Boiler heating surface soot blowing control method and device, equipment and storage medium

Abstract

The invention belongs to the technical field of intelligent control, and discloses a soot blowing control method and device for a boiler heating surface, equipment and a storage medium, digital twinborn reconstruction of a temperature field, a flow velocity field and a pressure field in the whole process of a boiler is achieved by introducing a deep learning model constrained by a physical equation, and based on the reconstructed temperature field, flow velocity field and pressure field, the soot blowing control method and device for the boiler heating surface are achieved. Executing cleanliness inversion to eliminate working condition interference; and finally, outputting an optimal soot blowing strategy under the guidance of a multi-dimensional reward function by utilizing a reinforcement learning agent, so that clear perception of a soot formation state and a multi-dimensional decision-making target can be realized, a soot blowing control strategy is optimized, and the intelligent degree of soot blowing control is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to a method, device, equipment, and storage medium for controlling soot blowing on boiler heating surfaces. Background Technology

[0002] During the operation of coal-fired power generating units, ash accumulation and coking on boiler heating surfaces, including furnace water-cooled walls, superheaters, reheaters, and economizers, are unavoidable physical phenomena. Ash accumulation increases the thermal resistance of the heating surfaces, leading to increased boiler flue gas temperature and decreased thermal efficiency; severe coking can also cause safety accidents such as coke shedding and flameout, and tube rupture due to overheating. Therefore, the soot blowing system is an important auxiliary device for ensuring the safe and economical operation of the boiler.

[0003] However, traditional methods for detecting ash accumulation are mostly based on cleaning factor calculations, which use the enthalpy increase of the inlet and outlet working fluids to infer the heat absorption. However, this method is a "black box" model and cannot obtain a detailed temperature field distribution inside the furnace. More seriously, the decrease in heat absorption is often not only due to ash accumulation but also to fluctuations in operating conditions, such as reduced unit load or slower flue gas velocity, leading to unclear ash accumulation status and a high risk of false or missed soot blowing. Furthermore, existing soot blowing methods typically use timed programming or simple threshold triggering. This soot blowing control method has a single decision objective and lacks sufficient intelligence.

[0004] It is evident that existing soot blowing control methods suffer from problems such as unclear perception of ash accumulation status and singular decision-making objectives, resulting in a low level of intelligence in soot blowing control. Summary of the Invention

[0005] To address the aforementioned deficiencies, the present invention aims to provide a method, apparatus, equipment, and storage medium for controlling soot blowing on boiler heating surfaces, which can achieve clear perception of ash accumulation status and multi-dimensional decision-making objectives, thereby optimizing soot blowing control strategies and improving the intelligence level of soot blowing control.

[0006] The first aspect of this invention discloses a method for controlling soot blowing on a boiler heating surface, comprising: A physical information neural network model is constructed and trained. The physical information neural network model includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section. The tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, flow velocity field and pressure field. Based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation, the first thermal cleanliness factor of the furnace water-cooled wall is calculated. Based on the physical field state inside the tail flue as simulated by the tail flue convection zone sub-model, the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface are calculated. A cleanliness matrix is ​​constructed based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor. The overall cleanliness matrix, overall flow velocity distribution, overall pressure gradient distribution, current unit load, load change rate, and the previous soot blowing operation record are used as the state space. The state space is input into a trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy. The reinforcement learning decision model is equipped with a multi-dimensional reward function. Verify the first soot blowing control strategy; If the verification passes, the first soot blowing control strategy will be used as the target soot blowing control strategy. The system controls the issuance of soot blowing commands corresponding to the target soot blowing control strategy, acquires the response data of the boiler's real physical field after executing the target soot blowing control strategy, and performs online correction of the physical information neural network model based on the response data.

[0007] In some embodiments, a physical information neural network model is constructed and trained, including: Using spatiotemporal coordinates and operating condition variables as inputs, the physical field states inside the furnace and at the outlet section as outputs, and the energy equations describing heat convection and heat conduction, the Navier-Stokes equations describing flue gas flow, and the boundary conditions of the velocity field as constraints, a sub-model of the furnace radiation zone is constructed based on a neural network. Using spatial coordinates as input, the physical field state inside the tail flue as output, and the momentum equation as constraint, a sub-model of the convection zone of the tail flue is constructed based on a neural network. The physical field states of the furnace interior and outlet section output by the furnace radiation zone sub-model are directly used as the inlet boundary conditions of the tail flue convection zone sub-model to obtain a cascaded coupled physical information neural network model, and the physical information neural network model is trained.

[0008] In some embodiments, the first thermal cleanliness factor of the furnace water-cooled wall is calculated based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation, including: The first normal temperature gradient is calculated based on the temperature of the furnace water-cooled wall surface output by the furnace radiation zone sub-model simulation. The first actual heat flux density is calculated based on the first normal temperature gradient and thermal conductivity. Based on the local flow velocity and flue gas temperature outside the boundary layer of the furnace water-cooled wall surface, which are output from the furnace radiation zone sub-model simulation, the first theoretical clean heat flux density is calculated by substituting them into the standard heat transfer formula, which includes radiation and convection terms. The ratio of the first actual heat flux density to the first theoretical clean heat flux density is calculated and used as the first thermal cleanliness factor of the furnace water-cooled wall surface after eliminating operating condition interference.

[0009] In some embodiments, based on the physical field state inside the tail flue as simulated by the tail flue convection zone sub-model, the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface are calculated, including: The second normal temperature gradient is calculated based on the temperature of the heated surface of the tail flue, which is output by the simulation of the tail flue convection zone sub-model. The second actual heat flux density is calculated based on the second normal temperature gradient and thermal conductivity. Based on the local flow velocity and flue gas temperature outside the boundary layer of the tail flue heat-receiving surface, which are output by the simulation of the tail flue convection zone sub-model, the second theoretical clean heat flux density is calculated by substituting it into the standard heat transfer formula, which includes radiation and convection terms. The ratio of the second actual heat flux density to the second theoretical clean heat flux density is calculated and used as the second thermal cleanliness factor of the tail flue heating surface after eliminating operating condition interference. The simulated pressure gradient inside the tail flue, output by the simulation of the convection zone sub-model, is obtained. The ratio between the simulated pressure gradient and the baseline pressure gradient under theoretical clean conditions is calculated and used as the hydraulic cleanliness factor of the tail flue.

[0010] In some embodiments, validating the first soot blowing control strategy includes: The first soot blowing control strategy is verified using one or more of the following strategies: wear-blocking verification, thermal shock protection, and silent duration; wherein... The wear prevention verification includes: reconstructing the local flow velocity of the proposed operation area in the first soot blowing control strategy through the physical information neural network model; if the local flow velocity of the proposed operation area does not exceed the preset wear critical flow velocity threshold, the verification passes; if the local flow velocity of the proposed operation area exceeds the preset wear critical flow velocity threshold, it is determined that there is a wear risk, and the verification fails. The thermal shock protection includes: calculating the third thermal cleanliness factor of the proposed operating area in the first soot blowing control strategy through the physical information neural network model; if the third thermal cleanliness factor of the proposed operating area is not higher than the preset thermal shock threshold, the verification passes; if the third thermal cleanliness factor of the proposed operating area is higher than the preset thermal shock threshold, it is determined that there is a risk of thermal shock, and the verification fails. The silent duration strategy includes: recording the time since the last soot blowing when the soot blowing gun selected in the first soot blowing control strategy has not been blown; if the time since the last soot blowing is not less than the preset soot blowing silent duration threshold, the verification passes and soot blowing must be carried out; if the time since the last soot blowing is less than the preset soot blowing silent duration threshold, the verification fails.

[0011] In some embodiments, if the verification passes, before using the first soot blowing control strategy as the target soot blowing control strategy, the method further includes: If the verification passes, determine whether there is a serious blockage. If no serious blockage is found, execute the operation of using the first soot blowing control strategy as the target soot blowing control strategy; If severe blockage exists, the first soot blowing control strategy is modified to obtain a second soot blowing control strategy based on the principle of prioritizing unblocking, and the second soot blowing control strategy is used as the target soot blowing control strategy.

[0012] A second aspect of the present invention discloses a boiler heating surface soot blowing control device, comprising: The reconstruction unit is used to construct and train a physical information neural network model, which includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section, and the tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, flow velocity field and pressure field. The first calculation unit is used to calculate the first thermal cleanliness factor of the furnace water-cooled wall based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation. The second calculation unit is used to calculate the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface based on the physical field state inside the tail flue output by the simulation output of the tail flue convection zone sub-model. The matrix construction unit is used to construct a cleanliness matrix of the entire field based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor. The decision unit is used to take the overall cleanliness matrix, overall flow velocity distribution, overall pressure gradient distribution, current unit load, load change rate, and the soot blowing operation record of the previous moment as the state space, and input the state space into the trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy; the reinforcement learning decision model is set with a multi-dimensional reward function. A verification unit is used to verify the first soot blowing control strategy; The determining unit is used to take the first soot blowing control strategy as the target soot blowing control strategy when the verification passes. The control unit is used to control the issuance of soot blowing commands corresponding to the target soot blowing control strategy; The correction unit is used to acquire the response data of the boiler's real physical field after the target soot blowing control strategy is executed, and to perform online correction of the physical information neural network model based on the response data.

[0013] In some embodiments, the apparatus further includes: The judgment unit is used to determine whether there is a serious blockage when the verification is passed and before the determination unit uses the first soot blowing control strategy as the target soot blowing control strategy; if there is no serious blockage, the determination unit is triggered to perform the operation of using the first soot blowing control strategy as the target soot blowing control strategy. The correction unit is used to correct the first soot blowing control strategy to obtain a second soot blowing control strategy when the verification passes and the judgment unit determines that there is a serious blockage, based on the principle of prioritizing unblocking, and to use the second soot blowing control strategy as the target soot blowing control strategy.

[0014] A third aspect of the present invention discloses an electronic device, including a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the boiler heating surface soot blowing control method disclosed in the first aspect.

[0015] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the boiler heating surface soot blowing control method disclosed in the first aspect.

[0016] Compared with existing technologies, the beneficial effects of this invention are that by introducing a deep learning model constrained by physical equations, a "digital twin" reconstruction of the temperature field, flow velocity field, and pressure field of the entire boiler process is achieved. Based on the reconstructed temperature field, flow velocity field, and pressure field, cleanliness inversion is performed to eliminate operating condition interference. Finally, by using a reinforcement learning agent, the optimal soot blowing strategy is output under the guidance of a multi-dimensional reward function. This can achieve clear perception of ash accumulation status and multi-dimensional decision-making objectives, thereby optimizing the soot blowing control strategy and improving the intelligence level of soot blowing control. Attached Figure Description

[0017] Figure 1 This is a flowchart of a boiler heating surface soot blowing control method disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a boiler heating surface soot blowing control device disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device disclosed in an embodiment of the present invention.

[0018] Explanation of reference numerals in the attached figures: 201. Reconstruction unit; 202. First calculation unit; 203. Second calculation unit; 204. Matrix construction unit; 205. Decision unit; 206. Verification unit; 207. Determination unit; 208. Control unit; 209. Correction unit; 301. Memory; 302. Processor. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0020] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 110, 120, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0021] It will be understood by those skilled in the art that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0022] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Throughout the description, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 This invention discloses a method for controlling soot blowing on boiler heating surfaces. This method is applicable to a 660MW supercritical tangential combustion boiler. The executing entity of this method can be an electronic device such as an industrial controller, computer, laptop, or tablet, or a boiler heating surface soot blowing control device embedded in an electronic device; this invention does not limit this to any particular type.

[0025] like Figure 1 As shown, the method includes the following steps 110-180: 110. Construct and train a physical information neural network model, which includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section, and the tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, velocity field and pressure field.

[0026] In this embodiment of the invention, the Physics-Informed Neural Networks (PINN) model includes a cascaded furnace radiation zone sub-model (Model-Furnace) and a tail flue convection zone sub-model (Model-Tail). This enables the reconstruction of the entire physical field during the actual operation of the boiler, including the three-dimensional temperature, pressure, and velocity fields, under meshless conditions. The physical field states inside the furnace and at the outlet section calculated by the furnace radiation zone sub-model are directly used as the inlet boundary conditions of the tail flue convection zone sub-model without intermediate compression, achieving a cascaded reconstruction of the entire physical field. This ensures that when the flame is deflected inside the furnace, the velocity field at the flue inlet is also deflected.

[0027] As an optional implementation, step 110 includes the following steps 1101-1103 (not shown): 1101. Using spatiotemporal coordinates and operating condition variables as inputs, the physical field states inside the furnace and at the outlet section as outputs, and the energy equations describing heat convection and heat conduction, the Navier-Stokes equations describing flue gas flow, and the boundary conditions of the velocity field as constraints, a sub-model of the furnace radiation zone is constructed based on a neural network.

[0028] The input to the furnace radiation zone sub-model is spatiotemporal coordinates. The output includes operating condition variables, representing the physical field states inside the furnace and at the outlet section, including temperature. ,pressure Velocity vector The operating condition variables and the constraint variables for the velocity field boundary conditions are all real-time data that can be collected from the power plant's distributed control system (DCS) or supervisory information system (SIS). After collecting the real-time data, further data preprocessing can be performed, including removing bad pixels, time alignment, and normalization, thereby accelerating the convergence of the neural network.

[0029] Operating variables include unit load and physical properties of the burner region, such as total coal quantity, total air volume, and burner tilt angle. Constraint variables for the velocity field boundary conditions include economizer inlet feedwater temperature, water-cooled wall outlet steam temperature, temperature sensors of each stage of the heating surface, flue gas inlet and outlet temperatures, and static pressure. Based on these constraints, the velocity field boundary conditions can be set to force the air volume at the burner inlet to equal the total air volume collected from the DCS system, force the velocity at the furnace water-cooled wall surface to be zero (i.e., no slip condition), and so on.

[0030] The loss function of the furnace radiation zone sub-model incorporates the Navier-Stokes equations for describing flue gas flow, the energy equations for describing thermal convection and conduction, and the velocity field boundary conditions. Specifically, the energy equations include a radiation source term, calculated using the P-1 radiation approximation model.

[0031] 1102. Using spatial coordinates as input, the physical field state inside the tail flue as output, and momentum equation as constraint, a sub-model of the convection zone of the tail flue is constructed based on a neural network.

[0032] The input to the tail flue convection region sub-model is spatial coordinates. Unlike the spatiotemporal coordinate system, the origin of this spatial coordinate system is shifted to the flue inlet. The output of the tail flue convection region sub-model is the physical field distribution at various locations within the tail flue, including temperature. ,pressure Velocity vector .

[0033] Preferably, the furnace radiation zone sub-model and the tail flue convection zone sub-model are constructed and trained using a fully connected neural network (FCNN) with the same network structure. The fully connected neural network includes 8 hidden layers, each with 100 neurons, and the activation function is the Tanh function.

[0034] In a preferred embodiment of the invention, the construction of the convection zone sub-model in the tail flue incorporates porous media theory. For densely packed tube bundle regions with small diameters and numerous parallel tube bundles within the tail flue, such as superheaters, reheaters, and economizers, instead of constructing a geometric mesh for a single tube bank, it is equivalent to a porous medium with permeability and inertial drag coefficients. Permeability characterizes the porosity caused by ash accumulation, and the inertial drag coefficient characterizes the nonlinear pressure drop loss at high flow velocities. Therefore, a porous media resistance model is constructed by adding a fluid resistance source term to the momentum equation, using the change in pressure gradient to inversely determine the ash accumulation and blockage state in the densely packed tube bundle region. Specifically, the fluid resistance source term is constructed using the Darcy-Forchheimer extended formula, thereby establishing a quantitative mapping relationship between the pressure gradient and the degree of ash accumulation.

[0035] Specifically, fluid resistance source terms As shown in equation (1): (1) in, Permeability is the rate of dust accumulation; the smaller the value, the heavier the dust accumulation and the greater the pressure drop. ρ is the inertial drag coefficient; ρ is the fluid density. For fluid dynamic viscosity, This refers to Darcy's speed.

[0036] When the pressure difference between the flue inlet and outlet increases during actual operation, PINN will automatically reduce the permeability of the tube bundle region or increase the pressure gradient to minimize data loss. This abnormal change in the pressure field is a direct signal of ash accumulation and blockage.

[0037] 1103. The physical field states of the furnace interior and outlet section output by the furnace radiation zone sub-model are directly used as the inlet boundary conditions of the tail flue convection zone sub-model to obtain a cascaded coupled physical information neural network model, and the physical information neural network model is trained.

[0038] In practical applications, the trained PINN model is used to simulate the real physical field during the actual operation of the boiler, and can be further used to optimize soot blowing control.

[0039] The loss function during PINN model training includes at least three parts: physical residual loss, data fitting loss, and boundary condition loss. The physical residual loss includes at least the residuals of the mass continuity equation, the momentum conservation equation, and the energy conservation equation. The data fitting loss includes at least the measured data used to anchor the sparse sensors. The boundary condition loss includes at least the burner inlet airflow constraint loss and the furnace water-cooled wall surface no-slip condition loss.

[0040] During the training of the PINN model, a "offline pre-training + online fine-tuning" strategy was adopted. In the offline pre-training phase, one year's worth of historical data was used, and the Adam optimizer was employed for 10,000 epochs to learn the fundamental physical characteristics of the furnace. In the online fine-tuning phase, every 15 minutes, the latest measured data collected by sparse sensors was used, and the L-BFGS second-order optimizer was employed for 50 rapid iterations to make the PINN model state approximate the actual physical field at the current moment.

[0041] 120. Based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation, the first thermal cleanliness factor of the furnace water-cooled wall is calculated.

[0042] Step 120 may include the following steps 1201-1203 (not shown): 1201. Based on the temperature of the furnace water-cooled wall surface output by the furnace radiation zone sub-model simulation, the first normal temperature gradient is calculated; based on the first normal temperature gradient and thermal conductivity, the first actual heat flux density is calculated.

[0043] The water-cooled wall surface of the furnace is the heating surface. The normal temperature gradient of the heating surface can be calculated by automatic differentiation, and the actual heat flux density can be calculated by combining the thermal conductivity. The calculation formula is shown in equation (2) below: (2) in, The vector represents the normal temperature gradient, where T represents temperature, and n represents the unit normal vector of the heated surface, with the direction of n perpendicular to the surface of the heated surface. This represents the thermal conductivity.

[0044] 1202. Based on the local flow velocity and flue gas temperature outside the boundary layer of the furnace water-cooled wall surface output by the furnace radiation zone sub-model simulation, the first theoretical clean heat flux density is calculated by substituting it into the standard heat transfer formula including radiation and convection terms.

[0045] The calculation process utilizes local velocity data to correct the theoretical heat transfer coefficient, eliminating interference from unit load and airflow fluctuations on ash accumulation assessment. Specifically, a grid point 10cm from the heated surface is taken as a local location outside the boundary layer, and the flue gas temperature at that local location is obtained. and the temperature of the heated surface The theoretical clean heat flux density, including radiation and convection terms, is calculated using the following formula (3): (3) in, For the theoretical clean heat flux density, To consider the overall heat transfer coefficient, The temperature of the heated surface. For flue gas temperature, The Stefan-Boltzmann constant is... The effective emissivity of the wall surface. This refers to the local flue gas velocity.

[0046] 1203. Calculate the ratio of the first actual heat flux density to the first theoretical clean heat flux density, and use it as the first thermal cleanliness factor of the furnace water-cooled wall surface after eliminating operating condition interference.

[0047] The ratio of the two is used as the thermal cleanliness factor. This index eliminates the influence of load and flow field fluctuations and can more accurately reflect the degree of ash accumulation.

[0048] 130. Based on the physical field state inside the tail flue as simulated by the sub-model of the convection zone of the tail flue, the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface are calculated.

[0049] The calculation method for the second thermal cleanliness factor of the tail flue heating surface is the same as that for the first thermal cleanliness factor of the furnace water-cooled wall surface. The detailed formula will not be elaborated here, only the steps will be described.

[0050] That is, the calculation of the second thermal cleanliness factor may include the following steps 1301-1303 (not shown): 1301. Based on the temperature of the heated surface of the tail flue from the simulation output of the tail flue convection zone sub-model, the second normal temperature gradient is calculated; based on the second normal temperature gradient and the thermal conductivity, the second actual heat flux density is calculated.

[0051] 1302. Based on the local flow velocity and flue gas temperature outside the boundary layer of the tail flue heat-receiving surface, which are output by the simulation of the tail flue convection region submodel, the second theoretical clean heat flux density is calculated by substituting it into the standard heat transfer formula, which includes radiation and convection terms.

[0052] 1303. Calculate the ratio of the second actual heat flux density to the second theoretical clean heat flux density, and use it as the second thermal cleanliness factor of the tail flue heating surface after eliminating operating condition interference.

[0053] The calculation of the hydraulic cleanliness factor of the tail flue heating surface can specifically include: obtaining the simulated pressure gradient inside the tail flue from the simulation output of the tail flue convection zone sub-model, and calculating the ratio between the simulated pressure gradient and the reference pressure gradient under theoretical clean conditions, which is used as the hydraulic cleanliness factor of the tail flue. Its calculation formula is shown in equation (4) below: (4) in, To simulate pressure gradient, As the reference pressure gradient, The hydraulic cleanliness factor is used to reflect whether bridging or blockage has occurred in the tube bundles within the tail flue.

[0054] 140. Construct a cleanliness matrix for the entire field based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor.

[0055] 150. The cleanliness matrix of the entire field, the velocity distribution of the entire field, the pressure gradient distribution of the entire field, the current unit load, the load change rate, and the soot blowing operation record of the previous moment are used as the state space. The state space is input into the trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy. The reinforcement learning decision model is set with a multi-dimensional reward function.

[0056] The overall velocity distribution includes the furnace velocity output from the furnace radiation zone sub-model simulation and the tail flue velocity output from the tail flue convection zone sub-model simulation; the overall pressure gradient distribution includes the furnace heating surface pressure gradient output from the furnace radiation zone sub-model simulation and the tail flue heating surface pressure gradient output from the tail flue convection zone sub-model simulation.

[0057] The process of constructing a reinforcement learning decision-making model includes: The state space is defined by including the overall cleanliness matrix, overall velocity distribution, and overall pressure gradient distribution calculated using PINN, as well as the current unit load, load change rate, and the soot blowing operation record from the previous moment. The action space is defined as discrete actions, including the soot blowing gun sequence and continuous soot blowing parameters. The soot blowing gun sequence is used to characterize the selected soot blowing gun; the continuous soot blowing parameters are used to set the soot blowing steam pressure, soot blowing step speed, etc.

[0058] The reward function is defined as a four-dimensional comprehensive reward function, including performance gains, cost penalties, safety penalties, and control stability penalties. Performance gains include improved boiler thermal efficiency and reduced desuperheating water flow; cost penalties include increased media consumption and reduced equipment lifespan; safety penalties include penalties based on local flow velocity calculations for pipe wall wear rate and thermal shock; and control stability penalties are used to suppress frequent operations during unit load changes.

[0059] Based on this, the reward value is calculated as shown in equation (5): (5) in, This indicates performance gains, specifically the reduction in flue gas temperature and the reduction in desuperheating water flow rate. This indicates cost penalties, including the consumption of steam and the shortening of motor life; This represents a safety penalty; for example, if PINN predicts that the current flue gas velocity in a certain soot blowing area exceeds 15 m / s, this penalty will become extremely large. This forces the agent to learn to avoid blowing soot through high-speed flue gas corridors, effectively preventing pipe damage and rupture. This indicates a control stability penalty. When the unit is in a rapid load change phase, such as during Automatic Gain Control (AGC) peak shaving, it increases the negative reward to suppress unnecessary soot blowing and avoid interfering with combustion control.

[0060] The reward function is designed as a four-dimensional hierarchical dynamic reward system, which deeply integrates economic efficiency, thermodynamic characteristics, equipment health management, grid demand response, and control stability.

[0061] The training process of a reinforcement learning decision model includes: First, the agent performs actions in the environment, observes the environmental state and the rewards obtained, forming experience data, including state, action, reward, and next state. This experience is stored in an experience replay buffer for subsequent learning.

[0062] Secondly, the value function is updated using the Temporal-Difference Learning (TD) method. Specifically, the Q-function can be used as the value function, and the Bellman equation can be used for iterative updates.

[0063] Then, based on the updated value function, the strategy is adjusted to select a better action. A greedy strategy can be used to balance trying new actions with selecting the current best action. In deep reinforcement learning, a target network is used to stabilize the training process. The parameters of the target network are periodically copied from the main network to calculate the target Q-value, reducing oscillations during training.

[0064] Next, small batches of experience are randomly sampled from the experience replay buffer for training. Experience replay can break the correlation between data and improve sample utilization efficiency.

[0065] Finally, the above process is repeated iteratively. Through continuous interaction and learning with the environment, the agent gradually optimizes its strategy and eventually learns to make optimal decisions in complex environments. The entire training process is a cycle of trial and error, evaluation and improvement, with the agent starting from random exploration and gradually converging to the optimal strategy.

[0066] 160. Verify the first soot blowing control strategy.

[0067] The first soot blowing control strategy is verified based on the full-process physical field data reconstructed by the physical information neural network model. The full-process physical field data includes the physical field state of the furnace interior and outlet section output by the furnace radiation zone sub-model simulation, and the physical field state of the tail flue interior output by the tail flue convection zone sub-model simulation.

[0068] Specifically, this may include: verifying the first soot blowing control strategy using one or more of the following strategies: wear-blocking verification, thermal shock protection, and silent duration, based on the full-process physical field data reconstructed from the physical information neural network model; wherein, Wear prevention verification includes: reconstructing the local flow velocity of the proposed operating area in the first soot blowing control strategy through PINN; if the local flow velocity of the proposed operating area does not exceed the preset wear critical flow velocity threshold, the verification passes; if the local flow velocity of the proposed operating area exceeds the preset wear critical flow velocity threshold, it is determined that there is a wear risk, and the verification fails.

[0069] Thermal shock protection includes: calculating the third thermal cleanliness factor of the proposed operating area in the first soot blowing control strategy using PINN; if the third thermal cleanliness factor of the proposed operating area is not higher than the preset thermal shock threshold, the verification passes; if the third thermal cleanliness factor of the proposed operating area is higher than the preset thermal shock threshold, it is determined that there is a risk of thermal shock, and the verification fails.

[0070] The silent duration strategy includes: recording the time since the last soot blowing when the soot blowing gun selected in the first soot blowing control strategy has not been blown. If the time since the last soot blowing is not less than the preset soot blowing silent duration threshold, the verification passes and soot blowing must be carried out; if the time since the last soot blowing is less than the preset soot blowing silent duration threshold, the verification fails.

[0071] Furthermore, if the verification fails, the blowing command corresponding to the first blowing control strategy can be forcibly intercepted, and a negative reward can be fed back to the reinforcement learning decision model.

[0072] This invention takes into account the risk of wear on the tube walls caused by soot blowing. For example, high-pressure purging in areas with excessively high local flue gas velocities can exponentially accelerate tube wall wear, which is one of the main causes of boiler tube rupture. Therefore, wear-blocking verification can prevent tube rupture caused by tube wall wear, achieving intelligent decision-making with controllable wear risk. Furthermore, thermal shock protection can prevent steam from directly impacting the exposed metal tube walls and causing thermal stress fatigue, thereby ensuring optimized safety.

[0073] 170. If the verification passes, the first soot blowing control strategy shall be adopted as the target soot blowing control strategy.

[0074] If the verification passes, before using the first soot blowing control strategy as the target soot blowing control strategy, it can be determined whether there is a serious blockage. If there is no serious blockage, the operation of using the first soot blowing control strategy as the target soot blowing control strategy is executed. If there is a serious blockage, the first soot blowing control strategy is modified according to the principle of prioritizing unblocking to obtain the second soot blowing control strategy, and the second soot blowing control strategy is used as the target soot blowing control strategy.

[0075] Specifically, determining whether there is a serious blockage includes: Monitor whether the hydraulic cleanliness factor of the tail flue is greater than the preset critical threshold for blockage. If the hydraulic cleanliness factor of the tail flue is greater than the preset critical threshold for blockage, it is determined that there is a serious blockage. If the hydraulic cleanliness factor of the tail flue is not greater than the preset critical threshold for blockage, it is determined that there is no serious blockage.

[0076] A second soot blowing control strategy is obtained by modifying the first soot blowing control strategy, including: The second soot blowing control strategy is obtained by replacing the selected soot blowing gun in the first soot blowing control strategy with an acoustic soot blower or a specific high-power unblocking soot blowing gun, and by adjusting the continuous soot blowing parameters in the first soot blowing control strategy accordingly.

[0077] 180. The soot blowing command corresponding to the target soot blowing control strategy is issued, the response data of the boiler's real physical field after the target soot blowing control strategy is executed is obtained, and the physical information neural network model is corrected online based on the response data.

[0078] The soot blowing control strategy corresponding to the control target is issued to the soot blowing control system for execution. Based on the feedback from the actual physical field response of the boiler after execution, the parameters of the PINN model and the reinforcement learning decision model are updated online. This achieves closed-loop control and online updates.

[0079] In summary, by implementing the embodiments of this invention and introducing a deep learning model constrained by physical equations, a "digital twin" reconstruction of the temperature, velocity, and pressure fields of the entire boiler process is achieved. This upgrades the perception dimension from simple temperature perception to multi-physics field perception encompassing "temperature + velocity + pressure," filling a gap in flow field monitoring. Physical equation constraints effectively filter sensor noise and enable high-precision virtual measurements in areas without measurement points, thereby improving monitoring accuracy. Furthermore, a reinforcement learning decision-making model considering efficiency, cost, wear, and safety is constructed. By incorporating flow field wear risk into the decision-making closed loop, the potential for blow-through tube rupture is eliminated at the algorithmic level, enhancing decision-making safety. The physical information-based model exhibits good generalization ability, adapting to changes in coal type and significant load fluctuations, thus enhancing its adaptability.

[0080] like Figure 2 As shown in the figure, this embodiment of the invention also discloses a boiler heating surface soot blowing control device, including a reconstruction unit 201, a first calculation unit 202, a second calculation unit 203, a matrix construction unit 204, a decision unit 205, a verification unit 206, a determination unit 207, a control unit 208, and a correction unit 209, wherein, Reconstruction unit 201 is used to construct and train a physical information neural network model. The physical information neural network model includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section, and the tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, velocity field and pressure field. The first calculation unit 202 is used to calculate the first thermal cleanliness factor of the furnace water-cooled wall based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation. The second calculation unit 203 is used to calculate the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface based on the physical field state inside the tail flue output by the simulation output of the tail flue convection zone sub-model. The matrix construction unit 204 is used to construct the overall cleanliness matrix based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor. Decision unit 205 is used to take the cleanliness matrix of the whole field, the flow velocity distribution of the whole field, the pressure gradient distribution of the whole field, the current unit load, the load change rate, and the soot blowing operation record of the previous moment as the state space, and input the state space into the trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy; the reinforcement learning decision model is set with a multi-dimensional reward function. Verification unit 206 is used to verify the first soot blowing control strategy; The determining unit 207 is used to take the first soot blowing control strategy as the target soot blowing control strategy when the verification is passed. Control unit 208 is used to control the issuance of soot blowing commands corresponding to the target soot blowing control strategy; The correction unit 209 is used to acquire the response data of the boiler's real physical field after the target soot blowing control strategy is executed, and to perform online correction of the physical information neural network model based on the response data.

[0081] Optionally, the boiler heating surface soot blowing control device may also include the following units not shown: The judgment unit is used to determine whether there is a serious blockage when the verification is passed and before the determination unit 207 sets the first soot blowing control strategy as the target soot blowing control strategy; if there is no serious blockage, the determination unit 207 is triggered to perform the operation of setting the first soot blowing control strategy as the target soot blowing control strategy. The correction unit is used to correct the first soot blowing control strategy to obtain a second soot blowing control strategy when the verification passes and the judgment unit determines that there is a serious blockage, based on the principle of prioritizing unblocking, and to use the second soot blowing control strategy as the target soot blowing control strategy.

[0082] Furthermore, the specific method by which the judgment unit determines whether a serious blockage exists is as follows: The judgment unit is used to monitor whether the hydraulic cleanliness factor of the tail flue is greater than the preset critical threshold for blockage. If it is greater than the preset critical threshold for blockage, it is determined that there is a serious blockage; if it is not greater than the preset critical threshold for blockage, it is determined that there is no serious blockage.

[0083] Optionally, the reconfigurable unit 201 includes the following sub-units (not shown): The first construction sub-unit is used to take spatiotemporal coordinates and operating condition variables as inputs, the physical field state inside the furnace and the outlet section as outputs, and the energy equation describing heat convection and heat conduction, the Navier-Stokes equation describing flue gas flow, and the flow velocity field boundary conditions as constraints to construct a furnace radiation zone sub-model based on a neural network. The second construction sub-unit is used to construct a convection zone sub-model of the tail flue based on a neural network, with spatial coordinates as input, the physical field state inside the tail flue as output, and momentum equation as constraint. The coupling sub-unit is used to directly use the physical field state of the furnace interior and outlet section output by the furnace radiation zone sub-model as the inlet boundary condition of the tail flue convection zone sub-model, so as to obtain a physical information neural network model of the cascaded coupling architecture. The training subunit is used to train the physical information neural network model.

[0084] Optionally, the first computational unit includes the following sub-units (not shown): The first simulation calculation subunit is used to calculate the first normal temperature gradient based on the temperature of the furnace water-cooled wall surface output by the furnace radiation zone sub-model simulation; and to calculate the first actual heat flux density based on the first normal temperature gradient and the thermal conductivity. The first theoretical calculation subunit is used to calculate the first theoretical clean heat flux density by substituting the local flow velocity and flue gas temperature outside the boundary layer of the furnace water-cooled wall surface, which are output by the furnace radiation zone sub-model simulation, into the standard heat transfer formula including radiation and convection terms. The first cleanliness inversion subunit is used to calculate the ratio of the first actual heat flux density to the first theoretical clean heat flux density, which serves as the first thermal cleanliness factor of the furnace water-cooled wall surface after eliminating operating condition interference.

[0085] Optionally, the second computational unit includes the following sub-units (not shown): The second simulation calculation subunit is used to calculate the second normal temperature gradient based on the temperature of the heated surface of the tail flue from the simulation output of the tail flue convection zone sub-model; and to calculate the second actual heat flux density based on the second normal temperature gradient and the thermal conductivity. The second theoretical calculation subunit calculates the second theoretical clean heat flux density by substituting the local flow velocity and flue gas temperature outside the boundary layer of the tail flue heat-receiving surface, which are output by the simulation of the tail flue convection zone sub-model, into the standard heat transfer formula including radiation and convection terms. The second cleanliness inversion subunit is used to calculate the ratio of the second actual heat flux density to the second theoretical clean heat flux density, which serves as the second thermal cleanliness factor of the tail flue heating surface after eliminating operating condition interference. The hydraulic cleanliness calculation subunit is used to obtain the simulated pressure gradient inside the tail flue from the simulation output of the tail flue convection zone sub-model, and calculate the ratio between the simulated pressure gradient and the reference pressure gradient under theoretical clean conditions, which is used as the hydraulic cleanliness factor of the tail flue.

[0086] Optionally, the verification unit 206 is specifically used to verify the first soot blowing control strategy using any one or more of the following strategies: wear-blocking verification, thermal shock protection, and silent duration; wherein, Wear prevention verification includes: reconstructing the local flow velocity of the proposed operation area in the first soot blowing control strategy through a physical information neural network model; if the local flow velocity of the proposed operation area does not exceed the preset wear critical flow velocity threshold, the verification passes; if the local flow velocity of the proposed operation area exceeds the preset wear critical flow velocity threshold, it is determined that there is a wear risk, and the verification fails. Thermal shock protection includes: calculating the third thermal cleanliness factor of the proposed operating area in the first soot blowing control strategy through a physical information neural network model; if the third thermal cleanliness factor of the proposed operating area is not higher than the preset thermal shock threshold, the verification passes; if the third thermal cleanliness factor of the proposed operating area is higher than the preset thermal shock threshold, it is determined that there is a risk of thermal shock, and the verification fails. The silent duration strategy includes: recording the time elapsed since the last soot blowing when the soot blowing gun selected in the first soot blowing control strategy was not blown. If the time elapsed since the last soot blowing is not less than the preset soot blowing silent duration threshold, the verification passes and soot blowing must be carried out; if the time elapsed since the last soot blowing is less than the preset soot blowing silent duration threshold, the verification fails.

[0087] Optionally, the boiler heating surface soot blowing control device also includes an interception unit (not shown), which is used to forcibly intercept the soot blowing command corresponding to the first soot blowing control strategy when the first soot blowing control strategy fails to pass the verification, and to feed back negative rewards to the reinforcement learning decision model.

[0088] like Figure 3 As shown, this embodiment of the invention also discloses an electronic device, including a memory 301 storing executable program code and a processor 302 coupled to the memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the boiler heating surface soot blowing control method described in the above embodiments.

[0089] like Figure 4 As shown in the illustration, this invention also discloses a computer device. This computer device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store relevant data for the boiler heating surface soot blowing control method. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the boiler heating surface soot blowing control method described in the above embodiments.

[0090] This invention also discloses a computer-readable storage medium storing a computer program that causes a computer to execute the boiler heating surface soot blowing control method described in the above embodiments. The storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0091] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for controlling soot blowing on boiler heating surfaces, characterized in that, include: A physical information neural network model is constructed and trained. The physical information neural network model includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section. The tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, flow velocity field and pressure field. Based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation, the first thermal cleanliness factor of the furnace water-cooled wall is calculated. Based on the physical field state inside the tail flue as simulated by the tail flue convection zone sub-model, the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface are calculated. A cleanliness matrix is ​​constructed based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor. The overall cleanliness matrix, overall flow velocity distribution, overall pressure gradient distribution, current unit load, load change rate, and the previous soot blowing operation record are used as the state space. The state space is input into the trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy. The reinforcement learning decision model is configured with a multi-dimensional reward function; Verify the first soot blowing control strategy; If the verification passes, the first soot blowing control strategy will be used as the target soot blowing control strategy. The system controls the issuance of soot blowing commands corresponding to the target soot blowing control strategy, acquires the response data of the boiler's real physical field after executing the target soot blowing control strategy, and performs online correction of the physical information neural network model based on the response data.

2. The boiler heating surface soot blowing control method according to claim 1, characterized in that, The physical information neural network model is constructed and trained, including: Using spatiotemporal coordinates and operating condition variables as inputs, the physical field states inside the furnace and at the outlet section as outputs, and the energy equations describing heat convection and heat conduction, the Navier-Stokes equations describing flue gas flow, and the boundary conditions of the velocity field as constraints, a sub-model of the furnace radiation zone is constructed based on a neural network. Using spatial coordinates as input, the physical field state inside the tail flue as output, and the momentum equation as constraint, a sub-model of the convection zone of the tail flue is constructed based on a neural network. The physical field states of the furnace interior and outlet section output by the furnace radiation zone sub-model are directly used as the inlet boundary conditions of the tail flue convection zone sub-model to obtain a cascaded coupled physical information neural network model, and the physical information neural network model is trained.

3. The boiler heating surface soot blowing control method according to claim 1, characterized in that, Based on the physical field state inside the furnace output from the furnace radiation zone sub-model simulation, the first thermal cleanliness factor of the furnace water-cooled wall is calculated, including: The first normal temperature gradient is calculated based on the temperature of the furnace water-cooled wall surface output by the furnace radiation zone sub-model simulation. The first actual heat flux density is calculated based on the first normal temperature gradient and thermal conductivity. Based on the local flow velocity and flue gas temperature outside the boundary layer of the furnace water-cooled wall surface, which are output from the furnace radiation zone sub-model simulation, the first theoretical clean heat flux density is calculated by substituting them into the standard heat transfer formula, which includes radiation and convection terms. The ratio of the first actual heat flux density to the first theoretical clean heat flux density is calculated and used as the first thermal cleanliness factor of the furnace water-cooled wall surface after eliminating operating condition interference.

4. The boiler heating surface soot blowing control method according to claim 1, characterized in that, Based on the physical field state inside the tail flue as simulated from the tail flue convection zone sub-model, the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface are calculated, including: The second normal temperature gradient is calculated based on the temperature of the heated surface of the tail flue, which is output by the simulation of the tail flue convection zone sub-model. The second actual heat flux density is calculated based on the second normal temperature gradient and thermal conductivity. Based on the local flow velocity and flue gas temperature outside the boundary layer of the tail flue heat-receiving surface, which are output by the simulation of the tail flue convection zone sub-model, the second theoretical clean heat flux density is calculated by substituting it into the standard heat transfer formula, which includes radiation and convection terms. The ratio of the second actual heat flux density to the second theoretical clean heat flux density is calculated and used as the second thermal cleanliness factor of the tail flue heating surface after eliminating operating condition interference. The simulated pressure gradient inside the tail flue, output by the simulation of the convection zone sub-model, is obtained. The ratio between the simulated pressure gradient and the baseline pressure gradient under theoretical clean conditions is calculated and used as the hydraulic cleanliness factor of the tail flue.

5. The boiler heating surface soot blowing control method according to claim 1, characterized in that, Verification of the first soot blowing control strategy includes: The first soot blowing control strategy is verified using one or more of the following strategies: wear-blocking verification, thermal shock protection, and silent duration; wherein... The wear prevention verification includes: reconstructing the local flow velocity of the proposed operation area in the first soot blowing control strategy through the physical information neural network model; if the local flow velocity of the proposed operation area does not exceed the preset wear critical flow velocity threshold, the verification passes; if the local flow velocity of the proposed operation area exceeds the preset wear critical flow velocity threshold, it is determined that there is a wear risk, and the verification fails. The thermal shock protection includes: calculating the third thermal cleanliness factor of the proposed operating area in the first soot blowing control strategy through the physical information neural network model; if the third thermal cleanliness factor of the proposed operating area is not higher than the preset thermal shock threshold, the verification passes; if the third thermal cleanliness factor of the proposed operating area is higher than the preset thermal shock threshold, it is determined that there is a risk of thermal shock, and the verification fails. The silent duration strategy includes: recording the time since the last soot blowing when the soot blowing gun selected in the first soot blowing control strategy has not been blown; if the time since the last soot blowing is not less than the preset soot blowing silent duration threshold, the verification passes and soot blowing must be carried out; if the time since the last soot blowing is less than the preset soot blowing silent duration threshold, the verification fails.

6. The boiler heating surface soot blowing control method according to any one of claims 1 to 5, characterized in that, If the verification passes, before using the first soot blowing control strategy as the target soot blowing control strategy, the method further includes: If the verification passes, determine whether there is a serious blockage. If no serious blockage is found, execute the operation of using the first soot blowing control strategy as the target soot blowing control strategy; If severe blockage exists, the first soot blowing control strategy is modified to obtain a second soot blowing control strategy based on the principle of prioritizing unblocking, and the second soot blowing control strategy is used as the target soot blowing control strategy.

7. A boiler heating surface soot blowing control device, characterized in that, include: The reconstruction unit is used to construct and train a physical information neural network model, which includes a cascaded and coupled furnace radiation zone sub-model and tail flue convection zone sub-model. The furnace radiation zone sub-model is used to simulate the physical field state inside the furnace and the outlet section, and the tail flue convection zone sub-model is used to simulate the physical field state inside the tail flue. The physical field state includes temperature field, flow velocity field and pressure field. The first calculation unit is used to calculate the first thermal cleanliness factor of the furnace water-cooled wall based on the physical field state inside the furnace output by the furnace radiation zone sub-model simulation. The second calculation unit is used to calculate the second thermal cleanliness factor and hydraulic cleanliness factor of the tail flue heating surface based on the physical field state inside the tail flue output by the simulation output of the tail flue convection zone sub-model. The matrix construction unit is used to construct a cleanliness matrix of the entire field based on the first thermal cleanliness factor of the furnace heating surface, the second thermal cleanliness factor of the tail flue heating surface, and the hydraulic cleanliness factor. The decision unit is used to take the overall cleanliness matrix, overall flow velocity distribution, overall pressure gradient distribution, current unit load, load change rate and the previous soot blowing operation record as the state space, and input the state space into the trained reinforcement learning decision model to obtain the optimization result as the first soot blowing control strategy. The reinforcement learning decision model is configured with a multi-dimensional reward function; A verification unit is used to verify the first soot blowing control strategy; The determining unit is used to take the first soot blowing control strategy as the target soot blowing control strategy when the verification passes. The control unit is used to control the issuance of soot blowing commands corresponding to the target soot blowing control strategy; The correction unit is used to acquire the response data of the boiler's real physical field after the target soot blowing control strategy is executed, and to perform online correction of the physical information neural network model based on the response data.

8. The boiler heating surface soot blowing control device according to claim 7, characterized in that, Also includes: The judgment unit is used to determine whether there is a serious blockage when the verification is passed and before the determination unit uses the first soot blowing control strategy as the target soot blowing control strategy. If there is no serious blockage, the determining unit is triggered to execute the operation of using the first soot blowing control strategy as the target soot blowing control strategy. The correction unit is used to correct the first soot blowing control strategy to obtain a second soot blowing control strategy when the verification passes and the judgment unit determines that there is a serious blockage, based on the principle of prioritizing unblocking, and to use the second soot blowing control strategy as the target soot blowing control strategy.

9. An electronic device, characterized in that, It includes a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the boiler heating surface soot blowing control method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to perform the boiler heating surface soot blowing control method according to any one of claims 1 to 6.

Images