Direct air cooling unit backpressure optimization and control method based on digital twinning

Abstract

The invention belongs to the field of direct air cooling unit backpressure optimization and control, particularly relates to a direct air cooling unit backpressure optimization and control method based on digital twinning, and aims to realize real-time optimization and accurate control of backpressure under variable working conditions. The method comprises the steps that operation parameters of the direct air cooling unit are collected in real time, and standardized operation data are obtained after data preprocessing; and constructing a digital twinborn model containing a dynoNet parameter identification model on the basis of a mechanism modeling and data-driven fusion mode. And simulating the back pressure of the unit by using the standardized operation data, and monitoring an error between a simulated back pressure value and an actually measured back pressure value. And when the error exceeds a set threshold value, an online parameter identification and model correction mechanism is triggered. And utilizing the corrected digital twinborn model to carry out optimization solution by taking comprehensive economic benefit maximization as a target, and determining an optimal backpressure set value under the current working condition. And the optimal backpressure set value is used as a tracking target of model prediction control, and a model prediction control algorithm is adopted to realize optimal adjustment of the backpressure.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of direct air-cooled machine back pressure optimization and control, and particularly relates to a direct air-cooled unit back pressure optimization and control method based on digital twinning. BACKGROUND

[0002] The direct air-cooled system has been widely used in coal-rich and water-scarce areas of thermal power generating units due to its excellent water-saving performance and low initial investment cost, and has become one of the core cooling technologies for power production in the area. The system directly cools the turbine exhaust steam by driving the environmental air through the axial flow fan, and the stability and economy of the operation back pressure are crucial to the efficiency and safety of the entire unit.

[0003] However, with the deepening of energy structure transformation, coal-fired units need to frequently participate in deep peak regulation of the power grid, resulting in frequent rapid fluctuations in their load. Under this variable operating condition, the operating back pressure of the direct air-cooled system is difficult to maintain optimal, and the traditional optimization control method based on fixed parameter mechanism model or pure data-driven model exposes obvious limitations: the mechanism model has clear physical meaning, but the key parameters (such as heat transfer coefficient) are easy to drift with the environment and operating state, resulting in model inaccuracy; and the pure data-driven model lacks physical constraints, has poor dynamic characteristic capturing ability and weak generalization, and most existing methods fail to achieve online estimation and update of parameters and state variables. Therefore, under the influence of load and environmental disturbance, the existing strategy is difficult to guarantee the accuracy, economy and engineering real-time of back pressure control at the same time. SUMMARY

[0004] The purpose of the present application is to provide a direct air-cooled unit back pressure optimization and control method based on digital twinning, to realize real-time optimization and accurate control of back pressure under variable operating conditions, thereby effectively improving the comprehensive operating economy and control robustness of the direct air-cooled unit.

[0005] To achieve the above purpose, the present application adopts the following technical solutions: This invention provides a method for optimizing and controlling back pressure of a direct air-cooled unit based on digital twins, comprising: S1: Data acquisition and processing step: Real-time acquisition of operating parameters of the direct air-cooled unit through a plant-level monitoring information system, and preprocessing of the operating parameters to obtain standardized operating data. S2: Model construction and adaptive update step: Based on the fusion of mechanism modeling and data-driven approach, a digital twin model including a dynoNet parameter identification model is constructed. The unit back pressure is simulated in real time using standardized operating data, and the error between the simulated back pressure value and the measured back pressure value is monitored. When the error exceeds a set threshold, an online parameter identification and model correction mechanism is triggered to dynamically adjust the parameters of the digital twin model. S3: Back pressure optimization solution step: Using the digital twin model corrected in step S2, optimization is performed with the goal of maximizing comprehensive economic benefits to determine the optimal back pressure setpoint under the current operating conditions. S4: Closed-loop control steps: The optimal back pressure setpoint is used as the tracking target of model predictive control. The model predictive control algorithm that integrates the dynoNet parameter identification model is adopted to continuously optimize the fan speed and send the optimized control command to the field actuator to realize the closed-loop optimization and adjustment of back pressure.

[0006] In step S1, the operating parameters include coal feed rate, main steam flow rate, main steam pressure, main steam temperature, reheat steam flow rate, reheat steam pressure, reheat steam temperature, unit load, back pressure, ambient temperature, and fan speed. Data preprocessing includes outlier handling, missing value imputation, unit conversion, and timestamp synchronization.

[0007] In step S2, the online parameter identification and model correction mechanism is triggered, specifically including: caching the input-output data sequence for a set time period prior to the current moment using a sliding window. For a new sampling point, a one-step mechanism simulation is performed based on the current model parameters. When the obtained instantaneous simulation error exceeds a set error threshold, and the time interval since the last model parameter update exceeds a set duration, the online parameter identification process is activated. Using the data within the current sliding window as the sample set, lightweight iterative training is performed to update the network parameters of the dynoNet parameter identification model.

[0008] During lightweight iterative training, a partial parameter update strategy is adopted, which updates only the network parameters used to compensate for the nonlinear time-varying characteristics of the system, while fixing the network parameters used to describe the fundamental physical dynamics.

[0009] Lightweight iterative training uses a loss function that includes a regularization term to constrain the drift of model parameters.

[0010] In step S3, optimization is performed with the goal of maximizing comprehensive economic benefits. Specifically, this includes: constructing a comprehensive economic benefit objective function, which simultaneously quantifies the energy cost caused by changes in fan power consumption and the coal cost caused by changes in coal feed rate; determining the safe operating range of back pressure based on the current unit load rate; and within this safe operating range, using the comprehensive economic benefit objective function as the evaluation criterion to obtain the optimal back pressure setpoint.

[0011] When back pressure adjustment causes the unit load to deviate from the automatic power generation control command, the unit load is restored to the set value by dynamically correcting the coal feed rate.

[0012] In step S4, the model prediction control algorithm uses the online updated and corrected dynoNet parameter identification model as the internal prediction model.

[0013] In the forward prediction process of the model predictive control algorithm, the ambient temperature, fan speed, number of operating fans and initial back pressure are used as external inputs; among them, the ambient temperature and number of operating fans are kept constant in the prediction time domain.

[0014] The optimization process of the model predictive control algorithm must satisfy the constraints determined by the dynamic characteristics of the direct air-cooled unit system, as well as the control quantity variation constraints determined by the physical limits of the fan actuator.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This application provides a method for optimizing and controlling back pressure of a direct air-cooled unit based on digital twins. It constructs a digital twin model by fusing mechanistic modeling with dynoNet data-driven approaches. An online parameter identification mechanism is triggered by both error thresholds and time intervals, and real-time running data is cached via a sliding window to ensure timely and targeted model updates. Simultaneously, a strategy is adopted to update only a portion of the parameters—fixing the basic physical parameters and updating only the nonlinear time-varying compensation parameters—and parameter drift is constrained by regularization terms. This preserves the physical rationality of the mechanistic model while rapidly adapting to time-varying operating conditions such as frequent load fluctuations and ambient temperature changes. This significantly improves the accuracy of back pressure simulation and the long-term stability of the model, effectively addressing the technical challenge of traditional models failing to adapt to drastic load changes in deep peak-shaving scenarios for coal-fired units.

[0016] 2. The method provided in this application achieves synergistic optimization of two core costs by constructing a comprehensive economic benefit objective function that encompasses both the energy consumption cost of the wind turbine and the coal feed cost. This avoids the problem of wind turbine energy consumption exceeding limits due to either focusing solely on optimizing wind turbine energy consumption while ignoring the increase in coal costs or focusing solely on coal savings. Furthermore, to address the load deviation issue that may be caused by back pressure adjustments, the method dynamically corrects the coal feed to restore the unit load to the requirements of the automatic power generation control command. This ensures the stability of power generation revenue while accurately searching for the optimal back pressure based on maximizing economic benefits, effectively reducing the total operating cost of the unit and creating significant economic value for power generation companies.

[0017] 3. The method provided in this application constructs a closed-loop control system that integrates physical information. It uses an online-updated and corrected dynoNet parameter identification model as the internal predictive model for model predictive control, and adjusts the fan speed in real time through a rolling optimization strategy to achieve precise tracking of optimal back pressure. During the control process, it strictly adheres to the system dynamic constraints of the direct air-cooled unit and the physical limit constraints of the fan actuator, effectively resisting external disturbances such as ambient temperature fluctuations and rapid load changes, thus solving the problems of lag and weak anti-interference capability of traditional control strategies. Simultaneously, back pressure optimization is performed within the safe operating range determined based on the current load rate, fundamentally avoiding the operational risks caused by back pressure exceeding the safe range and ensuring long-term stable operation of the unit. Attached Figure Description

[0018] Figure 1 This application provides a method for optimizing and controlling the back pressure of a direct air-cooled unit based on digital twins. Figure 2 This is a flowchart of an online parameter identification process provided in an embodiment of this application; Figure 3 This is a back pressure optimization flowchart provided in an embodiment of this application. Detailed Implementation

[0019] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] For example, refer to Figure 1 This application provides a method for optimizing and controlling back pressure of a direct air-cooled unit based on digital twins, including: S1: Data acquisition and processing steps: The operating parameters of the direct air-cooled unit are collected in real time through the plant-level monitoring information system, and the operating parameters are preprocessed to obtain standardized operating data.

[0021] In some embodiments, in step S1, the operating parameters include coal feed rate, main steam flow rate, main steam pressure, main steam temperature, reheat steam flow rate, reheat steam pressure, reheat steam temperature, unit load, back pressure, ambient temperature, and fan speed. Based on sensors arranged in the direct air-cooled unit, measurements of coal feed rate, main steam flow rate, main steam pressure, main steam temperature, reheat steam flow rate, reheat steam pressure, reheat steam temperature, load, back pressure, ambient temperature, and fan speed are collected.

[0022] Data preprocessing includes outlier handling, missing value imputation, unit conversion, and timestamp synchronization. Outlier handling in data preprocessing can eliminate erroneous data generated during the acquisition of operating parameters of direct air-cooled units due to sensor failures, interference, etc., avoiding misleading subsequent model building and optimization decisions; missing value imputation can fill data acquisition gaps and ensure the integrity of key parameter sequences such as coal feed, main steam parameters, and back pressure; unit conversion can unify the parameter measurement dimensions of different sensors and monitoring modules, eliminating calculation deviations caused by unit differences; timestamp synchronization can accurately align the time sequence information of various operating parameters, ensuring that the data reflects the unit's operating status at the same moment. The four functions work together to provide reliable, complete, and consistent standardized operating data for the digital twin model, directly ensuring the accuracy of subsequent model simulation, the effectiveness of online parameter identification, and the precision of optimal back pressure solution and closed-loop control, laying a solid data foundation for the safe and economical operation of the unit.

[0023] S2: Model Building and Adaptive Update Steps: Based on a fusion of mechanistic modeling and data-driven approaches, a digital twin model incorporating a dynoNet parameter identification model is constructed. Real-time simulation of unit back pressure is performed using standardized operational data, and the error between the simulated and measured back pressure values ​​is monitored. When the error exceeds a set threshold, an online parameter identification and model correction mechanism is triggered to dynamically adjust the parameters of the digital twin model.

[0024] More specifically, the model includes a dynoNet parameter identification model as the core data-driven unit of the digital twin model. It takes preprocessed standardized unit operating data as input and outputs the unit back pressure and the identified heat transfer coefficient. During the offline training phase, a multi-loss term collaborative optimization strategy is employed. These loss terms specifically include: data loss (the deviation between the model's output state variables and the measured state variables), physical loss (the deviation between the model's output state variables and the simulated state variables of the mechanistic model, ensuring physical consistency of the model), simulation error loss (the deviation between the simulated state variables of the mechanistic model and the measured state variables), and boundary loss (the loss formed by applying empirical upper and lower bound constraints to unknown physical parameters). Among these, data loss... Physical loss and simulation error All are calculated using the mean square error, and the formulas are as follows: in, , and The simulated state variables are obtained by substituting the measured state variables, the output state variables of the dynoNet parameter identification model, and the identified dynamic parameters into the discretized mechanism model. The number of samples; For the first The adaptive weights for each sample point are used. The discretized mechanism model is the core component of the mechanism modeling part of the digital twin model. Its construction is entirely based on the physical operating laws of the direct air-cooled unit, including the heat transfer mechanism, fluid flow laws, and energy conservation law of the cold-end system of the direct air-cooled unit as core principles. For the key physical processes of back pressure formation, continuous-time domain mechanism equations are established and converted into discrete-time domain mathematical models using numerical discretization methods (such as the Euler method and the Runge-Kutta method). The inputs to the discretized mechanism model are key physical parameters of unit operation, such as coal feed rate, main steam parameters, and ambient temperature. The outputs are simulation state variables derived from physical laws, such as simulated back pressure and intermediate parameters during heat transfer.

[0025] To address the issue that while the total loss of a neural network decreases during training, certain temporal features may not be fully captured, this application proposes setting a set of local weight multipliers that can adaptively adjust during training. This allows the optimization process to focus more on high residual samples and improve the overall approximation ability. The update rule is as follows: in, It is the first The residuals of each loss term for each sample. It is the attenuation parameter. This is the learning rate.

[0026] For example, boundary loss Represented as: in, The first output of the dynoNet parameter identification model One parameter, For the number of parameters, and The first The lower and upper limits of each parameter.

[0027] For example, the total loss function for: in, , , and These are the weights for different loss terms.

[0028] For example, in step S2, an online parameter identification and model correction mechanism is triggered, specifically including: caching the input-output data sequence for a set time period prior to the current moment using a sliding window method. For new sampling points... A one-step mechanism simulation is performed based on the current model parameters. When the obtained instantaneous simulation error exceeds the set error threshold... Furthermore, the time interval since the last model parameter update exceeds the set duration. ( Indicates the length of the sliding window. When the sampling time is reached, the online parameter identification process is activated. Using the data within the current sliding window as the sample set, and historical parameters and network weights as initial values, lightweight iterative training is performed to update the network parameters of the dynoNet parameter identification model.

[0029] As one possible implementation, a partial parameter update strategy is adopted during lightweight iterative training, which updates only the network parameters used to compensate for the nonlinear time-varying characteristics of the system, while fixing the network parameters used to describe the fundamental physical dynamics.

[0030] The network parameters used to describe the fundamental physical dynamics are the core link between the dynoNet parameter identification model and the discretized mechanism model. Essentially, they are a digital mapping of the fundamental physical laws governing cold-end heat transfer and fluid flow in direct-cooled units. Fixing these parameters ensures that the model remains anchored to the core physical characteristics of the unit, avoiding parameter drift that might occur during full-parameter updates. The core objective of lightweight iterative training is to adapt to online parameter correction under engineering scenarios. The network parameters used to compensate for the nonlinear time-varying characteristics of the system, such as dynamic compensation parameters to adapt to load fluctuations and sudden changes in ambient temperature, have a much lower dimensionality than the full network parameters. Updating only these parameters significantly reduces the computational load and convergence time of iterative training, avoids computational delays caused by full-parameter retraining, and ensures that online parameter identification can quickly respond to changes in operating conditions, meeting the real-time requirements of backpressure optimization and closed-loop control. The fundamental physical dynamics of direct air-cooled units, such as the flow channel characteristics determined by the equipment structure and the basic heat transfer capacity determined by the materials, are relatively stable during the unit's operating cycle and do not require frequent adjustments. However, the nonlinear time-varying characteristics of the system, such as drastic load fluctuations and changes in heat transfer efficiency caused by dust accumulation, are key to adapting to operating conditions. Targeted updates to these compensation parameters can achieve a balance between stable fundamental physical characteristics and accurate adaptation to time-varying characteristics, avoiding abrupt changes in model output that may be caused by full parameter updates. This allows the model to maintain smooth simulation and control performance under scenarios with frequent load changes and environmental disturbances, improving the robustness of back pressure control.

[0031] As one possible implementation, lightweight iterative training uses a loss function that includes a regularization term to constrain the drift of the model parameters, as shown below: in, express Before the moment The time corresponding to each sample; regularization term Constraint parameter drift magnitude; is the coefficient of the regularization term.

[0032] After lightweight iterative training is completed, the estimated heat transfer coefficient obtained through identification and the network state of the dynoNet parameter identification model are updated and used as the initial parameters and state benchmark for the next cycle of digital twin model operation. At the same time, the sliding window is shifted forward by one time step along the time sequence to continuously cache the latest running data, ultimately forming a closed-loop adaptive update mechanism for model parameters and state.

[0033] For example, such as Figure 2As shown, the system continuously caches the input-output data sequence for a set time period prior to the current moment using a sliding window. For newly acquired sampling points, it performs a one-step mechanism simulation based on the discretized mechanism model and existing parameters in the current digital twin model, calculating the instantaneous simulation error between the simulated back pressure value and the measured back pressure value. If the instantaneous simulation error does not exceed the set error threshold, the identification process is not triggered, and historical data is used, i.e., the current model parameters are used to continue the real-time back pressure simulation. The sliding window time sequence is moved forward and new sampled data is cached, entering the next cycle of error monitoring. When the instantaneous simulation error exceeds the set error threshold, and the time interval since the last model parameter update exceeds the set duration, the online parameter identification process is activated. Subsequently, the cached data in the sliding window is used as the sample set, along with historical model parameters and dynoN... The network weights of the dynoNet parameter identification model are used as initial values ​​to initiate lightweight iterative training. During training, a partial parameter update strategy is adopted (only the network parameters used to compensate for the nonlinear time-varying characteristics of the system are updated, while the network parameters used to describe the basic physical dynamics are fixed). A regularization term is added to the loss function to constrain the parameter drift amplitude, and an early stopping mechanism is introduced to avoid overfitting. After training, the estimated heat transfer coefficient obtained by identification and the network state of the dynoNet parameter identification model are updated and used as the initial parameters and state benchmark for the next cycle of digital twin model operation. The sliding window is pushed forward by one time step along the time sequence, and finally a closed-loop adaptive update mechanism for model parameters is formed to ensure that the digital twin model always adapts to the time-varying operating conditions of the unit and provides accurate support for subsequent back pressure optimization and closed-loop control.

[0034] S3: Back pressure optimization solution steps: Using the digital twin model corrected in step S2, optimize the solution with the goal of maximizing comprehensive economic benefits to determine the optimal back pressure setting value under the current working conditions.

[0035] For example, in step S3, optimization is performed with the goal of maximizing comprehensive economic benefits. This specifically includes: constructing a comprehensive economic benefit objective function, which simultaneously quantifies the energy cost caused by changes in fan power consumption and the coal cost caused by changes in coal feed rate. The safe operating range of back pressure is determined based on the current unit load rate. Within this safe operating range, the optimal back pressure setpoint is obtained by using the comprehensive economic benefit objective function as the evaluation criterion.

[0036] As one possible approach, if back pressure adjustment causes the unit load to deviate from the automatic power generation control command, the unit load can be restored to the set value by dynamically correcting the coal feed rate.

[0037] More specifically, establish a total power consumption model for the wind turbine group, and the total power consumption of the axial flow wind turbine group. The expression is: in, This refers to the number of operating wind turbines. This refers to the rated power of a single fan. and The first The actual speed and rated speed of the typhoon.

[0038] The energy costs caused by changes in wind turbine power consumption and the coal costs caused by changes in coal supply are jointly included in the evaluation indicators. Electricity and coal prices are used as the basis for economic quantification, ensuring that decisions directly reflect changes in economic benefits. The constructed economic benefit objective function is as follows: in, To optimize the cost changes caused by the change in the rotational speed of the wind turbine groups before and after; This represents the change in coal supply. For coal prices; This represents the change in power consumption of the wind turbine; For electricity prices.

[0039] The safe operating range of back pressure is determined based on the unit load rate under the current operating conditions. The system back pressure is changed by adjusting the fan speed, and the operating cost corresponding to different back pressures is calculated based on the impact of back pressure fluctuations on the unit load. Using the comprehensive economic benefit objective function as the evaluation criterion, the back pressure corresponding to the optimal comprehensive economic benefit is determined as the optimal back pressure setpoint under this operating condition. When the back pressure adjustment causes the unit load to deviate from the automatic power generation control command, the unit load is restored to the setpoint by dynamically correcting the coal feed rate. To meet the requirements of online optimization and engineering applications, linear mapping relationships between unit load and main steam flow rate and between main steam flow rate and coal feed rate are established, which can be updated online. During the optimization iteration process, if there is a deviation between the simulated unit load output by the digital twin model and the automatic power generation control command, the main steam flow rate is dynamically adjusted according to the magnitude of the deviation, and a corresponding coal feed command is generated based on the linear mapping relationship between the main steam flow rate and the coal feed rate.

[0040] For example, the linear mapping relationship between unit load and main steam flow that can be updated online is as follows: ,in Main steam flow rate, For unit load, and The coefficients are those of a first-order polynomial; the linear mapping relationship between the main steam flow rate and the coal feed rate is: ,in For coal feed rate, and denoted as the coefficients of a linear polynomial.

[0041] S4: Closed-loop control steps: The optimal back pressure setpoint is used as the tracking target of model predictive control. The model predictive control algorithm that integrates the dynoNet parameter identification model is adopted to continuously optimize the fan speed and send the optimized control command to the field actuator to realize the closed-loop optimization and adjustment of back pressure.

[0042] For example, in step S4, the model prediction control algorithm uses the online updated and corrected dynoNet parameter identification model as the internal prediction model.

[0043] As one possible implementation, in step S4, a back pressure controller based on model predictive control (MPC) is constructed. This controller aims to track the optimal back pressure, satisfy system constraints, and achieve rolling optimization. Its core modules include a predictive model, a rolling optimizer, and a feedback correction unit. The predictive model is provided by an online-updated and corrected dynoNet parameter identification model. The rolling optimizer is responsible for solving the optimal control sequence for the fan speed in the prediction time domain. The feedback correction unit corrects the control command based on the deviation between the measured and predicted back pressure.

[0044] As one possible implementation, in the forward prediction process of the model predictive control algorithm, the ambient temperature, fan speed, number of operating fans, and initial back pressure are taken as external inputs; among them, the ambient temperature and number of operating fans are kept constant in the prediction time domain.

[0045] The online-updated and corrected dynoNet parameter identification model for jointly estimating back pressure is used as the internal prediction model and combined with model predictive control to construct a predictive control architecture that integrates physical information. The optimal back pressure setpoint determined in step S3 is used as the control setpoint in subsequent control processes. In the forward recursive prediction process, the ambient temperature and the number of operating wind turbines are kept constant in the prediction time domain. The back pressure controller based on model predictive control (MPC) uses a model predictive control algorithm that integrates the dynoNet parameter identification model to achieve online optimization generation of wind turbine speed. Ambient temperature, wind turbine speed, number of operating wind turbines, and initial back pressure are all used as external input parameters to be fed into the prediction model. The future time period provided by the external scheduling system ( (Step) Automatic power generation control forward information, constructing exhaust steam flow With this automatic power generation control forward information linear mapping relationship ( It is updated online, and a feedforward input sequence for exhaust steam flow is generated accordingly; the back pressure control process is a nonlinear model predictive control process, specifically defined as follows: in, To predict the step size; To control the step size; Output back pressure for the model; This is the optimal back pressure reference value; To control the amount of input variation; It is the weight matrix of the output deviation and the reference value; It is a weight matrix; As a possible implementation, the optimization process of the model predictive control algorithm needs to satisfy the constraints determined by the dynamic characteristics of the direct air-cooled unit system, as well as the control quantity variation constraints determined by the physical limits of the fan actuator.

[0046] For example, the system dynamics constraints that need to be satisfied are: in, It is a system dynamics mapping function; For controlling input.

[0047] For example, the constraint on the change of control quantity is: For example, the inequalities and equality constraints imposed by the system dynamics constraints and control variable change constraints must be satisfied as follows: in, This refers to the system state variable, specifically the back pressure. and These represent the applied inequality and equality constraints, respectively.

[0048] For example, such as Figure 3 As shown, based on the digital twin model that has been updated and corrected online, the safe operating range of back pressure is first determined according to the unit load rate under the current operating conditions to ensure that the optimization process meets the unit's operational safety requirements. Then, the system back pressure is changed by traversing the fan speed, and the operating cost corresponding to different back pressures is calculated by combining the comprehensive economic benefit objective function (quantifying the energy cost caused by changes in fan power consumption and the coal cost caused by changes in coal feed). During this process, it is judged in real time whether the unit load deviates from the automatic generation control (AGC) command. If a deviation occurs, the coal feed is dynamically corrected by coordinating the system to restore the unit load to the set value and ensure the stability of power generation revenue. Finally, by comparing the comprehensive economic benefits corresponding to different back pressures, the cost-optimal operating state is recorded and the optimal back pressure set value under this condition is output, providing a precise tracking target for subsequent model predictive control and realizing the synergistic unity of back pressure optimization and safe and economical unit operation.

[0049] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0050] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for back pressure optimization and control of direct air-cooled units based on digital twins, characterized in that, include: S1: Data acquisition and processing steps: The operating parameters of the direct air-cooled unit are acquired in real time through the plant-level monitoring information system, and the operating parameters are preprocessed to obtain standardized operating data; S2: Model building and adaptive update steps: Based on the fusion of mechanism modeling and data-driven approach, a digital twin model containing a dynoNet parameter identification model is constructed; the standardized operating data is used to perform real-time simulation of the unit back pressure and monitor the error between the simulated back pressure value and the measured back pressure value; when the error exceeds a set threshold, an online parameter identification and model correction mechanism is triggered to dynamically adjust the parameters of the digital twin model; S3: Back pressure optimization solution step: Using the digital twin model corrected in step S2, optimize the solution with the goal of maximizing comprehensive economic benefits to determine the optimal back pressure setpoint under the current operating conditions; S4: Closed-loop control step: Use the optimal back pressure setpoint as the tracking target of model predictive control, adopt the model predictive control algorithm that integrates the dynoNet parameter identification model, continuously optimize the fan speed, and send the optimized control command to the field actuator to realize the closed-loop optimization adjustment of back pressure.

2. The method for back pressure optimization and control of a direct air-cooled unit based on digital twins according to claim 1, characterized in that, In step S1, the operating parameters include coal feed rate, main steam flow rate, main steam pressure, main steam temperature, reheat steam flow rate, reheat steam pressure, reheat steam temperature, unit load, back pressure, ambient temperature, and fan speed; the data preprocessing includes outlier handling, missing value imputation, unit conversion, and timestamp synchronization.

3. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 1, characterized in that, In step S2, the triggering of the online parameter identification and model correction mechanism specifically includes: caching the input-output data sequence for a set time period prior to the current moment using a sliding window; performing a one-step mechanism simulation based on the current model parameters for a new sampling point; activating the online parameter identification process when the obtained instantaneous simulation error exceeds a set error threshold and the time interval since the last model parameter update exceeds a set duration; and performing lightweight iterative training using the data within the current sliding window as the sample set to update the network parameters of the dynoNet parameter identification model.

4. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 3, characterized in that, During the lightweight iterative training process, a partial parameter update strategy is adopted, which updates only the network parameters used to compensate for the nonlinear time-varying characteristics of the system, while fixing the network parameters used to describe the fundamental physical dynamics.

5. The method for back pressure optimization and control of a direct air-cooled unit based on digital twins according to claim 3, characterized in that, The loss function used in the lightweight iterative training includes a regularization term to constrain the drift of the model parameters.

6. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 1, characterized in that, In step S3, the optimization solution aimed at maximizing comprehensive economic benefits specifically includes: constructing a comprehensive economic benefit objective function, which simultaneously quantifies the energy cost caused by changes in wind turbine power consumption and the coal cost caused by changes in coal feed rate; determining the safe operating range of back pressure based on the current unit load rate; and within the safe operating range, using the comprehensive economic benefit objective function as the evaluation criterion, solving for the optimal back pressure setpoint.

7. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 6, characterized in that, When back pressure adjustment causes the unit load to deviate from the automatic power generation control command, the unit load is restored to the set value by dynamically correcting the coal feed rate.

8. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 1, characterized in that, In step S4, the model prediction control algorithm uses the dynoNet parameter identification model, which has been updated and corrected online, as the internal prediction model.

9. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 8, characterized in that, In the forward prediction process of the model predictive control algorithm, the ambient temperature, fan speed, number of operating fans and initial back pressure are used as external inputs; wherein, the ambient temperature and number of operating fans are kept constant in the prediction time domain.

10. The method for back pressure optimization and control of a direct air-cooled unit based on digital twin as described in claim 8, characterized in that, The optimization process of the model predictive control algorithm must satisfy the constraints determined by the dynamic characteristics of the direct air-cooled unit system, as well as the control quantity variation constraints determined by the physical limits of the fan actuator.

Images