Heat supply period extraction condensing unit high back pressure optimization control system and method

Abstract

The invention provides a heat supply period extraction condensing unit high back pressure optimization control system and method, and the system comprises an agent modeling implementation module which is used for building a multi-agent model of a boiler, a steam turbine, a condenser and a heat supply network, achieves the real-time data exchange through a dynamic coupling interface, and is used for carrying out the real-time data exchange through a partitioning strong coupling method, the inter-field flux transfer accuracy is ensured; the three-stage closed-loop parameter calibration module is used for establishing a complete parameter calibration assembly line; the physical information deep learning module is used for combining a neural differential equation with physical constraints to determine a mechanism enhanced model; the third-level toughness control module is used for performing safety protection control according to the working state of the extraction condensing unit; and the full-link intelligent collaborative optimization module is used for extracting multi-source data related to the condensing unit and performing collaborative control on each module of the condensing unit. At least five innovative modules can be systematically integrated, so that intelligent optimization control of the back pressure of the extraction condensing unit is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optimal operation of a heat supply system of a thermal power plant, in particular to a high back pressure optimization control system and method for an extraction condensing unit during a heat supply period. BACKGROUND

[0002] At present, an extraction condensing unit is often used for heat supply during the operation of a thermal power plant. The selection of the back pressure of the extraction condensing unit directly affects the thermal economy of the heat supply system. The traditional back pressure control method relies on manual experience or static optimization table, and is difficult to adapt to complex and dynamic external conditions, such as load fluctuation, environmental temperature change and heat supply demand change, etc., resulting in low overall enthalpy efficiency and high coal consumption rate of the system.

[0003] Although the prior art attempts to combine simulation and machine learning, it lacks system-level closed-loop self-evolution capability and cannot realize whole life cycle optimization. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a high back pressure optimization control system and method for an extraction condensing unit during a heat supply period, which can realize intelligent optimization control of the back pressure of the extraction condensing unit through the system integration of five innovation modules.

[0005] In a first aspect, a high back pressure optimization control system for an extraction condensing unit during a heat supply period comprises:

[0006] An intelligent agent modeling implementation module is configured to establish a multi-agent model of a boiler, a steam turbine, a condenser and a heat network, realize real-time data exchange through a dynamic coupling interface, and ensure the accuracy of inter-field flux transmission by using a partition strong coupling method.

[0007] A three-stage closed-loop parameter calibration module is configured to establish a complete parameter calibration pipeline to form a closed loop from experimental design to final verification.

[0008] A physical information deep learning module is configured to combine a neural differential equation with physical constraints to determine a mechanism-enhanced model.

[0009] A three-level resilience control module is configured to perform safety protection control according to the working state of the extraction condensing unit to realize zero instability operation of the extraction condensing unit.

[0010] A full-link intelligent collaborative optimization module is configured to perform collaborative control on each module of the extraction condensing unit based on the multi-source data related to the extraction condensing unit to optimize the operating state of the extraction condensing unit.

[0011] Optionally, the intelligent agent modeling implementation module performs the following steps:

[0012] S11, obtaining a load change rate of the extraction condensing unit.

[0013] S12, determining whether the load change rate of the extraction condensing unit is greater than a pre-set load change rate threshold;

[0014] S13, if the load change rate of the extraction condensing unit is greater than the pre-set load change rate threshold, activating a fine model;

[0015] S14, performing multi-physical field coupling modeling based on the fine model to obtain a first intermediate model;

[0016] S15, performing POD reduction processing on the first intermediate model and embedding thermodynamic constraints to obtain a second intermediate model;

[0017] S16, verifying the second intermediate model, if the second intermediate model passes the verification, confirming the second intermediate model as a target model, and deploying the target model;

[0018] S17, if the second intermediate model fails to pass the verification, recalibrating the multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification.

[0019] Optionally, the method further comprises:

[0020] S21, if the load change rate of the extraction condensing unit is less than or equal to the pre-set load change rate threshold, maintaining a normal model;

[0021] S22, performing multi-physical field coupling modeling based on the normal model to obtain a third intermediate model;

[0022] S23, performing POD reduction processing on the third intermediate model and embedding thermodynamic constraints to obtain a fourth intermediate model;

[0023] S24, verifying the fourth intermediate model, if the fourth intermediate model passes the verification, confirming the fourth intermediate model as a target model, and deploying the target model;

[0024] S25, if the fourth intermediate model fails to pass the verification, recalibrating the multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification.

[0025] Optionally, the three-stage closed-loop parameter calibration module performs the following steps:

[0026] Based on a pre-designed space filling experiment, an optimal Latin hypercube sampling technique is used to ensure uniform coverage of the parameter space;

[0027] Performing data analysis on the parameter space sampling data to determine the Sobol index;

[0028] determining whether the sobol index is greater than a preset standard sobol index;

[0029] if the sobol index is greater than the preset standard sobol index, performing Bayesian optimization, and determining whether a fifth intermediate model after Bayesian optimization meets performance standards;

[0030] if the fifth intermediate model meets performance standards, updating a parameter library according to a plurality of parameters output by the fifth intermediate model.

[0031] Optionally, the step of performing data analysis on the parameter space sampling data comprises:

[0032] performing offline MOPSO optimization;

[0033] generating a Pareto parameter set;

[0034] performing online correction EnKF correction;

[0035] performing sensitivity analysis to determine a sobol index.

[0036] Optionally, the physical information deep learning module performs the following steps:

[0037] performing data preparation and determining physical feature engineering;

[0038] performing physical information model training based on a neural differential equation, a physical constraint layer design, and a predefined loss function;

[0039] verifying the physical information model according to an enhanced adversarial sample and an uncertainty quantification method to obtain a target physical information model;

[0040] performing performance evaluation on the target physical information model to determine whether the performance of the target physical information model meets standards;

[0041] if the performance of the target physical information model meets standards, saving and deploying the target physical information model.

[0042] Optionally, the method further comprises:

[0043] if the performance of the target physical information model does not meet standards, performing hyperparameter adjustment and retraining the physical information model.

[0044] Optionally, the three-level resilience control architecture performs the following steps:

[0045] performing state monitoring and extracting manifold features;

[0046] Nonlinear control law design and Lyapunov stability analysis are performed to generate control actions;

[0047] Safety constraint checking is performed to determine whether the constraint condition is met;

[0048] If the constraint condition is met, the control actions are executed;

[0049] Performance feedback is obtained, parameter adaptive adjustment is performed, and state monitoring is continued;

[0050] If the constraint condition is not met, safety compensation and system protection are performed;

[0051] Performance feedback is obtained, parameter adaptive adjustment is performed, and state monitoring is continued.

[0052] Optionally, the full-link intelligent collaborative optimization module performs the following steps:

[0053] Multi-source data is collected and preprocessed;

[0054] Intelligent agent modeling is performed according to the preprocessed multi-source data;

[0055] A target physical AI model is obtained through parameter calibration and multi-modal fusion optimization calculation;

[0056] Performance monitoring is performed to determine whether the performance of the target physical AI model meets the standard;

[0057] If the performance of the target physical AI model meets the standard, the target physical AI model is periodically evaluated to determine whether the target physical AI model needs to be updated;

[0058] If the target physical AI model needs to be updated, intelligent agent modeling is performed again.

[0059] On the other hand, the application also provides a high back pressure optimization control method for a heat extraction condensing unit during the heating period, comprising:

[0060] The control intelligent agent modeling implementation module establishes multi-agent models of the boiler, steam turbine, condenser, and heat network, realizes real-time data exchange through dynamic coupling interfaces, and adopts a partition strong coupling method to ensure the accuracy of inter-field flux transmission;

[0061] The control three-stage closed-loop parameter calibration module establishes a complete parameter calibration pipeline to form a closed loop from experimental design to final verification;

[0062] The control physical information deep learning module combines neural differential equations with physical constraints to determine a mechanism-enhanced model;

[0063] The three-stage toughness control module controls the safe protection according to the working state of the extraction condensing unit, so as to realize zero instability operation of the extraction condensing unit.

[0064] The full-link intelligent collaborative optimization module extracts the multi-source data related to the extraction condensing unit, and collaboratively controls each module of the extraction condensing unit, so as to optimize the operation state of the extraction condensing unit.

[0065] The embodiment of the application provides a high back pressure optimization control system and method for an extraction condensing unit in a heating period.

[0066] In order to make the above-mentioned purposes, characteristics and advantages of the application more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0068] Figure 1 A schematic diagram of a high back pressure optimization control system for an extraction condensing unit in a heating period provided by the embodiment of the application;

[0069] Figure 2 A flowchart of an intelligent agent modeling implementation module provided by the embodiment of the application;

[0070] Figure 3 A flowchart of a three-stage closed-loop parameter calibration module provided by the embodiment of the application;

[0071] Figure 4 A flowchart of a physical information deep learning module provided by the embodiment of the application;

[0072] Figure 5 A flowchart of a three-stage toughness control module provided by the embodiment of the application;

[0073] Figure 6 A flowchart of a full-link intelligent collaborative optimization module provided by the embodiment of the application;

[0074] Figure 7 A flowchart of a high back pressure optimization control method for an extraction condensing unit in a heating period provided by the embodiment of the application. DETAILED DESCRIPTION

[0075] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application and not all embodiments of the present application. The components of the embodiments of the present application generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by a person skilled in the art without creative work belongs to the scope of protection of the present application.

[0076] Firstly, the application scenarios applicable to the present application are introduced. The present application can be applied to the technical field of heat supply system optimization operation of thermal power plants.

[0077] At present, the extraction condensing unit is often used for heat supply in the operation of thermal power plants. The selection of the back pressure of the extraction condensing unit directly affects the thermal economy of the heat supply system. The traditional back pressure control mode depends on artificial experience or static optimization table, and it is difficult to adapt to complex and dynamically changing external working conditions, such as load fluctuation, environmental temperature change and heat supply demand change, etc., resulting in low overall enthalpy efficiency and high coal consumption rate of the system.

[0078] The traditional back pressure control of the extraction condensing unit faces four technical bottlenecks:

[0079] 1. Insufficient model rigidity: static model is difficult to adapt to severe load fluctuations (more than 5% change rate per minute)

[0080] 2. Single optimization dimension: ignoring multi-objective coordination such as equipment life and risk constraints

[0081] 3. Mechanism data fragmentation: pure AI model lacks physical interpretability and has poor generalization ability

[0082] 4. Difficult cross-unit migration: specific optimization scheme is difficult to scale up and apply

[0083] Although the existing technology attempts to combine simulation and machine learning, it lacks system-level closed-loop self-evolution ability and cannot realize whole life cycle optimization.

[0084] In the first aspect, please refer to Figure 1 , Figure 1 A schematic diagram of a high back pressure optimization control system for an extraction condensing unit in a heat supply period provided by the embodiments of the present application. As Figure 1As shown in the middle, the heat supply period extraction condensing unit high back pressure optimization control system provided by the embodiment of the application comprises: an intelligent agent modeling implementation module 101, a three-stage closed-loop parameter calibration module 102, a physical information deep learning module 103, a three-level toughness control module 104, and a full-link intelligent collaborative optimization module 105.

[0085] The intelligent agent modeling implementation module is configured to establish multi-agent models of a boiler, a steam turbine, a condenser, and a heat network, realize real-time data exchange through a dynamic coupling interface, and ensure the accuracy of inter-field flux transmission by using a partition strong coupling method.

[0086] The three-stage closed-loop parameter calibration module is configured to establish a complete parameter calibration pipeline to form a closed loop from experimental design to final verification.

[0087] The physical information deep learning module is configured to combine a neural differential equation with physical constraints to determine a mechanism-enhanced model.

[0088] The three-level toughness control module is configured to perform safety protection control according to the working state of the extraction condensing unit to realize zero instability operation of the extraction condensing unit.

[0089] The full-link intelligent collaborative optimization module is configured to perform collaborative control on each module of the extraction condensing unit based on multi-source data related to the extraction condensing unit to optimize the operating state of the extraction condensing unit.

[0090] Specifically, the intelligent agent modeling implementation module comprises:

[0091] An adaptive mesh refinement algorithm: an hp adaptive finite element method based on residual estimation;

[0092] A multiscale coupling algorithm: a heterogeneous multiscale method is used for cross-scale information transmission;

[0093] Nonlinear model reduction: a local linear embedding algorithm based on manifold learning;

[0094] Real-time solver optimization: a preconditioned conjugate gradient method and algebraic multigrid acceleration;

[0095] In this way, the calculation efficiency is improved by more than 40%, the model compression rate is 90%, and the physical consistency is still maintained, solving the problem of response lag of traditional static models, and the contradiction between calculation efficiency and accuracy.

[0096] Wherein, please refer to Figure 2 The intelligent agent modeling implementation module performs the following steps:

[0097] S11, obtaining the load change rate of the extraction condensing unit;

[0098] S12, judging whether the load change rate of the extraction condensing unit is greater than a pre-set load change rate threshold value;

[0099] S13, if the load change rate of the extraction condensing unit is greater than the pre-set load change rate threshold value, activating a fine model;

[0100] S14, performing multi-physical field coupling modeling based on the fine model to obtain a first intermediate model;

[0101] S15, performing POD reduction processing on the first intermediate model and embedding thermodynamic constraints to obtain a second intermediate model;

[0102] S16, verifying the second intermediate model, if the second intermediate model passes the verification, confirming the second intermediate model as a target model and deploying the target model;

[0103] S17, if the second intermediate model fails to pass the verification, recalibrating multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification.

[0104] Please continue to refer to Figure 2 , the method further comprises:

[0105] S21, if the load change rate of the extraction condensing unit is less than or equal to the pre-set load change rate threshold value, maintaining a normal model;

[0106] S22, performing multi-physical field coupling modeling based on the normal model to obtain a third intermediate model;

[0107] S23, performing POD reduction processing on the third intermediate model and embedding thermodynamic constraints to obtain a fourth intermediate model;

[0108] S24, verifying the fourth intermediate model, if the fourth intermediate model passes the verification, confirming the fourth intermediate model as a target model and deploying the target model;

[0109] S25, if the fourth intermediate model fails to pass the verification, recalibrating multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification.

[0110] The intelligent agent modeling implementation module establishes multi-agent models of the boiler, the steam turbine, the condenser and the heat network, and realizes real-time data exchange through dynamic coupling interfaces. A partition strong coupling method is adopted to ensure the accuracy of inter-field flux transmission, and a Newton-Krylov method is used to solve the nonlinear system.

[0111] Specifically, the Delaunay triangulation is used to generate a calculation grid: advanced mesh generation techniques are used to ensure calculation accuracy and efficiency; control equations for each subsystem are established: accurate mathematical models are constructed based on the first and second laws of thermodynamics; implicit time integration methods are used: numerical stability and calculation accuracy are ensured; parallel computing techniques are used: multi-core processors and GPUs are used to accelerate large-scale calculations.

[0112] The three-stage closed-loop parameter calibration module includes:

[0113] Multi-objective optimization algorithm: NSGA-III algorithm based on reference point to process high-dimensional target space;

[0114] Uncertainty quantification: global sensitivity analysis based on polynomial chaos expansion;

[0115] Adaptive sampling strategy: acquisition function balancing expected improvement and confidence upper bound;

[0116] Distributed optimization framework: asynchronous parallel Bayesian optimization to accelerate convergence;

[0117] In this way, the cross-unit calibration speed is improved by 50%, and the long-term prediction error is ≤1.5%, solving the problem of parameter drift caused by equipment aging and offline calibration failure in the prior art.

[0118] Specifically, please refer to Figure 3 The three-stage closed-loop parameter calibration module performs the following steps:

[0119] Based on pre-designed space-filling experiments, the optimal Latin hypercube sampling technique is used to ensure uniform coverage of the parameter space;

[0120] Perform data analysis on the parameter space sampling data to determine the Sobol index;

[0121] Determine whether the Sobol index is greater than the pre-set standard Sobol index;

[0122] If the Sobol index is greater than the pre-set standard Sobol index, perform Bayesian optimization, and determine whether the performance of the fifth intermediate model after Bayesian optimization meets the performance standard;

[0123] If the performance of the fifth intermediate model meets the performance standard, update the parameter library according to the multiple parameters output by the fifth intermediate model.

[0124] Please continue to refer to Figure 3 The step of performing data analysis on the parameter space sampling data includes:

[0125] Perform offline MOPSO optimization;

[0126] Generating a Pareto parameter set;

[0127] Online correction EnKF correction;

[0128] Sensitivity analysis is performed to determine the Sobol index.

[0129] Among them, the physical information deep learning module comprises:

[0130] Neural differential equation solving: implicit differential equation solver with sensitivity analysis

[0131] Physical constraint embedding: hard constraint optimization method based on Lagrange multiplier

[0132] Multi-fidelity learning: multi-fidelity neural network combining high-fidelity simulation and low-fidelity data

[0133] Uncertainty propagation: Bayesian deep learning based on random weight averaging

[0134] In this way, the error of abnormal working conditions is reduced by 35%, the prediction confidence reaches 95%, and the problems of lack of physical interpretability of black box model and failure of extreme working conditions are solved.

[0135] The three-stage closed-loop parameter calibration module establishes a complete parameter calibration pipeline, forming a closed loop from experimental design to final verification. Optimal Latin hypercube sampling is used to ensure sufficient exploration of the parameter space.

[0136] Specifically, the design space filling experiment: adopt optimal Latin hypercube sampling technology to ensure uniform coverage of the parameter space; multi-objective evolutionary algorithm: apply NSGA-III algorithm to generate high-quality Pareto optimal solution set; global sensitivity analysis based on variance decomposition: accurately identify key parameters and optimize resource allocation; apply Bayesian optimization to achieve intelligent exploration: use TuRBO algorithm to solve high-dimensional optimization problems.

[0137] Specifically, please refer to Figure 4 The physical information deep learning module performs the following steps:

[0138] Data preparation and determination of physical feature engineering are performed;

[0139] Based on neural differential equations, physical constraint layers, and predefined loss functions, physical information model training is performed;

[0140] The physical information model is verified according to the enhanced adversarial samples and the uncertainty quantification method, and the target physical information model is obtained;

[0141] The performance of the target physical information model is evaluated to determine whether the performance of the target physical information model meets the requirements;

[0142] If the performance of the target physical information model meets the standard, the target physical information model is saved and deployed.

[0143] Please continue to refer to Figure 4 , the method further comprises:

[0144] If the performance of the target physical information model does not meet the standard, hyperparameter adjustment is performed and the physical information model is retrained.

[0145] Among them, the three-level resilience control module includes:

[0146] Predictive control algorithm: random model predictive control based on Gaussian process

[0147] Stability analysis algorithm: Lyapunov function construction based on harmony search

[0148] Manifold learning algorithm: hybrid method based on diffusion mapping and local linear embedding

[0149] Safety control algorithm: real-time safety guarantee based on control barrier function

[0150] In this way, the response to sudden change conditions is less than 300ms, realizing zero instability operation and avoiding oscillation caused by load mutation and extreme condition instability.

[0151] Specifically, please refer to Figure 5 , the three-level resilience control architecture performs the following steps:

[0152] State monitoring is performed and manifold features are extracted;

[0153] Nonlinear control law design and Lyapunov stability analysis are performed to generate control actions;

[0154] Safety constraint checking is performed to determine whether the constraint condition is met;

[0155] If the constraint condition is met, the control action is executed;

[0156] Performance feedback is obtained, parameter adaptive adjustment is performed, and state monitoring is continued;

[0157] If the constraint condition is not met, safety compensation and system protection are performed;

[0158] Performance feedback is obtained, parameter adaptive adjustment is performed, and state monitoring is continued.

[0159] Among them, the full-link intelligent collaborative optimization module includes:

[0160] Multi-modal feature alignment based on attention mechanism;

[0161] A distributed alternating direction method of multipliers optimization framework

[0162] A fast solution method for online convex optimization

[0163] Fast adaptation strategy for meta-learning.

[0164] In this way, after the whole system is integrated, the simulation speed under variable working conditions is improved by 5 times, the operation cost is reduced by 10.5%, the equipment life is improved by 20%, the operation and maintenance cost is reduced by 18%, the problem of isolated operation of each module is avoided, and the overall optimization effect is limited.

[0165] Specifically, please refer to Figure 6 The full-link intelligent collaborative optimization module performs the following steps:

[0166] Collecting multi-source data and performing multi-source data preprocessing;

[0167] Modeling the agent according to the preprocessed multi-source data;

[0168] Obtaining the target physical AI model through parameter calibration and multi-modal fusion optimization calculation;

[0169] Performing performance monitoring to determine whether the performance of the target physical AI model meets the standard;

[0170] If the performance of the target physical AI model meets the standard, periodically evaluate the target physical AI model to determine whether the target physical AI model needs to be updated;

[0171] If the target physical AI model needs to be updated, re-model the agent.

[0172] In this way, by establishing a complete intelligent optimization control pipeline, full-link collaboration from data collection to control execution is realized. Through performance monitoring and parameter self-adaptation, continuous optimization of the system is realized.

[0173] Among them, the physical AI model is constructed by using a deep learning architecture guided by design mechanism, which embeds physical knowledge into network structure and loss function. Neural ordinary differential equations are used to describe system dynamics.

[0174] Specifically, it includes: constructing physical guidance features based on mechanism analysis: extracting features with physical meaning such as enthalpy difference and exergy efficiency; designing neural differential equation layers to capture dynamic characteristics: using neural ordinary differential equations to accurately describe system evolution; embedding hard constraints using the Lagrange multiplier method: strictly ensuring that the model meets the physical law; applying the Monte Carlo method to quantify uncertainty: providing confidence intervals for prediction results.

[0175] The application is also based on the manifold learning theory, extracts the essential characteristics of the system state, and designs a nonlinear control law. The global stability of the control system is ensured through Lyapunov stability analysis.

[0176] The control strategy implementation steps include: state analysis: real-time extraction of system operating state characteristics, identification of key patterns; control design: design of a nonlinear control law based on manifold learning to improve control accuracy; stability guarantee: ensure the stability of the system under various operating conditions through Lyapunov functions; safety check: real-time check whether the control action meets the safety constraint conditions.

[0177] In a second aspect, referring to Figure 7 The application also provides a high back pressure optimization control method for a heat supply period extraction condensing unit. As shown in Figure 7 The method comprises the following steps:

[0178] S101, the control intelligent agent modeling implementation module establishes a multi-agent model of the boiler, steam turbine, condenser and heat network, realizes real-time data exchange through a dynamic coupling interface, and adopts a partition strong coupling method to ensure the accuracy of inter-field flux transmission;

[0179] S102, the control three-stage closed-loop parameter calibration module establishes a complete parameter calibration pipeline to form a closed loop from experimental design to final verification;

[0180] S103, the control physical information deep learning module combines neural differential equations with physical constraints to determine a mechanism-enhanced model;

[0181] S104, the control three-level resilience control module performs safety protection control according to the working state of the extraction condensing unit to realize zero instability operation of the extraction condensing unit;

[0182] S105, the control full-link intelligent collaborative optimization module extracts the multi-source data related to the extraction condensing unit, and performs collaborative control on each module of the extraction condensing unit to optimize the operating state of the extraction condensing unit.

[0183] The application provides a high-back pressure optimization control system and method for a heat supply period extraction condensing unit. The system is integrated through five innovative modules, and the smooth exchange between the modules is ensured through a unified data interface. The system has the following advantages: a standardized data format and communication protocol are formulated; the standardized communication protocol supports distributed deployment; the loose coupling and scalability of each module are realized; a visual monitoring interface provides real-time state display; a friendly user interface is developed to facilitate operation and monitoring; an automatic operation and maintenance function reduces manual intervention; the system has self-diagnosis and self-repair capabilities; the modeling speed of dynamic working conditions is improved by 5 times, the coal consumption prediction error is less than or equal to 1.5%; the extreme working condition response is less than 300 ms, and the number of control instability is reduced to zero; the economic benefits are as follows: the service life of equipment is improved by 20%, and the comprehensive coal consumption is reduced by 10.5%; the environmental benefits are as follows: carbon emissions are reduced by 12%, and operation and maintenance costs are reduced by 18%.

[0184] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not contradict, they should be considered as the scope of the application.

[0185] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not contradict, they should be considered as the scope of the application.

[0186] The above embodiments only express several implementation manners of the application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent of the application. It should be pointed out that for ordinary skilled persons in the art, some modifications and improvements can be made without departing from the concept of the application, and these all belong to the protection scope of the application. Therefore, the protection scope of the application should be subject to the appended claims.

Claims

1. A high back pressure optimization control system for a heat supply period extraction condensing unit, characterized in that, The method comprises the following steps: S11, obtaining the load change rate of the extraction condensing unit; S12, determining whether the load change rate of the extraction condensing unit is greater than a pre-set load change rate threshold; S13, if the load change rate of the extraction condensing unit is greater than the pre-set load change rate threshold, activating a fine model; S14, performing multi-physical field coupling modeling based on the fine model to obtain a first intermediate model; S15, performing POD reduction processing on the first intermediate model and embedding thermodynamic constraints to obtain a second intermediate model; 2. The system of claim 1, wherein, S16, verifying the second intermediate model, if the second intermediate model passes the verification, confirming the second intermediate model as a target model, and deploying the target model; S17, if the second intermediate model fails to pass the verification, re-calibrating the multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification. The method further comprises the following steps: S21, if the load change rate of the extraction condensing unit is less than or equal to the pre-set load change rate threshold, keeping a normal model; S22, performing multi-physical field coupling modeling based on the normal model to obtain a third intermediate model; S23, performing POD reduction processing on the third intermediate model and embedding thermodynamic constraints to obtain a fourth intermediate model; S24, verifying the fourth intermediate model, if the fourth intermediate model passes the verification, confirming the fourth intermediate model as a target model, and deploying the target model; S25, if the fourth intermediate model fails to pass the verification, re-calibrating the multi-physical field coupling modeling parameters and re-executing steps S14-S15 until the second intermediate model passes the verification.

3. The system of claim 2, wherein, The three-stage closed-loop parameter calibration module performs the following steps: Based on pre-designed space filling experiments, using optimal Latin hypercube sampling technology to ensure uniform coverage of the parameter space; Performing data analysis on the parameter space sampling data to determine the sobol index; Determining whether the sobol index is greater than a pre-set standard sobol index; ​ ​ 4. The system of claim 1, wherein, ​ ​ ​ ​ If the sobol index is greater than a preset standard sobol index, Bayesian optimization is performed, and it is judged whether the performance of the fifth intermediate model after Bayesian optimization meets the standard; If the performance of the fifth intermediate model meets the standard, the parameter library is updated according to the plurality of parameters output by the fifth intermediate model.

5. The system of claim 4, wherein, The step of performing data analysis on the parameter space sampling data comprises: Performing offline MOPSO optimization; Generating a Pareto parameter set; Performing online correction EnKF correction; Performing sensitivity analysis to determine the sobol index.

6. The system of claim 1, wherein, The physical information deep learning module performs the following steps: Performing data preparation and determining physical feature engineering; Performing physical information model training based on neural differential equations, physical constraint layers, and predefined loss functions; Verifying the physical information model according to the enhanced adversarial samples and the uncertainty quantification method to obtain a target physical information model; Performing performance evaluation on the target physical information model to determine whether the performance of the target physical information model meets the standard; If the performance of the target physical information model meets the standard, the target physical information model is saved and deployed.

7. The system of claim 6, wherein, The method further comprises: If the performance of the target physical information model does not meet the standard, hyperparameter adjustment is performed, and the physical information model is retrained.

8. The system of claim 1, wherein, The three-level resilience control architecture performs the following steps: Performing state monitoring and extracting manifold features; Performing nonlinear control law design and Lyapunov stability analysis to generate control actions; Performing safety constraint checking to determine whether the constraint condition is met; If the constraint condition is met, the control action is executed; Obtaining performance feedback, performing parameter adaptive adjustment, and continuing state monitoring; If the constraint condition is not met, safety compensation and system protection are performed; Obtaining performance feedback, performing parameter adaptive adjustment, and continuing state monitoring.

9. The system of claim 8, wherein, The full-link intelligent collaborative optimization module performs the following steps: Collecting multi-source data and performing multi-source data preprocessing; Performing agent modeling according to the preprocessed multi-source data; Obtaining a target physical AI model through parameter calibration and multi-modal fusion optimization calculation; Performing performance monitoring to determine whether the performance of the target physical AI model meets the standard; If the performance of the target physical AI model meets the standard, the target physical AI model is periodically evaluated to determine whether the target physical AI model needs to be updated; If the target physical AI model needs to be updated, agent modeling is performed again.

10. A method for high back pressure optimization control of a heat supply period extraction condensing unit, characterized in that, It comprises: The control agent modeling implementation module establishes multi-agent models of the boiler, steam turbine, condenser, and heat network, realizes real-time data exchange through dynamic coupling interfaces, and adopts a partition strong coupling method to ensure the accuracy of inter-field flux transmission; The control three-stage closed-loop parameter calibration module establishes a complete parameter calibration pipeline to form a closed loop from experimental design to final verification; The control physical information deep learning module combines neural differential equations with physical constraints to determine a mechanism-enhanced model; The control three-level resilience control module performs safety protection control according to the working state of the extraction-condensing unit to realize zero instability operation of the extraction-condensing unit; The control full-link intelligent collaborative optimization module extracts and condenses the related multi-source data of the extraction-condensing unit, and collaboratively controls each module of the extraction-condensing unit, so as to optimize the operation state of the extraction-condensing unit.

Images