Intelligent flushing control system and method for demister of thermal power plant

Abstract

The invention relates to the technical field of flue gas purification of thermal power plants, in particular to an intelligent flushing control system and method for a demister of a thermal power plant, the system comprises a sensing module, a decision and control module and an execution module, the sensing module is used for collecting running state parameters of a unit to which the current demister belongs; the decision and control module is in communication connection with the sensing module and is used for receiving the unit operation state parameters and generating a target flushing control instruction according to the unit operation state parameters; and the execution module is in communication connection with the decision and control module and is used for carrying out partition flushing operation on the current demister according to the target flushing control instruction. Therefore, through artificial intelligence prediction and optimization, the problems of response lag, control blindness and incapability of anticipating the blockage trend in a traditional demister washing control mode are solved, intelligent washing of the demister from passive treatment to active prevention is realized, and water consumption and power consumption are remarkably reduced while the pressure difference is stabilized.

Description

Technical Field

[0001] This application relates to the field of flue gas purification technology in thermal power plants, and in particular to an intelligent flushing control system and method for demisters in thermal power plants. Background Technology

[0002] Demisters are key equipment in wet desulfurization systems of thermal power plants, primarily used to separate particulate matter such as liquid droplets and gypsum slurry carried in flue gas, ensuring that the emission quality of clean flue gas meets standards. During operation, fine dust, sulfuric acid droplets, and other substances in the flue gas continuously accumulate on the surface of the demister blades and within the channels, reducing the gas flow area and causing an increase in the pressure difference across the demister, leading to increased system resistance. Excessive pressure difference in the demister not only increases the energy consumption of the induced draft fan and reduces the economic efficiency of unit operation, but long-term operation under high pressure difference can also lead to equipment structural fatigue, flue vibration, and even deformation. In severe cases, under conditions of sudden load increases, it can cause uncontrolled pressure difference, forcing the unit to reduce load or shut down urgently, posing a serious threat to the safe and stable operation of the power plant.

[0003] In related technologies, the flushing control of demisters in thermal power plants currently mainly adopts two methods: sequential control and differential pressure control. Sequential control performs periodic flushing based on fixed time intervals, which is simple to implement but lacks flexibility and cannot adjust the flushing strategy according to the actual blockage status, easily leading to water waste or insufficient flushing. Differential pressure control starts flushing when the measured differential pressure reaches a set threshold. Although it can respond to actual blockage, it is still a passive and lagging control method. Especially under transient conditions such as rapid load changes, it is difficult to effectively prevent the differential pressure from rising rapidly, which can easily lead to system parameter fluctuations and operational instability, and urgently needs to be addressed. Summary of the Invention

[0004] This application provides an intelligent flushing control system and method for demisters in thermal power plants, which solves the problems of response lag, blind control and inability to predict clogging trends in traditional demister flushing control methods. It realizes intelligent flushing of demisters from passive treatment to active prevention, and significantly reduces water and electricity consumption while stabilizing pressure differential.

[0005] A first aspect of this application provides an intelligent flushing control system for a demister in a thermal power plant, comprising: a sensing module, a decision-making and control module, and an execution module, wherein... The sensing module is used to collect the operating status parameters of the unit to which the current demister belongs; The decision and control module is communicatively connected to the sensing module. The decision and control module is used to receive the unit operating status parameters and generate target flushing control commands based on the unit operating status parameters. An execution module, which is communicatively connected to the decision and control module, is used to perform zoned flushing operations on the current demister according to the target flushing control command; The decision-making and control module includes: The differential pressure prediction unit is used to predict the differential pressure based on the unit's operating status parameters and historical operating status parameters according to a preset machine learning prediction model, and to obtain the differential pressure prediction result of the current demister within a preset time window. The flushing strategy optimization unit is used to generate a target flushing instruction based on a preset reinforcement learning decision model and a preset comprehensive optimization objective, when the differential pressure prediction result and / or the unit operating status parameters meet the preset flushing triggering conditions. The target flushing instruction includes at least one of flushing start time, flushing area, flushing duration and flushing water pressure.

[0006] According to one embodiment of this application, the reward function of the preset reinforcement learning decision model is: RF = -k1*ΔP - k2*W - k3*E; Where RF is the reward function, ΔP is the pressure difference, W is the flushing water consumption, E is the change in induced draft fan energy consumption, and k1, k2, and k3 are weighting coefficients.

[0007] According to one embodiment of this application, the sensing module includes: A differential pressure transmitter is used to measure the pressure difference between the inlet and outlet of the current demister; A data interface unit is used to obtain the unit's operating status parameters from the power plant's distributed control system or plant-level monitoring information system.

[0008] According to one embodiment of this application, the unit operating status parameters include at least one of the following: unit load, flue gas flow rate, flue gas temperature, fuel sulfur content, and ash content.

[0009] According to one embodiment of this application, the intelligent flushing control system for the demister in a thermal power plant further includes: The digital twin module is used to construct a virtual model corresponding to the current demister and to use the virtual model to simulate and verify the target flushing command, so that the target flushing control command can be sent to the execution module after the verification is successful.

[0010] According to one embodiment of this application, the preset machine learning prediction model employs a long short-term memory network or a gated recurrent unit.

[0011] According to the intelligent flushing control system for demisters in thermal power plants provided in this application embodiment, the decision and control module receives unit operating status parameters and inputs them into a preset machine learning prediction model for differential pressure prediction. The resulting differential pressure prediction is then used by a flushing strategy optimization unit. When the differential pressure prediction and / or unit operating status parameters meet preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. The execution module then performs zoned flushing operations on the current demister. Thus, through artificial intelligence prediction and optimization, the system solves the problems of response lag, blind control, and inability to proactively predict clogging trends in traditional demister flushing control methods. This achieves intelligent flushing of the demister from passive treatment to proactive prevention, significantly reducing water and electricity consumption while stabilizing the differential pressure.

[0012] A second aspect of this application provides an intelligent flushing control method for a demister in a thermal power plant, employing the aforementioned intelligent flushing control system for a thermal power plant demister, wherein the method includes the following steps: The sensing module is used to collect the unit operating status parameters to which the current demister belongs; The decision and control module receives the unit operating status parameters and inputs them into a preset machine learning prediction model. Based on the unit operating status parameters and historical operating status parameters, differential pressure prediction is performed to obtain the differential pressure prediction result of the current demister within a preset time window. Using the flushing strategy optimization unit, when the differential pressure prediction result and / or the unit operating status parameters meet the preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. Using the execution module, the current demister is subjected to zoned flushing operation according to the target flushing control command.

[0013] According to one embodiment of this application, the unit operating status parameters include the unit load, and the preset flushing trigger condition is: The rate of increase of the unit load is greater than the first preset threshold, and the differential pressure prediction result is less than or equal to the second preset threshold.

[0014] According to one embodiment of this application, after generating a target flushing instruction based on a preset reinforcement learning decision model and a preset comprehensive optimization objective, the method further includes: A virtual model corresponding to the current demister is constructed using a digital twin module; The target flushing command is simulated and verified using the virtual model, and after successful verification, the target flushing control command is sent to the execution module.

[0015] According to one embodiment of this application, after performing a zoned flushing operation on the current demister according to the target flushing control command, the method further includes: Collect system operation data after flushing the current demister; The system operation data is fed back to the decision and control module to optimize the model parameters of the preset machine learning prediction model and the preset reinforcement learning decision model.

[0016] According to the intelligent flushing control method for demisters in thermal power plants provided in this application, a decision and control module receives unit operating status parameters and inputs them into a preset machine learning prediction model for differential pressure prediction. The resulting differential pressure prediction is then used by a flushing strategy optimization unit. When the differential pressure prediction and / or unit operating status parameters meet preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. The execution module then performs zoned flushing operations on the current demister. Thus, through artificial intelligence prediction and optimization, the methods solve the problems of response lag, blind control, and inability to proactively predict clogging trends inherent in traditional demister flushing control methods. This achieves intelligent flushing of the demister, moving from passive handling to proactive prevention, significantly reducing water and electricity consumption while stabilizing the differential pressure.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a block diagram of an intelligent flushing control system for a demister in a thermal power plant, provided according to an embodiment of this application. Figure 2 This is a schematic diagram of the architecture of an intelligent flushing control system for a demister in a thermal power plant according to an embodiment of this application. Figure 3 This is a flowchart illustrating the decision and control module according to one embodiment of this application; Figure 4 This is a comparison graph of the pressure difference-time curves between intelligent preventive flushing according to an embodiment of this application and conventional methods; Figure 5 This is a schematic diagram comparing the differential pressure control effect of an intelligent flushing method according to an embodiment of this application with that of a conventional method when dealing with a surge in load; Figure 6This is a flowchart of an intelligent flushing control method for a demister in a thermal power plant according to an embodiment of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0020] Those skilled in the art will understand that the commonly used demister flushing control methods currently include the following two main types.

[0021] (1) Time-sequence control method: The flushing is started periodically according to a preset fixed time interval. This method is simple to implement, but its biggest drawback is its blindness. It cannot detect the actual degree of blockage of the demister, which may lead to over-flushing during cleaning, wasting water resources and plant power; or it may lead to insufficient flushing when the blockage is rapid, making it impossible to effectively control the pressure difference.

[0022] (2) Differential pressure control mode: Flushing is initiated when the measured differential pressure of the demister reaches a certain set value. This method is an improvement over sequential control and can respond to actual blockage conditions. However, it is essentially a passive and delayed response. Under transient conditions such as a surge in unit load, blockage develops rapidly. By the time the differential pressure rises to the activation value, the blockage is often quite severe. At this point, the flushing effect is poor, the time is long, and it is easy to cause drastic fluctuations in system parameters, making it impossible to prevent problems before they occur.

[0023] Therefore, existing technologies lack an intelligent method that can proactively predict clogging trends and dynamically formulate the optimal flushing strategy to cope with complex and ever-changing operating conditions and achieve a balance between safety, stability, energy saving, and environmental protection.

[0024] To overcome the shortcomings of existing technologies, this application provides an intelligent flushing control system and method for demisters in thermal power plants based on artificial intelligence. This system enables predictive maintenance of demister clogging and adaptive optimization of flushing strategies, thereby stabilizing system differential pressure and reducing operating costs.

[0025] The following describes, with reference to the accompanying drawings, an intelligent flushing control system and method for a demister in a thermal power plant, according to embodiments of this application.

[0026] Specifically, Figure 1 This is a block diagram illustrating an intelligent flushing control system for a demister in a thermal power plant, provided as an embodiment of this application.

[0027] like Figure 1As shown, the intelligent flushing control system for the demister in the thermal power plant includes a sensing module 100, a decision and control module 200, and an execution module 300.

[0028] The system includes a sensing module for collecting the operating status parameters of the unit to which the current demister belongs; a decision and control module that communicates with the sensing module for receiving the operating status parameters and generating target flushing control commands based on these parameters; and an execution module that communicates with the decision and control module for performing zoned flushing operations on the current demister according to the target flushing control commands.

[0029] In some embodiments, the unit operating status parameters include at least one of the following: unit load, flue gas flow rate, flue gas temperature, fuel sulfur content, and ash content.

[0030] Specifically, the sensing module in this application embodiment is used to collect real-time operating status parameters of the demister and its associated systems. For example, the sensing module monitors differential pressure changes in real time through differential pressure transmitter components installed at the inlet and outlet of the demister. At the same time, it interacts with the power plant's distributed control system or plant-level monitoring information system through a data interface unit to obtain multi-dimensional operating parameters, including unit load, flue gas flow rate, flue gas temperature, fuel sulfur content, and ash content, providing complete data support for subsequent intelligent decision-making.

[0031] The decision and control module in this embodiment is an AI decision and control module, which is used to receive and process operating status parameters and generate the optimal flushing strategy (i.e., target flushing control command) based on the built-in artificial intelligence model; the execution module is used to receive and execute the flushing strategy to perform precise zone flushing of the demister.

[0032] Furthermore, in some embodiments, the sensing module includes: a differential pressure transmitter for measuring the pressure difference between the inlet and outlet of the current demister; and a data interface unit for acquiring unit operating status parameters from the power plant's distributed control system or plant-level monitoring information system.

[0033] Specifically, the differential pressure transmitter is the core sensing device in the sensing module used to directly monitor the status of the demister itself. The differential pressure transmitter is precisely installed between the flue gas inlet and outlet of the demister to measure the pressure difference between the inlet and outlet in real time and continuously. This pressure difference data is the most direct and critical performance indicator characterizing the degree of demister blockage, and its changing trend directly reflects the severity of scaling and ash accumulation within the channel.

[0034] The data interface unit establishes a stable connection with the widely deployed distributed control system or plant-level monitoring information system in the power plant through standard communication protocols. Through the data interface unit, the system can seamlessly acquire macroscopic operating environment parameters that have a significant impact on the demister pressure difference, such as unit load characterizing boiler output level, flue gas flow rate reflecting flue gas volume flow rate, flue gas temperature affecting slurry evaporation and carryover, and fuel analysis data (such as fuel sulfur and ash data indicating the concentration of pollutants in the flue gas).

[0035] Furthermore, the decision-making and control module includes: a differential pressure prediction unit, used to predict differential pressure based on a preset machine learning prediction model, according to unit operating status parameters and historical operating status parameters, to obtain the differential pressure prediction result of the current demister within a preset time window; and a flushing strategy optimization unit, used to generate a target flushing instruction based on a preset reinforcement learning decision model and a preset comprehensive optimization objective, when the differential pressure prediction result and / or unit operating status parameters meet preset flushing trigger conditions, wherein the target flushing instruction includes at least one of flushing start time, flushing area, flushing duration, and flushing water pressure.

[0036] Specifically, the differential pressure prediction unit in this application embodiment is built based on machine learning algorithms. By deeply analyzing the real-time collected unit operating status parameters and long-term accumulated historical operating data, a nonlinear mapping relationship between the demister blockage state and multivariable operating conditions is established, thereby accurately predicting the demister differential pressure change trend within a future time window (such as the next 30-60 minutes).

[0037] Therefore, the embodiments of this application use machine learning models to predict differential pressure trends in advance, enabling preventive flushing to be initiated before the differential pressure increases significantly. This fundamentally avoids the problem of differential pressure runaway caused by transient conditions such as a surge in load, greatly improving the operational stability and reliability of the desulfurization system and the main unit, realizing predictive maintenance, and turning passive into proactive.

[0038] The flushing strategy optimization unit is built based on a reinforcement learning algorithm. When the predicted differential pressure exceeds the safety threshold, or when a sudden change in specific operating conditions such as a sharp increase in unit load is detected, the flushing strategy optimization unit immediately activates the decision engine based on the reinforcement learning algorithm. With the comprehensive optimization objectives of maintaining differential pressure stability, minimizing flushing water consumption, and reducing total system energy consumption, it dynamically generates the optimal flushing command, which includes flushing timing, flushing zones, flushing duration, and flushing water pressure, thereby achieving predictive flushing based on sudden changes in operating conditions.

[0039] In some embodiments, the reinforcement learning algorithm used by the rinsing strategy optimization unit has the following reward function: RF = -k1*ΔP - k2*W - k3*E; Where RF is the reward function, ΔP is the pressure difference, W is the flushing water consumption, E is the change in induced draft fan energy consumption, and k1, k2, and k3 are all weighting coefficients greater than zero.

[0040] Therefore, this application embodiment uses a reinforcement learning model to unify multiple sometimes conflicting objectives, such as reducing pressure differential, saving water, and reducing power consumption, under a single framework for optimization, i.e., multi-objective dynamic optimization. The system no longer simply performs flushing actions, but finds the globally optimal cost-effective flushing scheme through continuous interaction with the environment and self-learning, significantly reducing the power plant's operating costs and maximizing overall benefits.

[0041] Furthermore, in some embodiments, the machine learning algorithm (preset machine learning prediction model) used by the differential pressure prediction unit is a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU).

[0042] Furthermore, in some embodiments, the intelligent flushing control system for demisters in thermal power plants also includes: a digital twin module, used to construct a virtual model corresponding to the current demister, and to use the virtual model to simulate and verify the target flushing command, so as to send the target flushing control command to the execution module after the verification is passed.

[0043] Specifically, the intelligent flushing control system for demisters in thermal power plants in this application embodiment also includes a digital twin module. This digital twin module is used to construct a virtual model that corresponds to the physical demister system with high fidelity. The flushing instructions generated by the AI ​​decision and control module are first simulated and evaluated in this digital twin module. After the strategy is confirmed to be effective, they are then sent to the execution module.

[0044] Therefore, this application embodiment provides a highly realistic "sand table simulation" environment for AI decision-making by introducing a digital twin module. The proposed flushing strategy can be comprehensively verified in virtual space, potential risks can be identified in advance, and the flawless execution of actual actions can be ensured. This greatly reduces the safety risks that direct control may bring, enhances the trust of power plant operators in the intelligent system, and improves the safety and reliability of decision-making.

[0045] Furthermore, the execution module includes: zone-controlled electric or pneumatic flushing valves, each valve controlling a specific flushing area of ​​the demister. This allows the system to selectively initiate flushing procedures for corresponding areas based on the actual degree of blockage in different parts of the demister, avoiding the water waste caused by traditional overall flushing methods. The electric or pneumatic drive ensures rapid and reliable valve response, enabling precise execution of opening, closing, and adjustment commands, achieving accurate control of flushing duration and sequence. The execution module also includes a flushing water pump and main piping system to provide flushing water with predetermined pressure and flow rate, effectively removing demister blockages while maximizing water conservation, achieving a balance between safety and economy.

[0046] Furthermore, after the execution module performs zoned flushing operations on the current demister according to the target flushing control command, the system can also continuously collect actual effect data after flushing to form a closed-loop feedback, continuously iterate and optimize the AI ​​model, so that the system can adapt to complex external conditions such as coal type changes, equipment performance degradation, and seasonal changes, always maintain the best decision-making performance, and have continuous self-learning and self-adaptive capabilities.

[0047] To facilitate a clearer and more intuitive understanding of the AI-based intelligent flushing control system for thermal power plant demisters in this application, the following is a detailed explanation. Figures 2-5 This application provides a detailed description of the architecture, workflow, and differential pressure control effect of the intelligent flushing control system for demisters in thermal power plants when dealing with a surge in load.

[0048] First, such as Figure 2 As shown, the architecture of the intelligent flushing control system for the demister in this thermal power plant includes a sensing layer, an intelligent decision-making layer, and an execution layer. The sensing layer is responsible for data acquisition and includes components such as a differential pressure transmitter, an interface for acquiring unit load signals, flue gas parameter monitoring equipment, and a data interface unit. These components work together to collect all the real-time operating status parameters required by the system.

[0049] The intelligent decision-making layer receives data from the perception layer and processes it intelligently. First, the data is organized and saved by the data preprocessing and storage unit. The AI ​​model engine integrates machine learning prediction models and reinforcement learning optimization models to predict pressure differentials and generate optimal flushing strategies. Before issuing commands, the digital twin simulation unit verifies the strategy. Finally, the control command generation unit outputs specific control commands. Furthermore, the intelligent decision-making layer also receives feedback data from the execution results for continuous model optimization.

[0050] The execution layer is responsible for translating the instructions from the decision-making layer into physical actions. Upon receiving an instruction, the zone flushing controller drives specific valves in the flushing valve group and controls the flushing water pump to provide the required flushing water, thereby completing the precise flushing of specific areas of the demister.

[0051] Furthermore, such as Figure 3 As shown, the workflow of the AI ​​decision-making and control module includes the following steps: Step 1: Collect real-time differential pressure of the demister and unit operating status parameters to form a multi-dimensional real-time dataset, and perform data preprocessing and feature engineering on the raw data; Step 2: Input the processed real-time dataset into the trained machine learning prediction model to obtain the pressure difference prediction curve within the future time window; Step 3, (Intelligent Judgment): The prediction result triggers the system's early warning mechanism. The reinforcement learning decision engine is triggered to perform predictive flushing only if the unit load increase rate exceeds a critical value and the current predicted differential pressure does not exceed a preset threshold. If the predicted differential pressure exceeds the threshold but the load does not change drastically, the regular optimized flushing process is triggered. Step four: The reinforcement learning decision engine calculates the optimal flushing scheme, including flushing timing, flushing zones, flushing duration, and flushing water pressure, based on the current system state and comprehensive optimization objectives, and uses a digital twin model for simulation verification. Step 5: Convert the rinsing plan into control commands and send them to the execution layer to drive the execution module to perform precise rinsing on the specified area; Step six: Collect actual operational data after rinsing and send it as a feedback signal to the AI ​​decision engine for online learning and adaptive optimization of the model.

[0052] Furthermore, such as Figure 4 As shown, the horizontal axis (time axis) of the coordinate system marks the key time series from t0 to t4, representing: t0 (preventive flush): The moment when the method of the present invention initiates a pre-flushing.

[0053] t1 (load surge): The starting point when the unit load begins to rise sharply.

[0054] t2 (Conventional method starts flushing): The moment when flushing begins only after the conventional differential pressure control method detects that the differential pressure exceeds the limit.

[0055] t3 (peak pressure): The moment when the system pressure difference reaches its highest point.

[0056] t4 (System Stability): The moment when the system reaches stability again after flushing is complete.

[0057] Vertical axis (differential pressure axis): The unit is Pascal (Pa), indicating the differential pressure range from 0 to 300 Pa, and includes two key reference lines: Design differential pressure line (150 Pa): The ideal operating differential pressure of the demister under normal design conditions.

[0058] Alarm differential pressure line (250 Pa): The alarm threshold set by the system. If the differential pressure exceeds this line, it indicates that the operation has entered a dangerous condition, which may cause the equipment to trip.

[0059] Curve Comparison: Using the traditional method (represented by the dashed line), the differential pressure slowly increases from t0 to t1. After a surge in load at t1, the differential pressure rises sharply with a steeper slope, exceeding the alarm threshold at t2, at which point the system initiates flushing, resulting in a severely delayed response. At t3, the differential pressure reaches its peak (approximately 280 Pa), far exceeding the alarm threshold and posing a serious threat to system safety. After flushing, the differential pressure remains at a high level (260 Pa) at t4, recovering slowly.

[0060] The curve (represented by the solid line) of the method of this invention shows that at time t0, the system, based on AI prediction, performed a "preventative flush" in advance to prepare for the load increase. During the entire period of load surge from t1 to t3, the differential pressure only showed a gradual and slight increase. The curve was always controlled near the design differential pressure line (150 Pa), and remained far below the alarm differential pressure line (250 Pa) throughout. At time t4, the system had quickly recovered to its optimal stable state.

[0061] This comparative chart powerfully demonstrates that, compared to traditional passive and delayed control methods, the intelligent flushing system of this invention, through predictive maintenance and proactive intervention, can fundamentally prevent uncontrolled spikes in differential pressure during periods of rapid load increase. By consistently maintaining the system differential pressure within a safe and stable ideal range, it significantly improves the operational reliability and safety of the desulfurization system and main unit.

[0062] Furthermore, such as Figure 5 As shown, the horizontal axis represents time (t0-t4), and the vertical axis represents differential pressure (50Pa-300Pa). The design differential pressure line is located at approximately 150Pa, and the alarm differential pressure line is located at 250Pa. The differential pressure curve obtained using the traditional method shows a continuous increase in pressure as the load rises, only being forced to flush after t2 exceeds the alarm line. The peak pressure is close to 300Pa, far exceeding the design value, posing a high risk.

[0063] The method of this invention predicts the curve in advance and initiates preventive flushing near t1, keeping the pressure difference below 200Pa throughout the process, below the alarm line, and maintaining a stable range closer to the design value, thus achieving safe, energy-saving, and stable operation.

[0064] The intelligent flushing system of this application will be described below with specific examples.

[0065] Suppose that the intelligent flushing system of the present invention is implemented in the wet desulfurization system of a 660MW coal-fired power unit.

[0066] (1) System construction and model training stage.

[0067] Sensing module: Equipped with a high-precision differential pressure transmitter, it communicates with the power plant's SIS system via the OPC protocol to acquire data such as load, flue gas flow, and coal quality analysis (sulfur content, ash content) in real time.

[0068] AI Decision and Control Module: Deploy a high-performance industrial server. Collect historical operating data of the unit over the past two years (totaling over 1 million data points) to train the LSTM prediction model and the reinforcement learning model based on the PPO algorithm offline. The training objectives are: the LSTM model's error in predicting the pressure difference over the next 60 minutes should be less than 5%; the reinforcement learning model should achieve stable convergence.

[0069] Execution module: The original demister flushing system is transformed into 6 independently controlled zones (AF zones), with each zone equipped with a high-speed electric ball valve.

[0070] (2) Online operation phase (taking a typical working condition as an example).

[0071] S1 (Data Acquisition): The system is running in real time, collecting data showing the current differential pressure is 180Pa (design value is 150Pa), the current unit load is 500MW, and the received dispatch command indicates that the load will rise to 660MW within the next 40 minutes. Coal quality data shows that the current coal ash content is high.

[0072] S2 (Trend Prediction): After comprehensively analyzing the current data, the differential pressure prediction unit (LSTM model) outputs the prediction result: "It is expected that the differential pressure of the demister will exceed the alarm value of 250Pa in 35 minutes."

[0073] S3 (Intelligent Judgment): The prediction result triggered the system's early warning mechanism, and the rapid increase in load also met the conditions for starting reinforcement learning.

[0074] S4 (Strategy Generation): The flushing strategy optimization unit (reinforcement learning model) is activated. It performs rapid simulations in the digital twin module, taking into account water conservation and stability under high load, and generates the optimal solution: "It is recommended to flush the upper nozzles of zones A, C and F, which have the highest risk of ash blockage, in sequence at a pressure of 4.0 Bar after 20 minutes, for 75 seconds in each zone."

[0075] S5 (Precise Execution): After the operators confirm the plan, the system automatically and precisely executes the flushing command at the scheduled time.

[0076] Therefore, during the flushing process, the pressure differential only fluctuated slightly. After flushing, as the load increased, the pressure differential stabilized at around 200 Pa, successfully preventing it from exceeding the limit. Analysis showed that this flushing method saved approximately 30% of flushing water compared to traditional differential pressure control methods.

[0077] (3) Learning optimization stage.

[0078] The entire process of "prediction-decision-execution-result" was fully recorded and fed back to the AI ​​model as new training samples. The system utilized idle computing power to fine-tune the model online based on the new data, making its decisions more accurate for high-ash coal types under conditions of sudden load increases.

[0079] The intelligent flushing control system for demisters in thermal power plants proposed in this application utilizes a decision and control module to receive unit operating status parameters and input them into a preset machine learning prediction model for differential pressure prediction. The resulting differential pressure prediction is then used by a flushing strategy optimization unit. When the differential pressure prediction and / or unit operating status parameters meet preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. The execution module then performs zoned flushing operations on the current demister. Thus, through artificial intelligence prediction and optimization, the system solves the problems of response lag, blind control, and inability to proactively predict clogging trends inherent in traditional demister flushing control methods. This achieves intelligent flushing of the demister, moving from passive handling to proactive prevention, significantly reducing water and electricity consumption while stabilizing the differential pressure.

[0080] Next, referring to the accompanying drawings, a method for intelligent flushing control of a demister in a thermal power plant, according to an embodiment of this application, is described.

[0081] In this embodiment, the intelligent flushing control method for the demister in a thermal power plant adopts the aforementioned intelligent flushing control system for the demister in a thermal power plant.

[0082] like Figure 6 As shown, the intelligent flushing control method for the demister in this thermal power plant includes the following steps: In step S601, the sensing module is used to collect the operating status parameters of the unit to which the current demister belongs.

[0083] In step S602, the decision and control module receives the unit operating status parameters and inputs them into a preset machine learning prediction model. Based on the unit operating status parameters and historical operating status parameters, differential pressure prediction is performed to obtain the differential pressure prediction result of the current demister within a preset time window.

[0084] In step S603, the flushing strategy optimization unit uses a preset reinforcement learning decision model to generate a target flushing command based on a preset comprehensive optimization objective, provided that the differential pressure prediction result and / or the unit operating status parameters meet the preset flushing triggering conditions.

[0085] In step S604, the execution module performs a zoned flushing operation on the current demister according to the target flushing control command.

[0086] Furthermore, in some embodiments, the unit operating status parameters include the unit load, and the preset flushing trigger condition is: the rate of increase of the unit load is greater than a first preset threshold, and the differential pressure prediction result is less than or equal to a second preset threshold.

[0087] Specifically, the embodiments of this application include a predictive flushing mode and an optimized flushing mode. When the rate of increase of the unit load is greater than a first preset threshold, and the differential pressure prediction result is less than or equal to a second preset threshold, the reinforcement learning decision engine is triggered to execute predictive flushing. If only the differential pressure is predicted to exceed the threshold, but the load does not change drastically, the conventional optimized flushing process is triggered, and water-saving and energy-saving strategies are generated through reinforcement learning.

[0088] Furthermore, in some embodiments, after generating the target flushing command based on a preset reinforcement learning decision model and a preset comprehensive optimization objective, the method further includes: constructing a virtual model corresponding to the current demister using a digital twin module; using the virtual model to simulate and verify the target flushing command; and sending the target flushing control command to the execution module after the verification is passed.

[0089] Furthermore, in some embodiments, after performing a zoned flushing operation on the current demister according to the target flushing control command, the method further includes: collecting system operation data after flushing the current demister; and feeding back the system operation data to the decision and control module to optimize the model parameters of the preset machine learning prediction model and the model parameters of the preset reinforcement learning decision model.

[0090] It should be noted that the foregoing explanation of the embodiment of the intelligent flushing control system for demisters in thermal power plants also applies to the intelligent flushing control method for demisters in thermal power plants in this embodiment, and will not be repeated here.

[0091] The intelligent flushing control method for demisters in thermal power plants proposed in this application utilizes a decision and control module to receive unit operating status parameters and input them into a preset machine learning prediction model for differential pressure prediction. The resulting differential pressure prediction is then used by a flushing strategy optimization unit. When the differential pressure prediction and / or unit operating status parameters meet preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. The execution module then performs zoned flushing operations on the current demister. Thus, through artificial intelligence prediction and optimization, the methods solve the problems of response lag, blind control, and inability to proactively predict clogging trends inherent in traditional demister flushing control methods. This achieves intelligent flushing of the demister, moving from passive handling to proactive prevention, significantly reducing water and electricity consumption while stabilizing the differential pressure.

[0092] Therefore, this invention discloses an intelligent flushing control system and method for demisters in thermal power plants based on artificial intelligence. The system consists of a sensing module, an AI decision-making and control module, and an execution module. The core of the system lies in the AI ​​decision-making and control module, which integrates machine learning prediction models and reinforcement learning optimization models. It can predict the demister pressure differential change trend based on operating data and dynamically generate an optimal flushing strategy with pressure stabilization, water conservation, and energy saving as comprehensive objectives. This method avoids pressure differential runaway under conditions such as sudden load increases through predictive flushing and adapts to complex operating conditions through continuous self-learning. This invention effectively solves the problems of lag and blindness in traditional sequential or differential pressure control methods, realizing intelligent operation and maintenance of demisters from "passive treatment" to "active prevention," significantly improving the safety and economy of desulfurization system operation.

[0093] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0094] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0095] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0096] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0097] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0098] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0099] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0100] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A smart flushing control system for a demister in a thermal power plant, characterized in that, include: The module consists of a perception module, a decision-making and control module, and an execution module. The sensing module is used to collect the operating status parameters of the unit to which the current demister belongs; The decision and control module is communicatively connected to the sensing module. The decision and control module is used to receive the unit operating status parameters and generate target flushing control commands based on the unit operating status parameters. An execution module, which is communicatively connected to the decision and control module, is used to perform zoned flushing operations on the current demister according to the target flushing control command; The decision-making and control module includes: The differential pressure prediction unit is used to predict the differential pressure based on the unit's operating status parameters and historical operating status parameters according to a preset machine learning prediction model, and to obtain the differential pressure prediction result of the current demister within a preset time window. The flushing strategy optimization unit is used to generate a target flushing instruction based on a preset reinforcement learning decision model and a preset comprehensive optimization objective, when the differential pressure prediction result and / or the unit operating status parameters meet the preset flushing triggering conditions. The target flushing instruction includes at least one of flushing start time, flushing area, flushing duration and flushing water pressure.

2. The system according to claim 1, characterized in that, The reward function of the preset reinforcement learning decision model is: RF = -k1*ΔP - k2*W - k3*E; Where RF is the reward function, ΔP is the pressure difference, W is the flushing water consumption, E is the change in induced draft fan energy consumption, and k1, k2, and k3 are weighting coefficients.

3. The system according to claim 1, characterized in that, The sensing module includes: A differential pressure transmitter is used to measure the pressure difference between the inlet and outlet of the current demister; A data interface unit is used to obtain the unit's operating status parameters from the power plant's distributed control system or plant-level monitoring information system.

4. The system according to claim 1, characterized in that, The unit operating status parameters include at least one of the following: unit load, flue gas flow rate, flue gas temperature, fuel sulfur content, and ash content.

5. The system according to claim 1, characterized in that, Also includes: The digital twin module is used to construct a virtual model corresponding to the current demister and to use the virtual model to simulate and verify the target flushing command, so that the target flushing control command can be sent to the execution module after the verification is successful.

6. The system according to claim 1, characterized in that, The preset machine learning prediction model employs a long short-term memory network or a gated recurrent unit.

7. A method for intelligent flushing control of a demister in a thermal power plant, characterized in that, The intelligent flushing control system for demisters in thermal power plants, as described in any one of claims 1-6, comprises the following steps: The sensing module is used to collect the unit operating status parameters to which the current demister belongs; The decision and control module receives the unit operating status parameters and inputs them into a preset machine learning prediction model. Based on the unit operating status parameters and historical operating status parameters, differential pressure prediction is performed to obtain the differential pressure prediction result of the current demister within a preset time window. Using the flushing strategy optimization unit, when the differential pressure prediction result and / or the unit operating status parameters meet the preset flushing trigger conditions, a target flushing command is generated based on a preset reinforcement learning decision model and a preset comprehensive optimization objective. Using the execution module, the current demister is subjected to zoned flushing operation according to the target flushing control command.

8. The method according to claim 7, characterized in that, The unit operating status parameters include the unit load, and the preset flushing trigger condition is: The rate of increase of the unit load is greater than the first preset threshold, and the differential pressure prediction result is less than or equal to the second preset threshold.

9. The method according to claim 7, characterized in that, After generating the target flushing instruction based on the preset reinforcement learning decision model and the preset comprehensive optimization objective, the process also includes: A virtual model corresponding to the current demister is constructed using a digital twin module; The target flushing command is simulated and verified using the virtual model, and after successful verification, the target flushing control command is sent to the execution module.

10. The method according to claim 7, characterized in that, After performing a zoned flushing operation on the current demister according to the target flushing control command, the process further includes: Collect system operation data after flushing the current demister; The system operation data is fed back to the decision and control module to optimize the model parameters of the preset machine learning prediction model and the preset reinforcement learning decision model.

Images