Dry-wet state conversion self-adaptive adjusting system for deep peak regulation of thermal power generating unit

Abstract

The invention relates to the technical field of automatic control and deep peak shaving operation of thermal power generating units, and discloses a dry-wet state conversion self-adaptive adjusting system for deep peak shaving of a thermal power generating unit. Comprising a classified data acquisition module, a data preprocessing module, a hybrid dynamic modeling module, a deep learning adaptive decision module, a multi-model prediction optimization module, a full-automatic execution module, an integrated linkage module, a safety monitoring module and an operation and maintenance management module. By designing a cooperative working mechanism of full-automatic execution, integrated linkage, real-time safety monitoring and intelligent operation and maintenance management, a full-process closed-loop intelligent management and control system from instruction generation to precise execution, from a main system to an auxiliary system, and from process control to risk management and control and efficiency evaluation is constructed. And finally, the beneficial effects of greatly improving the automation degree, the operation safety and the comprehensive economy of the system and reducing the operation burden and the safety risk of operators are achieved.

Description

Technical Field

[0001] This invention relates to the field of automatic control and deep peak shaving operation technology for thermal power units, specifically to a dry-wet state conversion adaptive adjustment system for deep peak shaving of thermal power units. Background Technology

[0002] With the continuous increase in the proportion of renewable energy power generation, the power grid's requirements for the deep peak-shaving capability of thermal power units are becoming increasingly stringent. Once-through boilers and supercritical units with deep peak-shaving capability need to frequently switch between dry and wet operation modes under low load to ensure hydrodynamic safety and boiler efficiency.

[0003] However, existing dry-wet state conversion processes mainly rely on the experience of operators for manual or semi-automatic operation, which has the following prominent problems: First, the conversion process is slow to respond and cannot quickly track the grid peak-shaving instructions; second, there is a lack of accurate models that can accurately describe the dynamic characteristics of dry / wet states and their conversion processes under wide loads and multiple operating conditions, resulting in control strategies relying on empirical parameters and poor adaptive capabilities; third, the main control system and auxiliary control systems such as circulating water and soot blowing are not well linked, resulting in "information silos" and low coordination efficiency; fourth, the degree of automation is low, the operation steps are cumbersome, the safety risks are high, and the key parameters (such as main steam temperature, main steam pressure, and working fluid level) fluctuate greatly during the conversion process, affecting the unit's lifespan and economy.

[0004] Therefore, there is an urgent need for an intelligent control system for dry and wet state conversion that can achieve full automation, self-adaptation, high precision, safety and economy. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a dry-wet state conversion adaptive regulation system for deep peak shaving of thermal power units. It has the advantages of fast response, accurate modeling, intelligent decision-making, fully automatic execution, deep linkage between main and auxiliary systems, and full-process safety monitoring. It solves the problems of slow response, low modeling accuracy, poor linkage, low degree of automation, and poor safety and economy in existing technologies.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution: a dry-wet state conversion adaptive regulation system for deep peak shaving of thermal power units, comprising a classification data acquisition module, a data preprocessing module, a hybrid dynamic modeling module, a deep learning adaptive decision-making module, a multi-model prediction optimization module, a fully automatic execution module, an integrated linkage module, a safety monitoring module, and an operation and maintenance management module;

[0009] The classification-based data acquisition module collects raw data in all dimensions at four classification sites, and the collected data is transmitted to the data preprocessing module in real time.

[0010] The data preprocessing module preprocesses the raw data collected from the four sites across all dimensions, generating a high-quality dataset that is then transmitted to the hybrid dynamic modeling module and the deep learning adaptive decision-making module.

[0011] The hybrid dynamic modeling module constructs a dynamic model of dry-wet state transition and outputs accurate modeling guidance by fusing dry-wet state characteristics into a modeling formula.

[0012] The deep learning adaptive decision-making module, based on the preprocessed high-quality dataset and the output of the hybrid dynamic model, uses deep learning algorithms to make adaptive decisions on the timing of dry-wet state transitions and the adjustment parameters.

[0013] The multi-model prediction optimization module, based on the output of the hybrid dynamic model and the results of deep learning adaptive decision-making, introduces a multi-model prediction control algorithm to predict key parameters within the next 5-10 seconds, thereby performing secondary optimization on the adjustment parameter instructions output by the deep learning adaptive decision-making module, generating the optimal control sequence and transmitting it to the fully automatic execution module.

[0014] The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module, and drives the corresponding actuator to complete precise actions based on the fully automatic sequential control logic, thereby realizing the fully automatic execution of the dry-wet state conversion process.

[0015] The integrated linkage module is based on a DCS deep integration architecture, which fully integrates the auxiliary control system and the main control system to build a unified linkage control logic. At the same time, it receives the action instructions of the fully automatic execution module and synchronously coordinates the operating status of each auxiliary control system.

[0016] The safety monitoring module collects unit operating parameters and operating status data of each module in real time. By setting safety thresholds and trend warning curves, it monitors the risks of over-temperature, over-pressure and hydrodynamic instability during the transition process in real time and outputs warning information.

[0017] The operation and maintenance management module receives early warning information from the safety monitoring module and operating data from each module, constructs a unit operation status assessment model, performs real-time calculation and trend analysis of equipment loss and operating efficiency, and realizes immediate equipment maintenance.

[0018] Preferably, the categorized data acquisition module includes four stations: a boiler-side data acquisition unit, a turbine-side data acquisition unit, an auxiliary equipment-side data acquisition unit, and an electrical-side data acquisition unit.

[0019] Preferably, the boiler-side data acquisition unit acquires key parameters of the boiler combustion and steam-water system through distributed temperature / pressure transmitters and flow meters.

[0020] Preferably, the turbine-side data acquisition unit acquires the operating parameters of the turbine and its regulation system through vibration sensors, displacement sensors, and thermal instruments.

[0021] Preferably, the auxiliary equipment side data acquisition unit collects the status and process parameters of the main auxiliary equipment through the pump / fan intelligent monitoring device and valve positioner.

[0022] Preferably, the electrical side data acquisition unit acquires parameters of the generator and plant power system through electrical measurement and protection devices.

[0023] Preferably, the hybrid dynamic modeling module achieves accurate modeling guidance under wide loads and multiple operating conditions by fusing dry and wet characteristics into a modeling formula. The calculation formula is as follows: , In the formula, This represents the comprehensive dynamic characteristic function during the dry-wet state transition process. Represents the input parameter vector. Indicates time, This represents the dry / wet weighting coefficient. This represents the dynamic characteristic function of dry-state operation. This represents the dynamic characteristic function under wet conditions. This represents the dynamic correction term during the transition process; , In the formula, Represents the rate of change of the system state vector. Represents the system state vector. Represents the control input vector. This represents a mechanistic model based on physical laws such as the conservation of mass and energy. This represents the i-th fuzzy rule or data-driven sub-model. This indicates that it uses adaptive weights based on real-time evaluation using reinforcement learning. This represents the total number of fuzzy rules or sub-models, with a value ranging from 4 to 10.

[0024] Preferably, the multi-model prediction optimization module uses a state prediction model to construct a multi-objective prediction optimization model for the transition process, specifically expressed as follows: , In the formula, This represents the predicted output vector of key parameters at a future time (k+1). This represents the system state vector at the current time (k). This represents the current control input sequence to be optimized. and These represent the state matrix and control matrix obtained by linearizing the hybrid dynamic model, respectively.

[0025] Preferably, the multi-model prediction optimization module uses a multi-objective optimization function to construct a multi-objective prediction optimization model for the transition process, expressed as: In the formula, This represents the value of the objective function that needs to be minimized. This means minimizing the sum of costs for all prediction steps within the prediction time domain. This represents the vector of actual predicted values ​​of the key controlled parameters output by the hybrid dynamic model. This indicates the dynamic setpoint or safety range reference trajectory of the key controlled parameter during the transition process. This represents the rate of change or increment of the control input vector. The penalty or time cost term represents the degree of completion of the transition process. , , These represent the weighting coefficients used to balance the tracking accuracy of the balancing parameters, the smoothness of the control action, and the transition speed.

[0026] Preferably, the multi-model prediction and optimization module, based on the state prediction model and the calculation results of the multi-objective optimization function, predicts key parameters such as main steam temperature, main steam pressure, and working fluid level within the next 5-10 seconds, and performs secondary optimization on the adjustment parameter instructions output by the deep learning adaptive decision module to generate the optimal control sequence and transmit it to the fully automatic execution module; at the same time, it achieves smooth switching between dry state, wet state, and different models during the transition process through model switching logic;

[0027] The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module. Based on the fully automatic sequential control logic, it drives the corresponding water regulating valve, steam valve and circulating water pump to complete precise actions, realizing the fully automatic execution of the dry-wet state conversion process. This module replaces manual operation with standardized execution logic; at the same time, it collects the action feedback data of the actuator in real time to form a closed-loop control.

[0028] Compared with the prior art, the present invention provides a dry-wet state switching adaptive regulation system for deep peak shaving of thermal power units, which has the following beneficial effects:

[0029] 1. This invention establishes a high-precision dynamic model for dry-wet state conversion under wide load and multiple operating conditions by constructing a classification-based full-dimensional data acquisition network and a hybrid dynamic modeling method combining fusion mechanism, fuzzy rules and reinforcement learning. This achieves the beneficial effect of fundamentally solving the problems of insufficient accuracy and poor adaptability of traditional models, and providing a reliable model foundation for intelligent decision-making and control.

[0030] 2. This invention integrates deep learning adaptive decision-making and multi-model predictive optimization control to form a closed-loop control architecture of intelligent decision-making, rolling optimization, and feedback correction. This architecture enables autonomous optimization decisions on transition timing and adjustment parameters, and achieves multi-objective optimization control with the fastest transition process and the smallest parameter fluctuations while ensuring safety. This significantly improves the system's response speed and stability.

[0031] 3. This invention designs a collaborative working mechanism that integrates fully automated execution, coordinated linkage, real-time safety monitoring, and intelligent operation and maintenance management. It constructs a closed-loop intelligent management and control system covering the entire process from instruction generation to precise execution, from the main system to the auxiliary system, and from process control to risk management and performance evaluation. Ultimately, this invention achieves the beneficial effects of significantly improving the system's automation level, operational safety, and overall economy, while reducing the operational burden and safety risks for operators. Attached Figure Description

[0032] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figure 1 A dry-wet state conversion adaptive regulation system for deep peak shaving of thermal power units includes a classification data acquisition module, a data preprocessing module, a hybrid dynamic modeling module, a deep learning adaptive decision-making module, a multi-model prediction optimization module, a fully automatic execution module, an integrated linkage module, a safety monitoring module, and an operation and maintenance management module.

[0035] The classification-based data acquisition module collects raw data from four classification sites across all dimensions, providing raw data support for subsequent modeling, analysis and control. The collected data is transmitted to the data preprocessing module in real time.

[0036] The data preprocessing module cleans, denoises, normalizes, and aligns the raw data collected from the four sites across all dimensions, removes outlier data, standardizes the data format, and generates a high-quality dataset that is then transmitted to the hybrid dynamic modeling module and the deep learning adaptive decision-making module.

[0037] The hybrid dynamic modeling module is based on the integration of mechanism analysis, fuzzy modeling, and reinforcement learning to construct a high-precision dynamic model of dry-wet state transition. It achieves accurate modeling guidance under wide load and multiple working conditions through the fusion modeling formula of dry and wet state characteristics, and finally solves the problems of existing technologies lacking accurate dry / wet dynamic models and poor adaptability.

[0038] The deep learning adaptive decision-making module, based on the preprocessed high-quality dataset and the output of the hybrid dynamic model, uses deep learning algorithms to make adaptive decisions on the timing of dry-wet transitions and adjustment parameters, thus solving the problems of slow response and reliance on manual intervention in conventional control.

[0039] The hybrid dynamic modeling module constructs a dynamic model of dry-wet state transition and outputs accurate modeling guidance by fusing dry-wet state characteristics into modeling formulas.

[0040] The deep learning adaptive decision-making module uses a deep learning algorithm to make adaptive decisions on the timing of dry-wet transitions and the adjustment parameters based on the preprocessed high-quality dataset and the output of the hybrid dynamic model.

[0041] The multi-model prediction optimization module is based on the output of the hybrid dynamic model and the results of deep learning adaptive decision-making. It introduces a multi-model predictive control algorithm to predict key parameters in the next 5-10 seconds, realizes secondary optimization of the adjustment parameter instructions output by the deep learning adaptive decision-making module, generates the optimal control sequence and transmits it to the fully automatic execution module.

[0042] The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module, and drives the corresponding actuator to complete precise actions based on the fully automatic sequential control logic, thereby realizing the fully automatic execution of the dry-wet state conversion process.

[0043] The integrated linkage module is based on the DCS deep integration architecture, which fully integrates the auxiliary control system with the main control system to build a unified linkage control logic. At the same time, it receives the action instructions of the fully automatic execution module and synchronously coordinates the operating status of each auxiliary control system.

[0044] The safety monitoring module collects unit operating parameters and operating status data of each module in real time. By setting safety thresholds and trend warning curves, it monitors the risks of over-temperature, over-pressure and hydrodynamic instability during the transition process in real time and outputs warning information.

[0045] The operation and maintenance management module receives early warning information from the safety monitoring module and operational data from each module, constructs a unit operation status assessment model, performs real-time calculation and trend analysis of equipment wear and operating efficiency, and enables immediate equipment maintenance.

[0046] The advantages are: the above modules are executed in a process of classification data acquisition module → data preprocessing module → hybrid dynamic modeling module / deep learning adaptive decision-making module → multi-model prediction optimization module → fully automatic execution module → integrated linkage module. The safety monitoring module monitors the risks of each link of the core process in real time. The operation and maintenance management module conducts assessment and maintenance based on the full process data, forming a closed-loop control of the entire process of data acquisition-processing-modeling-decision-optimization-execution-linkage-monitoring-operation and maintenance. Ultimately, it realizes the adaptive adjustment of the dry and wet state conversion of thermal power units, and solves the problems of slow response, low modeling accuracy, poor linkage, low degree of automation and poor safety and economy in the existing technology.

[0047] The categorized data acquisition module includes four stations: boiler-side data acquisition unit, turbine-side data acquisition unit, auxiliary equipment-side data acquisition unit, and electrical-side data acquisition unit.

[0048] The boiler-side data acquisition unit collects key parameters of the boiler combustion and steam-water system through distributed temperature / pressure transmitters and flow meters, including main steam temperature, main steam pressure, reheat steam temperature, feedwater flow rate, steam flow rate, drum / separator water level, wall temperature, and flue gas oxygen content. The turbine-side data acquisition unit collects operating parameters of the turbine and its regulating system through vibration sensors, displacement sensors, and thermal instruments, including turbine speed, shaft vibration, bearing temperature, regulating stage pressure, extraction steam parameters, and DEH commands.

[0049] The auxiliary equipment side data acquisition unit collects the status and process parameters of the main auxiliary equipment through the pump / fan intelligent monitoring device and valve positioner, including the current, speed, and outlet pressure of the feed water pump / circulating water pump, the opening degree and valve position feedback of the regulating valve / stop valve, and the condensate flow rate and pressure; the electrical side data acquisition unit collects the parameters of the generator and plant power system through electrical measurement and protection devices, including the generator active / reactive power, stator / rotor current and voltage, frequency, plant transformer load, and voltage of each bus section.

[0050] The hybrid dynamic modeling module achieves accurate modeling guidance under wide loads and multiple operating conditions by fusing dry and wet characteristics into a modeling formula. This ultimately solves the problems of existing technologies lacking accurate dry / wet dynamic models and having poor adaptive capabilities. Its calculation formula is as follows:

[0051] In the formula, This represents the comprehensive dynamic characteristic function during the dry-wet state transition process. This represents the input parameter vector (including load, feedwater flow rate, steam flow rate, etc.). Indicates time, Represents the dry / wet weighting coefficient (in the dry state) →1, In wet state →0), This represents the dynamic characteristic function of dry-state operation. This represents the dynamic characteristic function under wet conditions. This represents the dynamic correction term for the transition process, used to compensate for the nonlinear characteristics during the transition period.

[0052] In the formula, Represents the rate of change of the system state vector. This represents the system state vector (such as main steam temperature, main steam pressure, working fluid level, etc.). This represents the control input vector (such as water valve command, fuel command, etc.). This represents a mechanistic model based on physical laws such as the conservation of mass and energy. This represents the i-th fuzzy rule or data-driven sub-model. This indicates that its adaptive weights, based on real-time evaluation using reinforcement learning, satisfy... , The total number of fuzzy rules or sub-models is represented by a value ranging from 4 to 10, covering typical load segments and transition intervals. This fusion formula combines the generalization of the mechanistic model with the compensation capability of the data-driven model to achieve high-precision adaptive modeling of the dry-wet transition process under wide loads and multiple operating conditions.

[0053] The advantages are: it achieves accurate modeling guidance under wide loads and multiple operating conditions by integrating dry and wet state characteristics into the modeling formula, and it calculates the rate of change of the system state vector. Used for real-time updates of system state estimates, it provides accurate initial conditions and dynamic constraints for rolling optimization in the multi-model prediction optimization module. By correcting the static and dynamic deviations of traditional pure mechanistic models under varying operating conditions online, it accurately compensates for the strong nonlinear and time-varying dynamic characteristics during the dry-wet transition period. It also links with the safety monitoring module to construct a closed-loop safety correction mechanism based on model prediction. Finally, it achieves high-precision dynamic simulation and real-time state tracking of the entire dry-wet transition process within a wide load range, significantly improving the model's adaptability and engineering applicability to complex operating conditions, and laying a precise model foundation for intelligent decision-making and control.

[0054] The multi-model prediction and optimization module, based on the output of a hybrid dynamic model and the results of deep learning adaptive decision-making, uses a state prediction model (i.e., the system state-space equations) to construct a multi-objective prediction and optimization model for the transition process, addressing the problems of slow response and poor transition stability in conventional control. The system state-space equations are as follows: , In the formula, This represents the predicted output vector for key parameters (main steam temperature, main steam pressure, etc.) at a future time (k+1). This represents the system state vector at the current time (k). This represents the current control input sequence to be optimized. and These represent the state matrix and control matrix obtained by linearizing the hybrid dynamic model, respectively.

[0055] The advantage is that by constructing a multi-objective predictive optimization model for the transition process using a state prediction model (i.e., the system state-space equations), its computation... It is used to continuously predict the future trends of key parameters such as main steam temperature, main steam pressure, and working fluid level in the rolling time domain (5-10 seconds in the future), thereby quantitatively evaluating different candidate control strategies. The potential impact on future system state and safety indicators; by calculating in real time the deviation between the predicted trajectory and the dynamic safety boundary (such as the maximum allowable rate of temperature rise and the rate of pressure change), it provides a quantitative reference for the system to generate advanced and gentle adjustment commands, and uses it as the trigger condition for the model switching logic. When it is predicted that a certain parameter will enter another steady-state region, it will be triggered immediately. Finally, the system can achieve closed-loop, advanced and precise control of the core parameters of the transition process, effectively avoid parameter overshoot and violent fluctuations, and ultimately ensure the stability of the system transition process.

[0056] The multi-model prediction optimization module, based on the output of a hybrid dynamic model and the adaptive decision-making results of deep learning, constructs a multi-objective prediction optimization model for the transition process using a multi-objective optimization function (i.e., the MPC cost function). This addresses the problems of slow response and poor transition stability in conventional control. The MPC cost function is expressed as follows: In the formula, This represents the value of the objective function that needs to be minimized. This represents minimizing (min) the sum of costs (Σ) for all prediction steps within the prediction time domain (e.g., the next 5-10 seconds). The MPC algorithm solves this optimization problem in each control cycle. This represents the actual predicted value vector of key controlled parameters output by the hybrid dynamic model, including main steam temperature, main steam pressure, and water level (working fluid level) in the steam drum or separator. It is a core indicator for measuring the stability and compliance of the transition process, and its dimensions are similar to... Consistent This represents the dynamic setpoint or safety range reference trajectory of the key controlled parameter during the transition process. It is not a fixed value, but rather an ideal transition path planned based on whether the target is in a dry or wet state, and safety constraints (such as the rate of temperature rise and the rate of pressure change). This refers to the rate of change or increment of the control input vector (or manipulated variable), specifically the change in the opening command of the water supply regulating valve, the fuel quantity command, the valve control command, etc., within two adjacent control cycles. Minimizing this step aims to smooth control actions, avoid frequent and large movements of the actuators, and improve equipment lifespan and operational stability. The penalty or time cost term representing the degree of completion of the transition process is a function related to the degree and / or time of deviation of the system state from the target state. For example, it can be defined as the weighted norm integral of the difference between the state vector and the target steady state, or an estimate function of the expected remaining transition time. Minimizing this term aims to drive the system to complete the transition from one state to another more quickly and efficiently. , , These represent the weighting coefficients used to balance the tracking accuracy of the balancing parameters, the smoothness of the control action, and the transition speed.

[0057] The advantage is that by using a multi-objective optimization function (i.e., the MPC cost function) to construct a multi-objective prediction optimization model for the transition process, the objective function value to be minimized can be calculated. It is used to quantitatively evaluate and comprehensively weigh multiple conflicting objectives of control performance (tracking accuracy, motion amplitude, and transition time) by adjusting the weighting coefficients online. , , This allows for dynamic optimization of the control strategy's focus; and under the premise of satisfying various hard constraints (such as valve stroke and rate limits), it selects the globally optimal control input sequence, thereby effectively suppressing the high-frequency impact of measurement noise and process disturbances on control quality, generating user-friendly and refined adjustment commands that balance fast response and equipment lifespan, and finally achieving dry-wet state conversion in a multi-objective comprehensive optimal manner while ensuring process safety. For example, while ensuring the fastest conversion speed, it reduces the fluctuation amplitude of the working fluid level and stably controls the main steam pressure change rate within a safe range of ±0.3MPa / min, significantly improving the economy and equipment safety of the conversion process.

[0058] The multi-model prediction and optimization module, based on the state prediction model and the calculation results of the multi-objective optimization function, predicts key parameters such as main steam temperature, main steam pressure, and working fluid level within the next 5-10 seconds. The optimization objective is to achieve the fastest transition speed and the minimum parameter fluctuation. It also performs secondary optimization on the adjustment parameter instructions output by the deep learning adaptive decision-making module, generating the optimal control sequence and transmitting it to the fully automatic execution module. Simultaneously, model switching logic enables smooth switching between dry and wet states and between different models during transitions, ensuring the stability of control performance across a wide load range.

[0059] The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module. Based on the fully automatic sequential control (SCS) logic, it drives the corresponding actuators (such as water supply regulating valves, steam valves, circulating water pumps, etc.) to complete precise actions, realizing the fully automatic execution of the dry-wet state conversion process. This module replaces manual operation with standardized execution logic, solving the problems of cumbersome and high-risk traditional state conversion steps. At the same time, it collects the action feedback data of the actuators in real time to form a closed-loop control, ensuring the accurate implementation of the adjustment commands.

[0060] The integrated linkage module is based on a DCS deep integration architecture, which fully integrates auxiliary control systems such as circulating water and soot blowing with the main control system to build a unified linkage control logic. It receives action commands from the fully automatic execution module to synchronously coordinate the operating status of each auxiliary control system (such as synchronously adjusting the circulating water flow and coordinating the timing of soot blowing during the transition process), solving the problems of information silos and low linkage efficiency in traditional auxiliary control systems. At the same time, it aggregates the operating data of each system to a unified monitoring platform to provide data support for subsequent monitoring and maintenance.

[0061] The safety monitoring module collects unit operating parameters (such as heating surface wall temperature, main steam pressure fluctuation amplitude, hydrodynamic parameters, etc.) and operating status data of each module in real time. By setting safety thresholds and trend warning curves, it monitors the risks of over-temperature, over-pressure, and hydrodynamic instability during the transition process in real time. When the monitored parameters exceed the safety boundary, it immediately outputs a warning signal to the deep learning adaptive decision-making module and the fully automatic execution module. The system automatically triggers emergency adjustment commands (such as adjusting valve opening and reducing the load change rate) and pushes the warning information to the operation and maintenance management module, solving the problem of high safety risks in traditional transition processes.

[0062] The operation and maintenance management module receives early warning information from the safety monitoring module and operational data from each module, constructs a unit operation status assessment model, performs real-time calculation and trend analysis of equipment wear and operating efficiency (such as coal consumption), and stores historical data and adjustment strategies for dry-wet state transitions to form a knowledge base, providing decision support for subsequent strategy optimization and equipment maintenance. In addition, this module also supports manual intervention. When encountering extreme operating conditions, operation and maintenance personnel can issue manual intervention commands through this module to ensure the reliability of system operation, thereby helping to solve the problems of parameter drift and equipment aging in long-term system operation.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dry-wet state switching adaptive regulation system for deep peak shaving in thermal power units, characterized in that, It includes a classification data acquisition module, a data preprocessing module, a hybrid dynamic modeling module, a deep learning adaptive decision-making module, a multi-model prediction and optimization module, a fully automated execution module, an integrated linkage module, a security monitoring module, and an operation and maintenance management module. The classification-based data acquisition module collects raw data in all dimensions at four classification sites, and the collected data is transmitted to the data preprocessing module in real time. The data preprocessing module preprocesses the raw data collected from the four sites across all dimensions, generating a high-quality dataset that is then transmitted to the hybrid dynamic modeling module and the deep learning adaptive decision-making module. The hybrid dynamic modeling module constructs a dynamic model of dry-wet state transition and outputs accurate modeling guidance by fusing dry-wet state characteristics into a modeling formula. The deep learning adaptive decision-making module, based on the preprocessed high-quality dataset and the output of the hybrid dynamic model, uses deep learning algorithms to make adaptive decisions on the timing of dry-wet state transitions and the adjustment parameters. The multi-model prediction optimization module, based on the output of the hybrid dynamic model and the results of deep learning adaptive decision-making, introduces a multi-model prediction control algorithm to predict key parameters within the next 5-10 seconds, thereby performing secondary optimization on the adjustment parameter instructions output by the deep learning adaptive decision-making module, generating the optimal control sequence and transmitting it to the fully automatic execution module. The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module, and drives the corresponding actuator to complete precise actions based on the fully automatic sequential control logic, thereby realizing the fully automatic execution of the dry-wet state conversion process. The integrated linkage module is based on a DCS deep integration architecture, which fully integrates the auxiliary control system and the main control system to build a unified linkage control logic. At the same time, it receives the action instructions of the fully automatic execution module and synchronously coordinates the operating status of each auxiliary control system. The safety monitoring module collects unit operating parameters and operating status data of each module in real time. By setting safety thresholds and trend warning curves, it monitors the risks of over-temperature, over-pressure and hydrodynamic instability during the transition process in real time and outputs warning information. The operation and maintenance management module receives early warning information from the safety monitoring module and operating data from each module, constructs a unit operation status assessment model, performs real-time calculation and trend analysis of equipment wear and operating efficiency, and realizes immediate equipment maintenance.

2. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 1, characterized in that, The categorized data acquisition module includes four stations: a boiler-side data acquisition unit, a turbine-side data acquisition unit, an auxiliary equipment-side data acquisition unit, and an electrical-side data acquisition unit.

3. The dry-wet state conversion adaptive adjustment system for deep peak shaving of thermal power units according to claim 2, characterized in that, The boiler-side data acquisition unit collects key parameters of the boiler combustion and steam-water system through distributed temperature / pressure transmitters and flow meters.

4. The dry-wet state conversion adaptive adjustment system for deep peak shaving of thermal power units according to claim 2, characterized in that, The turbine-side data acquisition unit collects operating parameters of the turbine and its regulation system through vibration sensors, displacement sensors, and thermal instruments.

5. The dry-wet state conversion adaptive adjustment system for deep peak shaving of thermal power units according to claim 2, characterized in that, The auxiliary equipment side data acquisition unit collects the status and process parameters of the main auxiliary equipment through the pump / fan intelligent monitoring device and valve positioner.

6. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 2, characterized in that, The electrical side data acquisition unit collects parameters of the generator and plant power system through electrical measurement and protection devices.

7. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 1, characterized in that, The hybrid dynamic modeling module achieves accurate modeling guidance under wide loads and multiple operating conditions by fusing dry and wet characteristics into a modeling formula. The calculation formula is as follows: ， In the formula, This represents the comprehensive dynamic characteristic function during the dry-wet state transition process. Represents the input parameter vector. Indicates time, This represents the dry / wet weighting coefficient. This represents the dynamic characteristic function of dry-state operation. This represents the dynamic characteristic function under wet conditions. This represents the dynamic correction term during the transition process; ， In the formula, Represents the rate of change of the system state vector. Represents the system state vector. Represents the control input vector. This represents a mechanistic model based on physical laws such as the conservation of mass and energy. This represents the i-th fuzzy rule or data-driven sub-model. This indicates that it uses adaptive weights based on real-time evaluation using reinforcement learning. This represents the total number of fuzzy rules or sub-models, with a value ranging from 4 to 10.

8. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 1, characterized in that, The multi-model prediction and optimization module uses a state prediction model to construct a multi-objective prediction and optimization model for the transition process, specifically expressed as follows: , In the formula, This represents the predicted output vector of key parameters at a future time (k+1). This represents the system state vector at the current time (k). This represents the current control input sequence to be optimized. and These represent the state matrix and control matrix obtained by linearizing the hybrid dynamic model, respectively.

9. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 1, characterized in that, The multi-model prediction optimization module uses a multi-objective optimization function to construct a multi-objective prediction optimization model for the transition process, expressed as: In the formula, This represents the value of the objective function that needs to be minimized. This means minimizing the sum of costs for all prediction steps within the prediction time domain. This represents the vector of actual predicted values ​​of the key controlled parameters output by the hybrid dynamic model. This indicates the dynamic setpoint or safety range reference trajectory of the key controlled parameter during the transition process. This represents the rate of change or increment of the control input vector. The penalty or time cost term represents the degree of completion of the transition process. , , These represent the weighting coefficients used to balance the tracking accuracy of the balancing parameters, the smoothness of the control action, and the transition speed.

10. The dry-wet state switching adaptive adjustment system for deep peak shaving of thermal power units according to claim 9, characterized in that, The multi-model prediction and optimization module, based on the state prediction model and the calculation results of the multi-objective optimization function, predicts key parameters such as main steam temperature, main steam pressure and working fluid level within the next 5-10 seconds. It also performs secondary optimization on the adjustment parameter instructions output by the deep learning adaptive decision module, generates the optimal control sequence, and transmits it to the fully automatic execution module. At the same time, it achieves smooth switching between dry state, wet state, and different models during the transition process through model switching logic. The fully automatic execution module receives the optimal control sequence output by the multi-model prediction and optimization module. Based on the fully automatic sequential control logic, it drives the corresponding water regulating valve, steam valve and circulating water pump to complete precise actions, realizing the fully automatic execution of the dry-wet state conversion process. This module replaces manual operation with standardized execution logic; at the same time, it collects the action feedback data of the actuator in real time to form a closed-loop control.

Images