Flue gas flow field dynamic adjusting system based on CFD simulation

Abstract

The invention relates to the technical field of dynamic regulation, in particular to a CFD (computational fluid dynamics) simulation-based flue gas flow field dynamic regulation system, which comprises a multi-dimensional data collaborative acquisition module, an anti-ash deposition sensor array is deployed at a key position of a flue, and a time-space fusion input data set is generated; the dual-mechanism CFD dynamic coupling module operates the main CFD prediction model and the simplified fluid network model in parallel, compares output differences through an evidence theory fusion algorithm, corrects boundary conditions and reconstructs network parameter characteristics when the confidence coefficient is lower than a threshold value, and updates a dynamic parameter knowledge base; the multi-target self-adaptive regulation and control module is used for solving a Pareto optimal strategy meeting the safety constraint by taking the baffle opening degree and the pump frequency as variables based on the updating model, driving an actuator and feeding back flow field response data; and the system forms a closed-loop adaptive mechanism, so that the CFD model continuously tracks the degradation state of the equipment, and the effectiveness of the regulation and control strategy is maintained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dynamic adjustment, in particular to a flue gas flow field dynamic adjustment system based on CFD simulation. BACKGROUND

[0002] In the research of industrial flue gas flow field, the complex three-dimensional flow characteristics of flue gas include temperature gradient, turbulent mixing and pollutant diffusion, etc. Traditional physical experiments are difficult to fully capture the details, so simulation based on computational fluid dynamics method provides an effective technical means. By discretizing the basic equations of fluid mechanics, using numerical solution techniques such as finite volume or finite element, the velocity field, temperature distribution and concentration field of flue gas are predicted, and the random fluctuation phenomenon is handled by using Reynolds averaged equation or large eddy simulation turbulence model. Such simulation supports engineers to evaluate the impact of emissions on the environment, such as identifying vortex regions at the outlet of the chimney to prevent the accumulation of pollutants, thereby guiding the design of flue structure and the adjustment of operating parameters, improving system energy efficiency and reducing the release of harmful substances, and ultimately promoting compliance with environmental protection regulations.

[0003] There are spatial and temporal differences between the CFD simulation results under dynamic conditions and the actual flue gas flow field characteristics: since the equipment state and operating conditions of the flue gas system may change significantly over time, the geometric boundary conditions and physical property parameters relied on by the initial CFD model will gradually deviate from the actual state, resulting in systematic deviations between the simulated flow field distribution and the real-time physical process, so that the adjustment strategies relying on offline simulation results fail in engineering applications. Typical cases are that the flow cross-section of the primary flue gas cooler is reduced and the heat transfer efficiency is decreased due to the accumulation of ash and hardening, at this time, if the CFD model constructed under clean operating conditions is still used to guide the opening degree of the flue damper, the generated flow field optimization instructions will be incorrect due to the model's failure to reflect the actual increase in flow resistance and heat transfer deterioration, which not only cannot eliminate the risk of cold end corrosion, but may even accelerate the deposition of ash in local areas and form a vicious cycle. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a flue gas flow field dynamic adjustment system based on CFD simulation, which solves the problem that the CFD offline model is inaccurate due to the drift of the equipment state in the long-term operation of the coal-fired power plant system, thereby causing the failure of the control instruction.

[0005] To solve the above technical problems, the specific technical solutions of the present application are as follows:

[0006] The flue gas flow field dynamic adjustment system based on CFD simulation provided by the present application comprises:

[0007] A data collaborative collection module is configured to deploy a sensor array at a preset position of the flue gas system, to obtain flue gas temperature, flow rate and pollutant concentration data with spatial coordinate information in real time, to synchronously integrate unit operation parameters, and to generate a spatio-temporal fusion input data set after time alignment processing;

[0008] A dual-mechanism CFD dynamic coupling module is connected to the data collaborative collection module, receives the input data set, is configured to run a main CFD prediction model and a simplified fluid network model in parallel, compares the difference between the flow field distribution prediction value output by the main CFD prediction model and the node transmission characteristics generated by the simplified fluid network model at the sensor position through an evidence theory fusion algorithm, corrects the boundary conditions of the main CFD prediction model and reconstructs the parameter characteristics of the simplified fluid network model when the confidence level is lower than a set threshold, and updates the equipment state parameters in the dynamic parameter knowledge base;

[0009] A multi-objective adaptive regulation module is connected to the dual-mechanism CFD dynamic coupling module and the dynamic parameter knowledge base, receives the updated model information, is configured to use a multi-objective optimization solving method, takes the flue gas damper opening combination and the frequency of the heat medium water circulating pump as the adjustment variables, takes the optimization of heat exchange efficiency, flow field uniformity and equipment risk suppression as the objective function, solves the optimal regulation strategy instruction under the condition of meeting the physical safety constraint, drives the field actuator to act, and acquires the flow field response data after execution to update the equipment characteristic parameters in the dynamic parameter knowledge base;

[0010] A digital twin parallel deduction system is connected to the dual-mechanism CFD dynamic coupling module and the dynamic parameter knowledge base, is configured to clone the main CFD prediction model, load the expected state parameters to construct a virtual environment, deduce the long-term operation effect of different regulation strategies in the virtual environment, compare and select the Pareto optimal solution through evaluation indexes, output the Pareto optimal solution to the multi-objective adaptive regulation module, and monitor the deviation between the real flow field data and the virtual prediction, and trigger the parameter backtracking analysis and model retraining of the dual-mechanism CFD dynamic coupling module when the deviation continuously exceeds the limit.

[0011] Further, the flue gas flow field dynamic regulation system based on CFD simulation disclosed in the application is characterized in that the data collaborative collection module is configured as follows:

[0012] High-frequency thermocouple arrays, differential pressure flowmeters and laser spectrum pollutant analyzers are deployed at key sections of the outlet of the air preheater, the flow channel of the first flue gas cooler, the flow channel of the second flue gas cooler and the entrance dome area of the chimney.

[0013] In the flue gas cooler heat exchange tube bundle, a micro electro-mechanical system hot film anemometer with a dustproof self-cleaning function is used to form a micro sensor cluster.

[0014] Further, the CFD simulation-based flue gas flow field dynamic adjustment system disclosed by the application, the processing of the input data set generated by the data collaborative acquisition module comprises:

[0015] The temperature, flow rate, pollutant concentration data and unit operation parameters from different sources are subjected to space-time registration to eliminate coordinate system deviation;

[0016] The registered heterogeneous data is fused according to the dynamic weight assigned according to the sensor confidence, the influence of abnormal data is reduced, and the input data set subjected to space-time fusion is generated.

[0017] Further, the CFD simulation-based flue gas flow field dynamic adjustment system disclosed by the application, the double-mechanism CFD dynamic coupling module is configured to:

[0018] The main CFD prediction model adopts a non-steady Realizable k-ε turbulence model coupled with a component transport equation for solving, and an enhanced wall treatment is applied to capture near-wall flow;

[0019] The simplified fluid network model abstracts the flue system as a weighted topological graph, and the pressure drop flow relationship between nodes is calculated in real time through a graph convolution network.

[0020] Further, the CFD simulation-based flue gas flow field dynamic adjustment system disclosed by the application, the evidence theory fusion algorithm adopts a Dempster-Shafer synthesis rule to process the output difference of the model;

[0021] The correction of boundary conditions includes updating the local resistance coefficient and wall heat flux density of the heat exchanger;

[0022] The reconstruction of parameter characteristics includes updating the edge weight matrix of the weighted topological graph.

[0023] Further, the CFD simulation-based flue gas flow field dynamic adjustment system disclosed by the application, in the multi-objective adaptive regulation and control module:

[0024] The multi-objective optimization solving method adopts a particle swarm algorithm with adaptive inertia weight and elite preservation strategy;

[0025] The physical safety constraints include prohibiting flue gas backflow and preventing component overheating.

[0026] Further, the CFD simulation-based flue gas flow field dynamic adjustment system disclosed by the application, the closed-loop processing of the multi-objective adaptive regulation and control module comprises:

[0027] The optimal regulation and control strategy instruction obtained by solving is filtered according to a preset priority by a fuzzy decision maker, and drives the action of the pneumatic regulating valve and the frequency converter;

[0028] The flow field response data of the response action is collected and returned for updating the baffle flow characteristic curve and the heat exchanger efficiency parameter in the dynamic parameter knowledge base.

[0029] Further, the virtual environment constructed in the digital twin parallel deduction system of the flue gas flow field dynamic adjustment system based on CFD simulation includes:

[0030] Clone the main CFD prediction model, load the expected state parameters generated based on the equipment degradation prediction model;

[0031] The time compression factor is set by applying the turbulent flow characteristic time scale similarity criterion to accelerate the long-term running effect simulation.

[0032] Further, the flue gas flow field dynamic adjustment system based on CFD simulation also includes:

[0033] The root mean square error of the real universe monitoring data and the virtual universe prediction value is calculated;

[0034] When the root mean square error exceeds the set limit value for multiple times in succession, the parameter backtracking analysis is triggered to locate the key error source, and the model is driven to retrain and update the simplified fluid network model weight.

[0035] Further, the flue gas flow field dynamic adjustment system based on CFD simulation also includes:

[0036] The data coordination collection module generates the input data set with dynamic weight, the real-time updated dynamic parameter knowledge base established by the double-mechanism CFD dynamic coupling module, and the flow field response data fed back by the multi-objective adaptive regulation module, and cooperates to act on the long-term deduction accuracy of the digital twin parallel deduction system;

[0037] The Pareto optimal solution selected by the digital twin parallel deduction system and the key error source identified guide the real-time decision priority of the multi-objective adaptive regulation module and the retraining focus of the double-mechanism CFD dynamic coupling module respectively, forming a cooperative adaptive enhancement mechanism under the condition of continuous drift of equipment state.

[0038] The present application has the following advantages:

[0039] The present application has the beneficial effect of systematically solving the model misalignment problem caused by equipment state drift: a high-precision spatiotemporal dataset is generated by the anti-ash accumulation sensor array to provide a real physical basis for model correction; a dual-mechanism model cross-validation architecture uses evidence theory fusion algorithm to detect master-slave model deviation, trigger boundary condition and macro parameter collaborative correction, so that the dynamic parameter knowledge base reflects the degradation state such as ash corrosion in real time; a multi-objective optimization strategy based on the refined model combines digital twin growth simulation to output the Pareto optimal solution that takes into account immediate regulation and long-term stability, avoiding the regulation failure caused by traditional offline models; the execution feedback and virtual reality deviation monitoring constitute a closed-loop reinforcement learning mechanism to continuously calibrate the equipment characteristic parameters and drive model retraining. The system forms an adaptive positive feedback loop, significantly improving the effectiveness and reliability of flow field regulation in continuous operation conditions of coal-fired power plants. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, for those skilled in the art, other drawings can also be obtained from the drawings without creative labor.

[0041] Figure 1 The system architecture diagram of the flue gas flow field dynamic regulation system based on CFD simulation provided by the embodiments of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present application clearer, the following will combine the specific embodiments of the present application and the corresponding drawings to clearly and completely describe the technical solutions of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The following will combine the drawings to specifically describe the technical solutions provided by the embodiments of the present application. In order to better understand the purpose of the present application, the following will further describe the present application in detail.

[0043] Please refer to Figure 1 The flue gas flow field dynamic regulation system based on CFD simulation provided by the present application comprises:

[0044] The data collaborative acquisition module installs a sensor network at key locations in the flue system. Specifically, high-frequency thermocouple arrays, differential pressure flow meters, and laser spectroscopy pollutant analyzers are arranged at the outlet of the air preheater, inside the flow channel of the first / second flue gas cooler, and at the entrance dome area of the chimney. To cope with the high-dust environment, a cluster of micro-sensors composed of MEMS hot-film anemometers with dust-proof coating and self-cleaning air path design is specially deployed in areas prone to dust accumulation, such as the windward and leeward sides of the flue gas cooler heat exchange tube bundle. This module collects real-time raw signals of flue gas temperature, flow rate, and pollutant concentration with three-dimensional spatial coordinate markers at these location points, and simultaneously acquires unit operation parameters such as boiler load, heat medium water flow, and induced draft fan frequency through industrial standard communication protocols (such as OPC-UA). For heterogeneous data streams from different physical locations and signal sources, the module first performs a space-time registration operation to calibrate the rigid transformation of the sensor measurement coordinate system, eliminating spatial coordinate shifts caused by installation deviations; then, based on the operating state and historical accuracy of each sensor, it assesses its confidence (such as considering the degree of thermocouple aging) and dynamically assigns fusion weights to data from different sources, reducing the influence of abnormal measurement values. After completing the above registration and weighted fusion, the module generates a four-dimensional input data set that integrates spatial location information and time series evolution, providing a high-precision, high-robustness data foundation for subsequent modeling.

[0045] The dual-mechanism CFD dynamic coupling module receives the input dataset from the data co-sampling module and adopts a master-slave model co-verification architecture. In this module, the master CFD prediction model is built based on the unsteady Realizable k-ε turbulence model, coupled with the component transport equation describing the pollutant diffusion and the physical model reflecting the wall heat exchange process. The calculation grid adopts the Poly-Hexcore hybrid topology structure with adaptive boundary layer encryption to accurately capture the complex flow in the near-wall region. At the same time, the module runs a lightweight simplified fluid network model in parallel. This model abstracts the entire flue system as a weighted topological graph structure, where the nodes represent key monitoring locations, and the connection edge weights between nodes are dynamically calculated by the real-time running graph convolution network, used to represent the macroscopic pressure drop-flow transfer characteristics between nodes. The core of the module is the dynamic parameter co-optimization mechanism: it compares the micro flow field distribution (velocity vector field, temperature field) output by the master CFD prediction model with the macro node transfer characteristics generated by the simplified fluid network model at the actual sensor location. This comparison is processed by the evidence theory fusion algorithm (specifically using the Dempster-Shafer combination rule) from the output evidence of the two models. When the confidence calculated by the fusion result is lower than the preset threshold (for example, 0.85), it indicates that there is a significant deviation between the model prediction and the actual monitoring data, at which time the module triggers the correction process. The correction operation includes updating the local resistance coefficient and wall heat flux density of key equipment (such as heat exchangers) in the master CFD model according to the measured pressure difference and other data, and simultaneously reconstructing the pipe segment impedance characteristics represented by the topological graph edge weight matrix in the simplified fluid network model. All the revised time-varying parameters (such as the change of wall roughness caused by ash deposition, the evolution of heat flux density) are continuously stored in the dynamic parameter knowledge base. This mechanism enables the model to continuously adapt to equipment state drift.

[0046] The multi-objective adaptive control module connects and receives the updated model information from the dual-mechanism CFD dynamic coupling module and the equipment characteristic parameters in the dynamic parameter knowledge base. The core of this module is an intelligent decision-making kernel, which uses an improved multi-objective particle swarm optimization algorithm for solving. This algorithm takes the adjustable aerodynamic damper opening combination in the flue gas system and the operating frequency of the heat medium water circulating pump as the optimization variables, and sets multiple mutually restrictive objective functions, including maximizing the effective utilization rate of the flue gas waste heat recovery equipment, minimizing the unevenness of the temperature distribution in the flue (temperature variance), and suppressing the corrosion risk index in a specific area (such as the low-temperature section). The optimization solving process strictly embeds a physical constraint rule engine, which explicitly prohibits operating conditions that may cause flue gas backflow, equipment overheating, or other operating hazards. The algorithm itself integrates an adaptive inertia weight adjustment mechanism and an elite reservation strategy to balance the global search and local development capabilities, and to avoid falling into a local optimal solution. The optimal control strategy instructions obtained by solving (usually a solution set on the Pareto frontier) are input into a fuzzy decision maker, which selects the final instructions to be executed according to the pre-set operating priority strategy (such as giving priority to corrosion prevention over energy efficiency improvement). These instructions are sent to the on-site aerodynamic regulating valve, frequency converter, and other actuators to drive their actions. After the actions are completed, the module obtains the flow field response data (such as the changed temperature and flow velocity distribution) through the data cooperative acquisition module in real time. These response data reflecting the actual control effect are fed back to update the characteristic parameters of the relevant equipment in the dynamic parameter knowledge base, such as the characteristic curve of the damper opening and the flow rate relationship, the actual efficiency parameters of the heat exchanger in the state of dust accumulation, etc., forming a closed-loop learning link.

[0047] The digital twin parallel deduction system constructs a virtual environment parallel to the real system. It first clones the master CFD prediction model in the dual-mechanism CFD dynamic coupling module as the basis, and then loads the expected state parameters of the equipment predicted and generated based on the equipment physical degradation model (such as considering the corrosion rate and the dust accumulation growth model) to construct a virtual universe. In order to efficiently evaluate the long-term impact of the control strategy (such as the dust distribution and corrosion hot spot after continuous operation for hundreds of hours), the system applies the time scale similarity criterion of turbulent flow characteristics to set the time compression factor, and accelerates the simulation of the virtual universe. In the virtual environment, the system pre-represents multiple groups of candidate control strategies generated by the multi-objective adaptive control module, and advances to the set equivalent running time. After the deduction is completed, the system extracts the indicators reflecting the long-term performance (such as the heat exchange tube bundle dust coverage rate and the corrosion risk thermal map distribution) and evaluates them. Through multi-attribute decision-making methods (such as TOPSIS), the comprehensive score of each strategy is calculated, and the Pareto optimal solution that takes into account the immediate control effect and the long-term operation stability is selected, and output to the multi-objective adaptive control module as the preferred solution for real execution. At the same time, the system continuously monitors the deviation between the actual flow field data obtained from the real universe (through the data collaborative acquisition module) and the predicted values of the virtual universe. The system calculates the root mean square error of the key indicators as the deviation measure. When the root mean square error exceeds the set limit value for multiple times (such as 3 times), it indicates that the prediction of the virtual model deviates too much from the actual system. At this time, the system triggers the parameter backtracking analysis process, locates the key time-varying parameters (such as the flow coefficient deviation of a certain baffle) in the dynamic parameter knowledge base that contribute most to the prediction deviation, and drives the dual-mechanism CFD dynamic coupling module to start the model retraining program, such as updating the weight parameters of the graph convolution network in the simplified fluid network model.

[0048] The whole system forms a closed-loop reinforcement learning mechanism under data-driven. The data collaborative acquisition module provides reliable input after high-precision, registration and weighting processing. The dual-mechanism CFD dynamic coupling module realizes the dynamic collaborative optimization of model boundary conditions and macroscopic characteristics through cross-validation and evidence fusion of master-slave models, and continuously corrects the dynamic parameter knowledge base. The multi-objective adaptive control module generates optimization strategies based on accurate models and executes them, and the feedback flow field response data further calibrate the knowledge base. The digital twin parallel deduction system uses knowledge base information and cloned models for long-term pre-representation and evaluation, and the Pareto optimal solution selected guides real-time control, and the key error source identified drives model retraining focus. The modules of the system are closely coordinated through data flow (input data set, dynamic parameter knowledge base update, flow field response data) and instruction flow (optimal control strategy, model retraining trigger), and together form a self-enhancing closed-loop system that can continuously adapt to equipment state drift and maintain model accuracy and control effect.

[0049] At the key monitoring positions of the outlet cross-section of the air preheater, the internal flow passage of the first-stage flue gas cooler, the internal flow passage of the second-stage flue gas cooler, and the dome area of the chimney inlet, an array of high-frequency response thermocouples is arranged to measure the temperature spatiotemporal distribution, a differential pressure flowmeter is installed to capture the main flow rate variation, and a laser spectrum pollutant analyzer is arranged to track the pollutant component concentration in real time. In particular, for the windward side and leeward side of the heat exchange tube bundle of the flue gas cooler and other areas prone to dust accumulation, a micro sensor cluster is formed by using a micro electro-mechanical system hot-film anemometer. The micro sensor integrates a dustproof coating protective layer and is designed with a periodic self-cleaning air path structure, and through the control of a micro electromagnetic valve, compressed air pulse blowing is maintained to maintain the measurement accuracy in a dusty environment.

[0050] The data collaborative acquisition module first performs spatiotemporal registration processing on the monitoring data from different sources such as thermocouples, flowmeters, and pollutant analyzers. Spatial registration converts the local coordinate system of each sensor to the global coordinate system of the flue through a coordinate transformation matrix, eliminating the measurement position deviation caused by installation deviation; the time registration uses an interpolation synchronization algorithm to align the timestamps of the distributed control system. Subsequently, the module evaluates the sensor health status (based on calibration records, running time, and historical failure rate) to generate a confidence index, and assigns dynamic fusion weights to the temperature, flow rate, pollutant concentration, and boiler load parameters after registration. Finally, a weighted fusion algorithm is used to generate an input data set with a unified spatiotemporal label, suppressing abnormal data interference.

[0051] In the dual-mechanism CFD dynamic coupling module, the main CFD prediction model uses the unsteady Realizable k-ε turbulence model to construct the control equation set, couples the component transport equation to describe the pollutant diffusion process, and applies an enhanced wall treatment technique in the near-wall region to capture the velocity gradient and temperature boundary layer through an adaptive grid in the boundary layer. The simultaneously running simplified fluid network model abstracts the flue into a node-edge topology graph, with nodes corresponding to key monitoring positions such as the outlet of the air preheater and the flow passage of the cooler, and edge weights being dynamically calculated by a real-time graph convolution network processor to generate a macroscopic pressure drop-flow rate transfer characteristic matrix between nodes.

[0052] When the difference in output between the main model calculated by the evidence theory fusion algorithm (Dempster-Shafer combination rule) and the simplified model is below the 0.85 threshold, the system triggers a boundary condition correction process: the local resistance coefficient of the heat exchanger is optimized in reverse according to the measured pressure difference data, and the heat flux boundary value is updated according to the wall thermocouple readings. The parameter characteristics of the simplified fluid network model are simultaneously reconstructed, and the edge matrix coefficient of each pipe segment in the topology graph is recalculated through the flow balance equation, maintaining the consistency of the macro-micro models.

[0053] The multi-objective adaptive control module adopts an improved particle swarm optimization algorithm for optimization solution. The algorithm integrates an adaptive inertia weight mechanism, dynamically adjusts the search step size in iterations according to the convergence state, and sets an elite solution reservation strategy to maintain population diversity. The optimization process strictly follows the limited conditions of the physical constraint rule base, including prohibiting any baffle opening combination that causes reverse flow of flue gas, and preventing the generation of operation instructions that exceed the material tolerance limit of the temperature at the monitoring point.

[0054] The optimal strategy solution set obtained by solving is input into the fuzzy decision maker, and the final execution scheme is selected according to the preset priority rules (such as the corrosion prevention weight being higher than the energy efficiency optimization). The selection instruction drives the field pneumatic control valve to adjust the baffle angle, and synchronously controls the variable frequency speed regulation of the heat medium water circulating pump. After the control action is implemented, the module re-collects the temperature field and flow field distribution at the key positions of the flue as response feedback data, and reversely calibrates the equipment characteristic parameters in the dynamic parameter knowledge base, especially updates the coefficient matrix of the baffle opening-flow characteristic curve, and corrects the effective heat transfer coefficient of the heat exchanger under the condition of ash deposition.

[0055] The digital twin parallel deduction system clones the main CFD prediction model as the virtual environment infrastructure, loads the expected state parameters generated by the equipment degradation prediction model (including the pipe wall thinning data calculated based on the corrosion dynamics equation and the surface roughness predicted by the ash growth model). To accelerate the simulation of long-term effects, the system analyzes the characteristic time scale of the main frequency of turbulent fluctuations, and sets a time compression scale factor accordingly, so that 1 hour of calculation time in the virtual environment is equivalent to simulating the actual system for tens of hours.

[0056] The system continuously calculates the root mean square error of the monitored data of the real system and the virtual prediction value in key parameters (such as temperature distribution and pressure drop). When this error exceeds the tolerance threshold for three consecutive times, the parameter backtracking analysis program is triggered: through the sensitivity contribution degree algorithm, the time-varying parameter item (such as the flow coefficient offset of a specific baffle) that has the greatest impact on the deviation in the dynamic parameter knowledge base is located. After positioning, drive the simplified fluid network model to retrain the process, and use the new data set to update the edge weight calculation weight of the graph convolution network.

[0057] The input data set with confidence weight generated by the data collaborative collection module provides a high-fidelity initial field for virtual deduction; the equipment state parameters updated by the dynamic parameter knowledge base ensure the authenticity of the degradation simulation; and the flow field response data fed back by the multi-objective adaptive control module provide verification samples for the control effect. The three work together to improve the long-term deduction accuracy of the digital twin system. At the same time, the Pareto optimal solution output by the digital twin system guides the decision priority allocation of real-time control; and the key error sources identified by it focus on the retraining calculation resources of the double-mechanism CFD dynamic coupling module. Form a cooperative adaptive enhanced closed loop under the condition of continuous drift of equipment state.

[0058] The flue gas flow field dynamic adjustment system provided by the application solves the model misalignment problem caused by equipment state drift through the following technical logic:

[0059] Firstly, the system establishes a real-time data-driven model correction mechanism. By deploying an anti-deposition sensor array at key positions in the flue to obtain high spatiotemporal resolution flow field data, combined with unit operation parameters to generate input data sets. The double mechanism CFD dynamic coupling module runs the main CFD model and the simplified fluid network model in parallel, and uses the evidence theory fusion algorithm to compare the output differences of the two models at the sensor nodes. When the confidence level is below the threshold, the boundary conditions of the main model are corrected in reverse (such as updating the local resistance coefficient of the heat exchanger) and the topology parameters of the simplified model are reconstructed, so that the equipment state parameters in the dynamic parameter knowledge base can reflect the actual degree of deterioration in real time, and the model parameter drift is eliminated from the root cause.

[0060] Secondly, the system constructs a long-term verification mechanism for the control strategy. When the multi-objective adaptive control module solves the optimal adjustment instruction based on the refined model, the digital twin parallel deduces the system to clone the model and load the expected degradation parameters of the equipment to construct a virtual environment. The long-term effect of the control strategy (such as 500 hours of ash deposition distribution) is pre-visualized under accelerated simulation conditions, and the Pareto optimal solution that takes into account both immediate effectiveness and long-term stability is selected through cross-universe evaluation. This mechanism enables the control decision to respond to current state changes and predict equipment degradation trends, avoiding strategy failure due to model misalignment.

[0061] Finally, the system forms a closed-loop reinforcement learning mechanism. The flow field response data collected after the execution of the control strategy is fed back to the dynamic parameter knowledge base for continuous calibration of the equipment characteristic curve; at the same time, the digital twin system monitors the deviation between virtual prediction and real data in real time, and triggers parameter backtracking analysis to locate the key error source when the deviation exceeds the limit, driving the simplified model to retrain. This closed loop enables the model accuracy, control effect and equipment state change to form a co-evolution, maintaining the adaptive ability of the system in the long run.

[0062] The specific embodiments of the application are based on the actual operation scene of the flue gas system of a coal-fired power plant. A sensor network composed of high-frequency thermocouple arrays, differential pressure flowmeters and laser spectrum pollutant analyzers is deployed at the outlet of the air preheater, the internal flow passage of the first / second flue gas cooler and the dome area of the chimney inlet; for positions prone to ash deposition such as the windward surface of the flue gas cooler heat exchange tube bundle, a cluster of micro-electro-mechanical system hot film anemometers with a dustproof coating and a self-cleaning air path is installed. The output signals of each sensor are time and space registered to eliminate coordinate offset, and dynamic weights are allocated according to the degree of equipment aging (such as the weight of a thermocouple serving for more than 2 years being reduced to 0.7), to generate an input data set with a three-dimensional space label.

[0063] When the primary flue gas cooler's flow cross-section shrinks due to ash accumulation, the dual-mechanism model's dynamic coupling module detects an anomaly: the pressure drop data predicted by the main CFD model (coupled with an unsteady Realizable k-ε turbulence model and enhanced wall treatment) deviates significantly from the macroscopic transfer characteristics calculated by the simplified fluid network model (topology graph constructed based on graph convolutional networks). The evidence theory fusion algorithm (Dempster-Shafer combination rule) calculates a confidence level below the 0.85 threshold, triggering a boundary condition correction: the cooler's local resistance coefficient is updated to the actual value of 1.8 times in the opposite direction based on the measured pressure difference, and the edge weight matrix of the corresponding pipe segment in the topology graph is reconstructed simultaneously. The updated device state parameters (such as the wall roughness increased to 120 μm) are stored in the dynamic parameter knowledge base.

[0064] The multi-objective adaptive control module generates control strategies based on the corrected model: a particle swarm optimization algorithm with adaptive inertia weight is used to optimize the baffle opening combination, and under the constraints of "flue gas backflow prohibition" (flow rate > 0.3 m / s) and "pipe wall over-temperature prohibition" (< 450 ℃), the Pareto solution set that maximizes the heat exchange efficiency is solved. The strategy is screened by a fuzzy decision maker (anti-corrosion weight accounts for 70%) and drives the pneumatic control valve to act. The collected temperature distribution data after execution are fed back to the knowledge base, and the slope of the baffle flow characteristic curve is updated to the corrected value of 0.68. At the same time, the digital twin system clones the main model, loads the pipe wall corrosion parameters predicted based on the Arrhenius equation, and applies the turbulent time scale similarity criterion (compression ratio 1:50) to deduce the equivalent running effect for 500 hours. When the root mean square error between the virtual predicted ash thickness (12 mm) and the actual monitored value (15 mm) exceeds the limit for three consecutive times, the system backtracks and locates the heat transfer coefficient in the dynamic parameter knowledge base as the key error source, triggering the simplified model to retrain and update the graph network weights, forming a closed-loop adaptive mechanism for equipment degradation.

Claims

1. A flue gas flow field dynamic adjustment system based on CFD simulation, characterized in that, The application relates to a system for real-time monitoring and adaptive control of a flue gas desulfurization (FGD) system, which comprises the following modules: a data collaborative acquisition module configured to deploy a sensor array at a preset position of a flue gas system, to obtain flue gas temperature, flow rate and pollutant concentration data with spatial coordinate information in real time, to synchronously integrate unit operation parameters, and to generate a spatio-temporal fusion input data set after time alignment processing; a dual-mechanism CFD dynamic coupling module connected to the data collaborative acquisition module, receiving the input data set, configured to run a main CFD prediction model and a simplified fluid network model in parallel, to compare the difference between the flow field distribution prediction value output by the main CFD prediction model and the node transmission characteristics generated by the simplified fluid network model at the sensor position through an evidence theory fusion algorithm, and to correct the boundary conditions of the main CFD prediction model and reconstruct the parameter characteristics of the simplified fluid network model when the confidence level is lower than a set threshold, and to update the equipment state parameters in the dynamic parameter knowledge base; a multi-objective adaptive control module connected to the dual-mechanism CFD dynamic coupling module and the dynamic parameter knowledge base, receiving the updated model information, configured to use a multi-objective optimization solving method, to take the flue gas damper opening combination and the frequency of the heat medium water circulating pump as the adjustment variables, to take the optimization of heat exchange efficiency, flow field uniformity and equipment risk suppression as the objective function, to solve the optimal control strategy instruction under the condition of meeting the physical safety constraint, to drive the field actuator to act, and to obtain the flow field response data after execution and feed back to the dynamic parameter knowledge base to update the equipment characteristic parameters; a digital twin parallel deduction system connected to the dual-mechanism CFD dynamic coupling module and the dynamic parameter knowledge base, configured to clone the main CFD prediction model and load the expected state parameters to construct a virtual environment, to deduce the long-term operation effect of different control strategies in the virtual environment, to select the Pareto optimal solution through evaluation index comparison and output to the multi-objective adaptive control module, and to monitor the deviation between the real flow field data and the virtual prediction, and to trigger the parameter backtracking analysis and model retraining of the dual-mechanism CFD dynamic coupling module when the deviation continuously exceeds the limit.

2. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 1, characterized in that, The data collaborative acquisition module is configured as follows: high-frequency thermocouple arrays, differential pressure flowmeters and laser spectrum pollutant analyzers are arranged at the key sections of the air preheater outlet, the flow channel of the first flue gas cooler, the flow channel of the second flue gas cooler and the dome area of the chimney inlet; micro electro mechanical system hot film anemometers with dustproof self-cleaning function are used to form micro sensor clusters in the easy dust deposition area of the flue gas cooler heat exchange tube bundle.

3. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 2, characterized in that, The processing of the data collaborative acquisition module to generate the input data set comprises the following steps: the temperature, flow rate, pollutant concentration data and unit operation parameters from different sources are subjected to spatio-temporal registration to eliminate coordinate system deviation; the registered heterogeneous data are fused by assigning dynamic weights according to the sensor confidence level, the influence of abnormal data is reduced, and the spatio-temporal fusion input data set is generated.

4. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 3, characterized in that, The dual-mechanism CFD dynamic coupling module is configured as follows: the main CFD prediction model adopts a non-steady Realizable k-epsilon turbulence model coupled with a component transport equation, and an enhanced wall surface treatment is applied to capture near-wall flow. The simplified fluid network model abstracts the flue system as a weighted topological graph, and calculates the pressure drop flow relationship between nodes in real time through a graph convolution network.

5. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 4, characterized in that, The evidence theory fusion algorithm adopts the Dempster-Shafer combination rule to process the model output differences; The revised boundary conditions include updating the local resistance coefficient of the heat exchanger and the wall heat flux density; The reconstructed parameter characteristics include updating the edge weight matrix of the weighted topological graph.

6. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 5, wherein, In the multi-objective adaptive control module: The multi-objective optimization solution method adopts a particle swarm optimization algorithm with adaptive inertia weight and elite preservation strategy; The physical safety constraints include prohibiting flue backflow and preventing component overheating.

7. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 6, characterized in that, The closed-loop processing of the multi-objective adaptive control module includes: The optimal control strategy instructions obtained by solving are filtered according to the preset priority by the fuzzy decision maker, and drive the action of the pneumatic regulating valve and the frequency converter; The flow field response data in response to the action are collected and fed back, which are used to update the baffle flow characteristic curve and the heat exchanger efficiency parameter in the dynamic parameter knowledge base.

8. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 1, wherein, In the digital twin parallel deduction system, the construction of the virtual environment includes: Cloning the main CFD prediction model, loading the expected state parameters generated based on the device degradation prediction model; Applying the turbulent characteristic time scale similarity criterion to set the time compression factor to accelerate the long-term running effect simulation.

9. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 8, wherein, Further comprising: Calculating the root mean square error of the real universe monitoring data and the virtual universe prediction value; When the root mean square error exceeds the set limit value for multiple times in succession, triggering parameter backtracking analysis to locate the key error source, and driving the model retraining to update the simplified fluid network model weight.

10. The CFD simulation based flue gas flow field dynamic adjustment system according to claim 9, wherein, Further comprising: The data coordination acquisition module generates the input data set with dynamic weight, the dynamic parameter knowledge base updated in real time by the double-mechanism CFD dynamic coupling module, and the flow field response data fed back by the multi-objective adaptive control module, which cooperatively act on the long-term deduction accuracy of the digital twin parallel deduction system; The Pareto optimal solution selected by the digital twin parallel deduction system and the key error source identified guide the real-time decision priority of the multi-objective adaptive control module and the retraining focus of the double-mechanism CFD dynamic coupling module respectively, forming a cooperative adaptive enhancement mechanism under the condition of continuous drift of the device state.

Images