the specific amount of limestone added through defuzzification.

[0028] (2) Fuzzy logic controller design

[0029] Fuzzification of input variables: Fuzzify input variables such as the predicted value of sulfur dioxide in the furnace, limestone inventory, and limestone price, and divide them into different fuzzy sets, such as "low", "medium", and "high".

[0030] Fuzzy rule base design: Design a fuzzy rule base based on expert experience and actual operation data. For example, when the predicted sulfur dioxide value in the furnace is high and the limestone inventory is sufficient, increase the amount of limestone added; when the limestone price is high, appropriately reduce the amount of limestone added to reduce costs.

[0031] Fuzzy reasoning: Based on the fuzzy set of input variables and the fuzzy rule base, fuzzy reasoning is performed to obtain the fuzzy set of output variables.

[0032] Defuzzification: Convert the fuzzy set of output variables into the specific amount of limestone. Common defuzzification methods include the maximum membership method and the centroid method.

[0033] (3) Control effect evaluation

[0034] By comparing the difference between the actual desulfurization efficiency and the set desulfurization efficiency, as well as the change in operating costs, the control effect of the control model is evaluated. According to the evaluation results, the fuzzy rule base and control strategy are continuously adjusted to improve control accuracy and economic benefits.

[0035] 3. Furnace and furnace coordination model

[0036] Furnace internal and external coordination model (see Figure 5 ) Comprehensively consider the boiler combustion efficiency, desulfurization system operation efficiency and desulfurization economy, optimize the amount of limestone and slaked lime added according to the purchase price of limestone and slaked lime and the desulfurization efficiency requirements, and realize the coordinated control of the desulfurization process inside and outside the furnace.

[0037] (1) Optimization goal

[0038] The optimization goals include improving desulfurization efficiency, reducing operating costs and reducing environmental pollution. Specifically:

[0039] 1) Desulfurization efficiency: Ensure that the sulfur dioxide concentration at the outlet of the desulfurization system meets environmental protection requirements.

[0040] 2) Operating cost: Under the premise of ensuring desulfurization efficiency, minimize the amount of desulfurizer added to reduce operating costs.

[0041] 3) Environmental pollution: Reduce environmental pollution problems such as solid waste and wastewater generated during the desulfurization process.

[0042] (2) Optimization algorithm

[0043] The amount of limestone and slaked lime is optimized by using multi-objective optimization algorithms, such as genetic algorithm and particle swarm optimization algorithm. During the optimization process, the objective functions such as desulfurization efficiency, operating cost and environmental pollution are converted into constraint conditions or penalty functions, and the optimal solution is found through iterative calculation.

[0044] (3) Collaborative control strategy

[0045] According to the results of the optimization algorithm, a coordinated control strategy for the furnace and the outside is formulated. When the boiler combustion efficiency is high and the sulfur dioxide concentration at the outlet of the desulfurization system is close to the set value, the amount of desulfurizer added is appropriately reduced; when the boiler combustion efficiency is low or the sulfur dioxide concentration at the outlet of the desulfurization system exceeds the standard, the amount of desulfurizer added is increased to improve the desulfurization efficiency. At the same time, according to the changes in the purchase price of limestone and slaked lime, the proportion of the two desulfurizers is flexibly adjusted to reduce operating costs.

[0046] 4. Early warning alarm model for monitoring and analyzing the health status of the desulfurization system

[0047] (1) Model Overview

[0048] The early warning alarm model for monitoring and analyzing the health status of the desulfurization system monitors and analyzes the operating status of the desulfurization system in real time, promptly discovers and warns of potential faults or abnormal conditions, and ensures the safe and stable operation of the desulfurization system.

[0049] (2) Monitoring indicators

[0050] Monitoring indicators include desulfurization efficiency, desulfurizer dosage, desulfurization tower pressure, desulfurization tower temperature, desulfurization tower liquid level, etc. These indicators can reflect the operating status and performance of the desulfurization system.

[0051] (3) Early warning and alarm mechanism

[0052] 1) Threshold alarm: Set the threshold range for each monitoring indicator, and trigger an alarm when the monitoring value exceeds the threshold range.

[0053] 2) Trend analysis alarm: Perform trend analysis on monitoring indicators and trigger an alarm when the monitoring values ​​show abnormal change trends.

[0054] 3) Correlation analysis alarm: Perform correlation analysis on multiple monitoring indicators and trigger an alarm when abnormal correlation occurs between the indicators.

[0055] (4) Early warning measures

[0056] According to the early warning alarm information, formulate corresponding treatment measures. For example, when the low efficiency alarm is triggered, increase the amount of desulfurizer or adjust the desulfurization tower operating parameters to improve the desulfurization efficiency; when the high pressure alarm is triggered, check whether the desulfurization tower inlet and outlet pipelines are blocked and clean them; when the leakage alarm is triggered, close the relevant valves and check whether the desulfurization tower and pipeline are leaking, etc. At the same time, the early warning alarm information is recorded and fed back to the relevant personnel so that timely measures can be taken to eliminate hidden dangers. IV. Description of the drawings

[0057] Figure 1 The system structure diagram can clearly show the various components of the system and their connections.

[0058] Figure 2 SO2 prediction model inside the furnace and SO2 prediction model architecture outside the furnace

[0059] Figure 3 LSTM Architecture

[0060] Figure 4 The framework of the control model of limestone as the SO2 removal agent inside the furnace and the control model of slaked lime as the SO2 removal agent outside the furnace

[0061] Figure 5 The collaborative model architecture of the furnace and outside the furnace

[0062] Please see the attached drawings of the specification for details. V. Specific implementation methods

[0063] The specific implementation methods of the present invention are as follows:

[0064] 1. Data collection implementation method

[0065] (1) Desulfurization system related data collection:

[0066] The research of this system requires the collection of historical and real-time data of boiler control system, limestone control system and absorption tower desulfurization system. The main purpose is to provide data support for sulfur generation prediction and desulfurization agent control through the correlation data of the desulfurization system. The detailed data collection points (different scenarios and different processes have different sample sizes, and the specific points are determined according to the actual situation) are shown in Table 1:

[0067] Table 1 System data

[0068]

[0069]

[0070]

[0071] (2) Data Flow

[0072] The flow path of data flow in the system architecture is as follows:

[0073] 1) The data collection layer transmits real-time operation data to the data center through sensors and instruments.

[0074] 2) The data processing layer pre-processes the received data and stores it in the database.

[0075] 3) The model prediction layer reads the preprocessed data from the database and uses deep learning algorithms to make predictions.

[0076] 4) The control decision layer formulates the control strategy based on the prediction results, using fuzzy logic control strategy and multi-objective optimization algorithm.

[0077] 5) The execution and feedback layer transmits the control strategy to the actuator of the desulfurization system and monitors the execution effect in real time.

[0078] 6) The execution results are returned to the data processing layer and model prediction layer through the execution and feedback layer for model optimization and updating.

[0079] 7) The user interface layer displays the system operation status, prediction results, control strategies, and early warning alarm information.

[0080] 2. Prediction model implementation

[0081] Considering the nonlinearity of real-time operating conditions in industrial production, boilers need to switch frequently and for a long time under a wide range of variable load conditions, and the units operate alternately in a "steady state-transition-steady state" mode, so there are a large number of non-steady state operating data in the operating data. In addition, the coal quality is variable. If conventional methods are used directly, even if the above-mentioned mechanism + data fusion method is used to accurately predict the quasi-steady state SO2 content, it is still impossible to cope with rapid and frequent start-stop and variable load conditions. The monitorable data is limited, and there are many disturbance factors, which cannot meet the control requirements; therefore, combined with deep residual learning and model online learning, based on CFB combustion and furnace and furnace reaction mechanism analysis, the load change of the boiler is determined to affect the boiler bed temperature, furnace wind speed, furnace pressure, excess air coefficient, oxygen content, circulation rate and other factors, and a high-precision adaptive SO2 generation and removal prediction model is established. The mechanism-data fusion model is modified to adapt to the changes in equipment, operation and environment under actual operating conditions, and the real-time concentration of SO2 at the desulfurization island inlet and the chimney inlet is directly corrected, thereby guiding accurate control strategies. This method is also the core advantage of artificial intelligence algorithms. The main algorithms involved include long short-term memory network (LSTM) and deep residual learning.

[0082] 3. Control model implementation method

[0083] At present, semi-dry desulfurization is a control system with multivariable coupling, large inertia and nonlinearity. Due to the complexity of the semi-dry desulfurization process, it is not only affected by important variables such as unit load, coal feed rate, bed temperature, oxygen content, flue gas flow rate, but also needs to consider the influence of unmeasurable disturbances (such as combustion lag, coal quality, desulfurization agent, flue gas measurement uncertainty). Conventional control strategies have been difficult to achieve good control effects. The present invention is based on real-time SO2 prediction results, constructs a multivariable disturbance suppression predictive control method that is adapted to peak-shaving and frequency-modulation tasks, adopts Kalman filter estimation and Bayesian probability weighting method, takes chimney inlet SO2 concentration, hourly mean SO2 concentration, hourly variance SO2 concentration, desulfurizer consumption, etc. as control targets, fully mines automatic control schemes and operator manual control schemes in historical data based on reinforcement learning, automatically integrates and coordinates the control strategies of desulfurizers inside and outside the furnace, determines the optimal desulfurization distribution ratio inside and outside the furnace, and finally achieves the suppression of SO2 emission disturbances of the desulfurization system when the coal quality changes or the unit changes load over a large range while meeting the requirements of SO2 emission standards, and reduces the energy consumption of the desulfurization system and the consumption of the desulfurizer as much as possible, thereby ensuring the efficient, economical and reliable operation of the desulfurization system.

[0084] (1) Non-discriminatory adaptive predictive control

[0085] Non-identification adaptive predictive control is essentially an adaptive control algorithm that does not require identification of system parameters, mathematical models of the control system, and specific test signals. This method only requires online detection of the actual output and expected output of the process to form an adaptive control system with good dynamic performance indicators.

[0086] (2) Multivariable disturbance rejection predictive control

[0087] The multivariable disturbance rejection model predictive control algorithm (DRMPC) is an improvement based on the general predictive control method, and is mainly divided into three parts: prediction model, rolling optimization and feedback correction. The multivariable disturbance rejection predictive control algorithm is different from the conventional predictive control in that it adopts an augmented state space model. On the basis of establishing a set of disturbance sub-models, it uses a weighted prediction model to perform output prediction by fitting a weighted prediction model based on the Bayesian theory, performs state estimation based on the Kalman filter, and finally solves the constrained optimization problem to achieve control of the desulfurization system.

[0088] (3) Reinforcement Learning Control Optimization

[0089] Reinforcement learning is an important field in machine learning, emphasizing how to act based on the environment to maximize expected benefits. Therefore, it is often used in optimal control to fully explore historical control data to select the best strategy to maximize expected benefits. The introduction of this algorithm can further optimize and improve the control robustness of the desulfurization system.

[0090] 4. Intelligent auxiliary analysis model for the health status of the desulfurization system

[0091] In the massive operating data recorded by the unit DCS (PLC), some measurement data in the operating data are missing or deviate from the normal range due to electromagnetic interference, abnormal measurement points, changes in operating conditions, etc., and corresponding processing is performed according to the importance of the information contained in the missing data. Abnormal data will directly lead to major deviations in data-driven methods or affect the reliability of prediction and control models. Data dimensionality reduction and clustering methods are used to screen out learnable operating conditions from historical operating data, and then physical reasonable operating data are further screened out based on manual experience to complete basic data cleaning work.

[0092] After data cleaning, it is also necessary to analyze the health status of the entire desulfurization process based on the status of each measuring point, such as the SO2 concentration monitoring results of each key measuring point (predicted values ​​and monitored values ​​at the desulfurization island inlet and chimney inlet), disturbance conditions within a certain time window length, and other monitorable operating parameters (including screw feeder status, limestone air supply duct pressure, air blower current, voltage and other data), to determine whether it is necessary to give an early warning signal for abnormal equipment operation.

[0093] Considering that the limestone feed of the furnace desulfurization is prone to blockage during operation, when limestone blockage occurs, the SO2 at the entrance of the desulfurization island will increase rapidly. The current monitoring system cannot directly determine the signal status of the blockage, and the missed judgment rate is high. It must be combined with manual judgment and processing, which leads to low reliability of the furnace desulfurization system. When considering blockage, there will be a problem of sudden and rapid increase in the pressure and current of the conveying fan. Therefore, the pressure of the measuring point and the model can be used to predict the concentration of SO2 at the entrance of the desulfurization island. Combined with the health status analysis results, pipeline blockage can be detected in advance, whether there is a fault, and whether manual intervention by the operator is required in advance.

[0094] The above is only a preferred specific implementation of the present invention; however, the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and its improved conception within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An automatic control system for removing sulfur from flue gas; its characteristics include Follow these steps: (1) Determine the data equipment parameter information related to the desulfurization system, including but not limited to the parameters of boiler equipment, limestone equipment, desulfurization tower equipment, and data points of boiler control system (DCS), limestone control system (DCS), desulfurization tower system (DCS), and flue gas system (DCS); (2) Realize real-time data collection through multiple communication protocols, including TCP / IP protocol, RS232 / RS485 serial communication protocol, HTTP / MQTT protocol and Modbus protocol; (3) The collected data is transmitted to the algorithm model unit, and the data is processed, analyzed and calculated using industrial Internet of Things, big data analysis technology, machine learning technology, PID neural network, and operations optimization technology to generate an optimization strategy for the sulfur removal process in flue gas; (4) Output the calculated optimization strategy to the DCS distributed control system through OPC or Modbus protocol to control the field equipment to operate safely and efficiently according to the optimization strategy; (5) The control system (DCS) data is collected in real time and entered into the algorithm model unit. The PID neural network and operations optimization technology are used to continuously optimize the sulfur removal process in the flue gas. Through model training, the self-optimization and adaptive adjustment function of the system is realized.

2. The adaptive adjustment system according to claim 1, characterized in that: The optimization strategy includes but is not limited to adjusting the operating frequency and feed flow rate of the control equipment and optimizing the operating parameters of the desulfurization tower.

3. The data information according to claim 1, characterized in that: Perform data alignment and data fusion (data-level fusion, feature-level fusion, decision-level fusion) on the data collected in real time.

4. The data processing and analysis according to claim 1, characterized in that: Perform statistical analysis and in-depth data analysis (principal component analysis, data cluster analysis, regression analysis) on the data.

5. The algorithm model unit according to any one of claim 1, characterized in that: Real-time optimization control function of input variables in the production process, in-depth analysis of data sources; use optimization control software modules (model data preprocessing, variable correlation analysis, control model creation, model parameter configuration, real-time optimization control, model adaptive adjustment) to solve the optimal solution of the optimization model, and with the help of DCS distributed control system, execute the optimal control strategy of the production process.

6. The self-optimizing adaptive regulation according to any one of claim 1, characterized in that: The PID neural network controller is used. After it is put into operation, it can complete real-time machine self-learning online without re-adjusting the controller parameters. It can effectively control key process variables, make the production process run stably for a long time, and improve the operating efficiency of the desulfurization system.

Images