Data and mechanism hybrid modeling method and device for coordinated control of thermal power generating unit

Abstract

The invention discloses a thermal power generating unit coordination control data and mechanism hybrid modeling method and device, computer equipment and a computer readable storage medium, and relates to the technical field of thermal energy engineering and power system modeling and control. Establishing a high-precision dynamic mechanism model of the working medium flow and heat transfer process of the boiler-steam turbine system; then, constructing an error correction model fusing the long-short-term memory network, the convolutional neural network and residual connection; and finally, fusing the mechanism model and the error correction model through parallel computing to form a dynamic hybrid model with physical interpretability and high precision. Through the hybrid modeling strategy, the prediction accuracy of key parameters such as the main steam pressure, the separator outlet steam enthalpy value and the unit load in the wide load range, especially under the dry-state operation working condition is remarkably improved, and a reliable model basis is provided for designing an advanced unit coordination control system.

Description

Technical Field

[0001] This invention relates to the fields of thermal energy engineering, power system modeling and control technology, and in particular to a data and mechanism hybrid modeling method and device for coordinated control of thermal power units. Background Technology

[0002] Building a new power system with new energy sources as the mainstay has become the core direction of my country's energy development. The large-scale grid connection of renewable energy sources such as wind and solar power, due to their inherent intermittency and volatility, poses a severe challenge to the safe and stable operation of the power grid. Against this backdrop, thermal power units, especially large-capacity, high-efficiency ultra-supercritical units, which serve as the ballast of power supply, are shifting from traditional baseload power sources to flexible peak-shaving power sources. There is an urgent need to improve their deep peak-shaving and rapid load response capabilities to mitigate the fluctuations in renewable energy and ensure grid stability.

[0003] A key prerequisite for achieving flexible and safe operation of ultracritical (supercritical) generator units is establishing a mathematical model that can accurately describe their dynamic characteristics under all operating conditions. This model is the cornerstone of advanced control strategies (such as model predictive control). Currently, modeling methods are mainly divided into two categories: mechanistic modeling and data-driven modeling. Mechanistic modeling is based on physical laws such as the conservation of mass, energy, and momentum. The model has clear physical meaning and good extrapolation properties. However, for highly complex and nonlinear objects like supercritical generator units, the model is often heavily simplified, resulting in limited accuracy under wide loads and varying operating conditions. Data-driven modeling learns system dynamics directly from operating data. It can achieve high accuracy within the training data coverage area, but the model lacks physical interpretability, its generalization ability depends on data quality and completeness, and its "black box" nature makes it difficult to meet the reliability requirements of industrial control.

[0004] To balance the physicality and accuracy of models, gray-box modeling (also known as hybrid modeling) has emerged. Its core idea is to integrate the strengths of mechanistic models and data-driven models. However, existing gray-box modeling methods still face challenges. On the one hand, if only data-driven methods are used to identify specific parameters in the mechanistic model, the improvement in accuracy is limited, and it is difficult to fully compensate for errors caused by unmodeled dynamics and mechanistic simplification. On the other hand, designing efficient data-driven modules that can not only compensate for steady-state errors but also accurately track the dynamic characteristics of the system is key to improving the performance of hybrid models. Furthermore, facing the increasingly complex dynamic characteristics of units under flexible operating conditions such as deep peak shaving, there is an urgent need for a new hybrid modeling method that can maintain the physical framework while performing high-precision dynamic error compensation through data-driven approaches. Summary of the Invention

[0005] The main objective of this invention is to provide a hybrid modeling method for coordinated control of thermal power units that combines data and mechanisms.

[0006] Another objective of this invention is to provide a data and mechanism hybrid modeling device for coordinated control of thermal power units.

[0007] The third objective of this invention is to provide a computer device.

[0008] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, a first aspect of the present invention proposes a data and mechanism hybrid modeling method for coordinated control of thermal power units, comprising: S1. Based on the lumped parameter method, a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system is established. S2, based on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections, constructs an error correction model; S3 integrates the mechanistic model and the error correction model through parallel computing to construct a dynamic hybrid model; S4 uses a test dataset that was not used for training to validate the dynamic hybrid model. By calculating the error index between the model's predicted values ​​and the actual operating values, the model's generalization ability and accuracy are evaluated, thus completing the data and mechanism hybrid modeling of the coordinated control of thermal power units.

[0010] Optionally, a hybrid modeling method combining data and mechanisms for coordinated control of thermal power units is applicable to a wide load range under dry-state operation of boiler-turbine units.

[0011] Optionally, the dynamic mechanism model is derived based on the laws of conservation of mass and energy, combined with the lumped parameter method. Its state-space equations are as follows:

[0012]

[0013] Where U represents the input to the mechanistic model, For coal feed rate, For water supply, X represents the turbine regulating valve opening; X represents the system state. This refers to the amount of coal fed into the furnace. The enthalpy of the working fluid at the economizer outlet. To achieve the pressure at the outlet of the gas-water separator, The enthalpy value at the outlet of the steam-water separator. Y represents the unit load; Y is the output of the mechanism model. Main steam pressure.

[0014] Optionally, an error correction model can be constructed based on a Long Short-Term Memory (LSTM) network, combined with a Convolutional Neural Network (CNN) and residual connections, including: S21, collect historical operating data of boiler-turbine units over a wide load range; S22, input the historical operation data into the established mechanism model, and calculate the predicted output of the mechanism model; S23, calculate the error between the predicted output of the mechanism model and the corresponding actual measured value; S24. Using the historical running data as input and the error calculated in step S23 as the target output, train the LSTM network that combines CNN and residual connections to obtain the error correction model.

[0015] Optionally, historical runtime data can be used as input, including: A time series sample set was constructed using the sliding window method, and the data was normalized. The inputs to the data model include: coal feed rate Water supply Steam turbine regulating valve opening Cooling water flow rate Primary air volume wind1, secondary air volume wind2; the output of the data model is: unit load error. Enthalpy error at the outlet of the steam-water separator Main steam pressure error .

[0016] Optionally, an LSTM network combining CNN and residual connections is configured with the following structure: An input layer receives a temporal input sequence composed of historical running data; at least two Long Short-Term Memory (LSTM) network layers, with a residual connection between the first and second LSTM layers to add the output sequences of the first and second LSTM layers element-wise to alleviate gradient vanishing and preserve shallow features; a one-dimensional convolutional neural network layer, whose input is the complete hidden state sequence output by the second LSTM layer, is used to extract local temporal dynamic patterns from high-order features; a batch normalization layer and a ReLU activation function layer are sequentially connected after the one-dimensional convolutional neural network layer to stabilize the training process and enhance nonlinear expressive power; and an output layer, a fully connected layer, is used to map the processed features to the final error prediction value.

[0017] Optionally, the dynamic mixture model can be validated using a test dataset not used in training. The model's generalization ability and accuracy can be evaluated by calculating the error metric between the model's predicted values ​​and the actual running values, including: The absolute mean percentage error (MAPE) is used as the main evaluation index to evaluate the model; the smaller the error, the higher the accuracy of the mixed model.

[0018] To achieve the above objectives, a second aspect of the present invention provides a data and mechanism hybrid modeling device for coordinated control of thermal power units, comprising: A mechanism model module is constructed to establish a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system based on the lumped parameter method. The error correction module is used to construct an error correction model based on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections. The model fusion module is used to fuse the mechanistic model and the error correction model through parallel computing to construct a dynamic hybrid model; The validation module is used to validate the dynamic hybrid model using a test dataset that was not used for training. By calculating the error index between the model's predicted values ​​and the actual operating values, it evaluates the model's generalization ability and accuracy, and completes the data and mechanism hybrid modeling of the coordinated control of thermal power units.

[0019] To achieve the above objectives, a third aspect of this application provides a computer device comprising a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading executable program code stored in the memory, for implementing the data and mechanism hybrid modeling method for coordinated control of thermal power units as described in the first aspect embodiment.

[0020] To achieve the above objectives, the fourth aspect of this application proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the data and mechanism hybrid modeling method for coordinated control of thermal power units as described in the first aspect embodiment.

[0021] The embodiments of the present invention have the following beneficial effects: This invention provides a data and mechanism hybrid modeling method for coordinated control of thermal power units. By combining the interpretability of the mechanism model with the high precision of the data-driven model, it overcomes the shortcomings of traditional modeling methods, thereby improving modeling accuracy and ensuring model interpretability. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating a data and mechanism hybrid modeling method for coordinated control of thermal power units, provided in an embodiment of the present invention; Figure 2 This is a simplified diagram of the boiler steam-water flow path according to the present invention; Figure 3 This is a diagram of the neural network structure of the data compensation model proposed in this invention; Figure 4 This is a structural diagram of a data and mechanism hybrid modeling device for coordinated control of thermal power units provided in an embodiment of the present invention. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0025] The following describes, with reference to the accompanying drawings, a method and apparatus for hybrid modeling of data and mechanisms for coordinated control of thermal power units according to embodiments of the present invention.

[0026] Example 1 This embodiment provides a hybrid data and mechanism modeling method for coordinated control of thermal power units, which includes the following steps: S1. Based on the lumped parameter method, a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system is established.

[0027] In order to identify the parameters of the mechanism model, this invention establishes a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system, and realizes it using operating data based on the lumped parameter method.

[0028] The mechanistic modeling of the coordination system described in step S1 can be divided into three key subsystems: the pulverizing system, the boiler system, and the turbine system. The following core assumptions are adopted to simplify the modeling process: The dynamics of the fuel delivery and pulverizing system are simplified to a first-order inertial link; the working fluid flow is assumed to meet quasi-steady-state conditions, and the mass flow rate in each control volume remains constant, ignoring complex dynamic effects caused by two-phase flow; it is assumed that there is no time delay in the combustion process and that the heat released by combustion is uniformly transferred to the working fluid, without distinguishing between heat transfer modes such as radiation and convection; the working fluid is treated as a one-dimensional flow in the heated surface and control volume, ignoring the temperature and pressure gradients in the radial direction; due to the excellent thermal conductivity of metal materials, their internal temperature gradient and thermal resistance are ignored, and it is assumed that the rate of temperature change of the metal is consistent with the temperature change of the contact working fluid.

[0029] S1.1, Pulverizing System Modeling: The pulverizing process mainly includes coal feeding, pulverizing, conveying, and classification. Among these, coal feeding regulation involves a transmission delay, represented by a pure delay element; the crushing and drying dynamics of the pulverizing process are described by a first-order inertial element; subsequent pneumatic conveying and classification also introduce delays, considered as additional pure delay elements. Finally, qualified pulverized coal is fed into the burner to ensure stable combustion. Therefore, the dynamic characteristics of the pulverizing system can be simplified to a series model of "first-order inertia + pure delay," which can be described by the following equation: (1) In the formula, It is the fuel quantity directive (kg / s). This is the actual amount of coal entering the boiler (kg / s). The delay time is in seconds. It is the inertial time (s) of the powder-making system.

[0030] S1.2 Boiler System Modeling: The boiler system adopts a modular modeling strategy, mainly divided into three major components: desuperheater, water-cooled wall-superheater combination module, and economizer.

[0031] The boiler steam-water system process described in this invention is as follows: feedwater is preheated by the economizer, then heated into saturated steam by the water-cooled walls, and then flows through multiple superheaters to form superheated steam, which is finally sent to the steam turbine to perform work. To control the main steam temperature, the system is equipped with desuperheating water taken from the economizer outlet and injected between each stage of the superheater.

[0032] To simplify calculations, this invention establishes a simplified calculation model for the boiler steam-water system. Its core lies in integrating the water-cooled wall, steam-water separator, and multi-stage superheater into a single heating module; simultaneously, to eliminate the disturbance to the model caused by deviations in the desuperheating water flow rate measurement, the original multi-point desuperheating path is equivalent to a single spray path located at the outlet of the high-temperature superheater, such as... Figure 2 As shown, this equivalent treatment conforms to the principle of energy conservation and does not change the dynamic characteristics of the system outlet steam.

[0033] Desuperheater modeling: Desuperheating water is directly sprayed into superheated steam to reduce the steam temperature. This process changes rapidly, and its internal dynamic process can be ignored. The mass and energy balance equations for the water spray desuperheating process can be expressed as: (2) (3) Solving equations (2) and (3) simultaneously yields: (4) In the formula, The steam flow rate at the outlet of the high-temperature superheater is (kg / s). Main steam flow rate (kg / s); To reduce the total flow rate of the heated water (kg / s); , , These are the steam enthalpy values ​​(kJ / kg) at the economizer, steam-water separator outlet, and high-temperature superheater outlet, respectively. Main steam enthalpy (kJ / kg); l It is a static function representing the ratio of the steam enthalpy at the outlet of the high-temperature superheater to the steam enthalpy at the outlet of the steam-water separator, and can be written as a linear function of the amount of coal fed into the furnace.

[0034] S1.2.2, Water-cooled wall-superheater modeling: During the operation of a supercritical once-through boiler, the steam pressure and enthalpy at the outlet of the steam-water separator can keenly reflect changes in the water-coal ratio and respond quickly to load fluctuations. Therefore, the working fluid parameters at the outlet of the steam-water separator are selected as lumped parameters.

[0035] The mass and energy equations for the water-cooled wall and superheater system are as follows: (5) (6) In equation (6), the second term on the left side of the equal sign represents the energy change of the heated metal tube. This indicates the volume of the heated section of the pipe wall; This indicates the mass of the pipe wall in the heated section; It is the flow rate of the working fluid entering the water-cooled wall (kg / s). It is the enthalpy value of the working fluid at the economizer outlet; , Q1 is the water vapor density (kg / m3) and internal energy (kJ / kg) at the outlet of the steam-water separator; Q2 is the total heat absorption (kJ / s) in the water-cooled wall and superheater system. It can be written as the following equation: (7) in, It is a linear function of the amount of coal fed into the furnace. This refers to the amount of coal fed into the furnace (kg / s). The water supply flow rate is (kg / s). Unit load (MW) and Both are constants. It is a function of the amount of coal fed into the furnace, and is obtained through steady-state operating conditions.

[0036] Equations (5) and (6) can be rewritten as: (8) (9) Equations (8) and (9) can be further rewritten as: (10) (11) In the formula, , , , .

[0037] Substituting the desuperheating water model (4) into (10) and (11) yields: (12) (13) In the formula, , , , .

[0038] therefore, , , , It can be obtained from the unit's own parameters and thermodynamic parameters.

[0039] The superheater differential pressure, which is the difference between the steam pressure at the outlet of the steam-water separator and the main steam pressure, can be considered as a function of the steam pressure at the outlet of the steam-water separator as the independent variable: (14) in, It is a linear function.

[0040] S1.2.3 Economizer Modeling: Since the inlet and outlet working fluid flow rates of the economizer vary relatively little, and the outlet flow rate is difficult to measure directly, its flow dynamics can be ignored during modeling, and the outlet flow rate can be approximated as equal to the main feedwater flow rate. Therefore, the economizer subsystem no longer relies on the mass conservation equation; its dynamic process is mainly dominated by the energy conservation equation. (15)

[0041] akin, .

[0042] S1.3, Steam Turbine System: The main steam flow rate is determined by both the opening of the steam turbine regulating valve and the main steam pressure, and their relationship is described by the following function: (16) The model is based on a key simplification: since the main steam temperature is strictly stabilized near the set value (about 600°C) through closed-loop control, the specific enthalpy of the main steam can be approximated as a single-valued function of the pressure. This simplification effectively reduces the complexity of the model.

[0043] The dynamic characteristics of the turbine's output power are described by a first-order differential equation, specifically: (17) in, It is a static function related to the load. This represents the inertial time constant of the steam turbine.

[0044] S1.4 In summary, the vector state-space equations of a supercritical once-through boiler unit can be summarized as follows:

[0045]

[0046] Where U represents system input, X represents system state, and Y represents system output.

[0047] The state-space equations can be described in detail as follows: (18) S1.5, Parameter Identification: The parameters that the above model needs to identify are divided into four categories: 1) Four static functions , , 1) These functions are identified based on the system's steady-state data; 2) Five dynamic parameters , , , , These parameters are determined by the inherent physical structural characteristics of the unit and the thermophysical properties of water and steam; 3) Four dynamic functions , , , These parameters can be directly calculated from the running data; 4) Three fixed parameters , , These parameters are optimized through dynamic running data.

[0048] S1.5.1, the static function is identified through the unit's steady-state operating data. Under steady-state conditions, the unit's state parameters can be considered constant values, therefore equation (18) can be rewritten as: (19) superscript This indicates that the state variable is in a steady state.

[0049] S1.5.2, the dynamic parameters are calculated from the physical structural characteristics of the unit and the thermophysical properties of water and steam.

[0050] S1.5.3, Dynamic Function Identification: Feedwater Enthalpy The feedwater enthalpy is positively correlated with the boiler operating level; therefore, it can be expressed as a single-valued function of the unit load. The main steam pressure can be written as a single-valued function of the steam pressure at the steam-water separator outlet. As mentioned earlier, ,and then The main steam flow rate can be written as a function of the valve opening and the steam pressure at the separator outlet.

[0051] S1.5.4, Fixed Parameter Identification: Fixed Parameters , , Based on optimization derived from unit operation data, this invention employs a particle swarm optimization algorithm.

[0052] S2 is an error correction model built on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections.

[0053] To construct the error correction model, this application embodiment is based on a Long Short-Term Memory (LSTM) network combined with a Convolutional Neural Network (CNN) and residual connections.

[0054] The model described in step S2 is an error compensation model that integrates multimodal deep learning techniques. The model adopts a three-stage processing flow of "temporal feature extraction - local feature enhancement - global feature fusion".

[0055] First, the input data is normalized to eliminate the influence of units and improve training stability. Then, a training set is constructed using a sliding window.

[0056] After being ground by the pulverizing system, raw coal is conveyed into the furnace by primary air, where it undergoes a series of unsteady combustion processes, including volatile matter release, ignition, and coke burnout. This process itself involves a certain chemical reaction time scale. More importantly, the enormous heat released by combustion must be transferred to the working fluid through metal heating surfaces with high heat capacity. The significant heat storage effect in this heat transfer path constitutes the enormous thermal inertia of the boiler system.

[0057] Based on the above analysis, this invention first introduces a temporal modeling module for Long Short-Term Memory (LSTM) networks. This module is implemented through a finely designed gating system, and the memory update mechanism includes three key control gates: Forget gate: controls the degree to which historical information is retained, and can be expressed by the following formula: (20) in, It is the output of the forget gate. It is a weight matrix. It is a bias term. For activation function, It is the hidden state from the previous moment. This is the current input.

[0058] Input gate: Adjusts the proportion of new information written, which can be expressed by the following formula: (twenty one) (twenty two) (twenty three) in, For input gate output, and It refers to the current and past memory states.

[0059] The past memory state is multiplied by the forget gate to discard the information that needs to be forgotten. The output gate then determines the external representation of the final state. The output gate equation is: (twenty four) (25) in, This is the output gate control signal. It is the hidden state at the current moment.

[0060] Some processes within the boiler also have relatively fast responses; for example, a tiny movement in the turbine regulating valve opening can result in a corresponding change in unit load within seconds. Therefore, this invention introduces a one-dimensional convolutional neural network (CNN) specifically designed to capture local dynamic features in time-series data. Its mathematical expression is: (26) in, Represents the output features of convolution. Represents the convolution kernel weight parameters. For bias terms, The kernel size is [size]. This module uses local window data of the input sequence as input, and then connects a normalization layer and an activation function to effectively enhance the nonlinear expressive power.

[0061] Establish a shortcut path between the first two LSTM layers. The residual calculation expression is: (27) During gradient backpropagation, the residual structure provides an identity path, allowing the gradient to be directly propagated back to the shallow layer without at least decay, thus alleviating the gradient vanishing problem.

[0062] After building the data-driven error compensation model, the hyperparameters of the model are optimized using the particle swarm optimization algorithm.

[0063] In this embodiment, an error compensation model that integrates multimodal deep learning technology is constructed, laying the foundation for the subsequent construction of a dynamic hybrid model.

[0064] S3 integrates the mechanistic model and the error correction model through parallel computing to construct a dynamic hybrid model.

[0065] In this embodiment of the application, this step aims to achieve deep integration of the mechanism model and the error correction model through parallel computing, thereby constructing a dynamic hybrid model with high reliability and high accuracy.

[0066] In this embodiment, the mechanistic model serves as the core foundation of the dynamic hybrid model, enabling it to provide fundamental predictions with clear physical meaning based on well-defined principles of mass and energy conservation and valve flow characteristics.

[0067] This dynamic hybrid model adopts an innovative parallel fusion architecture. The core design idea is to fully combine the physical interpretability of the mechanistic model with the dynamic error compensation capability of the data-driven error correction model. The error correction model serves as a supplement and optimization module for the mechanistic model, possessing online learning capabilities. It can capture in real time the dynamic residuals of the mechanistic model caused by changes in operating conditions, parameter fluctuations, or unconsidered complex factors in practical applications. Through continuous learning and adaptive adjustment, it can accurately compensate for the prediction results of the mechanistic model, effectively reducing prediction errors.

[0068] In this embodiment, the key technical design is to use parallel computing for model fusion. This method allows the basic prediction of the mechanism model and the residual compensation of the error correction model to be performed simultaneously, which not only ensures the computational efficiency of the model, but also ensures the real-time fusion of the output results of the two, avoiding the delay problem that may be caused by serial computing.

[0069] In the embodiments of this application, the design and optimization of the advanced controller needs to be based on an accurate and reliable object model. This dynamic hybrid model can provide the controller with real-time and accurate state feedback and loss prediction, helping the controller to adjust the control strategy more accurately.

[0070] In this embodiment of the application, by constructing a dynamic hybrid model, the respective advantages of the mechanistic model and the data-driven model are fully utilized, laying the foundation for subsequent evaluation of the model's generalization ability and accuracy.

[0071] S4 uses a test dataset that was not used for training to validate the dynamic hybrid model. By calculating the error index between the model's predicted values ​​and the actual running values, the model's generalization ability and accuracy are evaluated.

[0072] In this application, the accuracy of the fusion model is validated using a dataset not used for training, and the dimensionless mean absolute percentage error (MAPE) is used as the main evaluation metric. (28) in, These are measured values. These are the predicted values ​​from the mixture model.

[0073] In addition, to intuitively judge the magnitude of the error, the mean absolute error (MAE) is used as an auxiliary evaluation index.

[0074] (29).

[0075] In summary, the embodiments of this application significantly improve prediction accuracy over a wide load range through a hybrid modeling strategy, providing a reliable model foundation for designing advanced unit coordinated control systems.

[0076] Based on the principles of mass conservation and energy conservation, the nonlinear model of the unit coordination system is as follows: (30) According to the steady-state equation: (31) The static function identification results are as follows: (32) The dynamic parameters are calculated from the structural characteristics of the unit and the thermophysical properties of water and steam.

[0077] The thermal properties of water can be expressed by IAPWS-IF97, which can be simplified to... (33) Therefore, the identification results of the dynamic parameters are as follows: (34) 200,000 seconds of operating data were selected from a 1000MW unit of a power plant, with 80% used as the training set and 20% as the test set. The data sampling interval was TS=8s, therefore, the training set had 20,000 samples and the test set had 5,000 samples.

[0078] Dynamic function identification using training set running data: (35) Using MAPE as the primary evaluation metric, the dynamic parameters identified from the training set data are as follows: = 138, = 21, = 47.

[0079] The model error of the training set is obtained by subtracting the actual measurement values ​​from the output values ​​of the mechanistic model, and then a data-driven error compensation model is established.

[0080] After normalizing the input, such as Figure 3 As shown, the input temporal features are processed by two LSTM layers to extract dynamic information. The first two LSTM layers use the sequence output mode to generate the hidden state sequence step by step, ensuring the complete preservation of the temporal features. The last LSTM layer uses the last output mode to compress the entire sequence into a global feature vector, which is convenient for subsequent prediction tasks.

[0081] Residual connections are introduced between the first two LSTM layers. For example... Figure 3 As shown, the outputs of LSTM1 and LSTM2 are added element-wise through the addition operation. Building upon LSTM feature extraction, the model further incorporates a one-dimensional convolutional neural network (1DCNN) to enhance the perception of local dynamic patterns.

[0082] After the convolution operation, batch normalization (BN) and ReLU activation functions are applied. The former stabilizes the training process by standardizing the data distribution, while the latter increases the non-linear expressiveness of the model and alleviates the gradient vanishing problem.

[0083] The network uses a third-layer LSTM to perform higher-level temporal integration of the convolutional features, and then combines Dropout random deactivation technology to prevent model overfitting.

[0084] The Adam optimizer is used during model training.

[0085] This invention uses the particle swarm optimization algorithm to optimize the hyperparameters of the data compensation model.

[0086] The output of the mechanistic model and the output of the error compensation model are added together to obtain the output of the hybrid model.

[0087] Test set validation shows that the hybrid model has good accuracy and can be used to optimize the unit's operating strategy over a wide load range, thereby improving its variable load performance.

[0088] Example 2 This invention also provides a data and mechanism hybrid modeling device for coordinated control of thermal power units, such as... Figure 4 As shown, the device includes: The mechanism model module 100 is used to establish a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system based on the lumped parameter method. Error correction module 200 is used to construct an error correction model based on Long Short-Term Memory Network (LSTM), combined with Convolutional Neural Network (CNN) and residual connections. The model fusion module 300 is used to fuse the mechanism model and the error correction model through parallel computing to construct a dynamic hybrid model; The verification module 400 is used to verify the dynamic hybrid model using a test dataset that was not used for training. By calculating the error index between the model's predicted values ​​and the actual operating values, the generalization ability and accuracy of the model are evaluated, and the data and mechanism hybrid modeling of the coordinated control of thermal power units is completed.

[0089] Example 3 To implement the methods of the above embodiments, the present invention also provides a computer device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0090] Example 4 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0092] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0093] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A data-mechanism hybrid modeling method for coordinated control of thermal power units, characterized in that, include: S1. Based on the lumped parameter method, a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system is established. S2, based on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections, constructs an error correction model; S3 integrates the mechanistic model and the error correction model through parallel computing to construct a dynamic hybrid model; S4 uses a test dataset that was not used for training to validate the dynamic hybrid model. By calculating the error index between the model's predicted values ​​and the actual operating values, the model's generalization ability and accuracy are evaluated, thus completing the data and mechanism hybrid modeling of the coordinated control of thermal power units.

2. The method as described in claim 1, characterized in that, The aforementioned data and mechanism hybrid modeling method for coordinated control of thermal power units is applicable to a wide load range under dry-state operation conditions of boiler-turbine units.

3. The method as described in claim 2, characterized in that, The dynamic mechanism model is derived based on the laws of mass conservation and energy conservation, combined with the lumped parameter method. Its state-space equations are as follows: Where U represents the input to the mechanistic model, For coal feed rate, For water supply, X represents the turbine regulating valve opening; X represents the system state. This refers to the amount of coal fed into the furnace. The enthalpy of the working fluid at the economizer outlet. To achieve the pressure at the outlet of the gas-water separator, The enthalpy value at the outlet of the steam-water separator. Y represents the unit load; Y is the output of the mechanism model. Main steam pressure.

4. The method as described in claim 3, characterized in that, The error correction model constructed based on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections, also includes: S21, collect historical operating data of boiler-turbine units over a wide load range; S22, input the historical operation data into the established mechanism model, and calculate the predicted output of the mechanism model; S23, calculate the error between the predicted output of the mechanism model and the corresponding actual measured value; S24. Using the historical running data as input and the error calculated in step S23 as the target output, train the LSTM network that combines CNN and residual connections to obtain the error correction model.

5. The method as described in claim 4, characterized in that, The method of using historical operational data as input also includes: A time series sample set was constructed using the sliding window method, and the data was normalized. The inputs to the data model include: coal feed rate Water supply Steam turbine regulating valve opening Cooling water flow rate Primary air volume wind1, secondary air volume wind2; the output of the data model is: unit load error. Enthalpy error at the outlet of the steam-water separator Main steam pressure error .

6. The method as described in claim 4, characterized in that, The LSTM network that combines CNN and residual connections is specifically configured as follows: An input layer receives a temporal input sequence composed of historical running data; at least two Long Short-Term Memory (LSTM) network layers, with a residual connection between the first and second LSTM layers to add the output sequences of the first and second LSTM layers element-wise to alleviate gradient vanishing and preserve shallow features; a one-dimensional convolutional neural network layer, whose input is the complete hidden state sequence output by the second LSTM layer, is used to extract local temporal dynamic patterns from high-order features; a batch normalization layer and a ReLU activation function layer are sequentially connected after the one-dimensional convolutional neural network layer to stabilize the training process and enhance nonlinear expressive power; and an output layer, a fully connected layer, is used to map the processed features to the final error prediction value.

7. The method as described in claim 1, characterized in that, The method of validating the dynamic hybrid model using a test dataset not used in training, and evaluating the model's generalization ability and accuracy by calculating the error index between the model's predicted values ​​and the actual running values, also includes: The absolute mean percentage error (MAPE) is used as the main evaluation index to evaluate the model; the smaller the error, the higher the accuracy of the mixed model.

8. A data and mechanism hybrid modeling device for coordinated control of thermal power units, characterized in that, include: A mechanism model module is constructed to establish a high-precision dynamic mechanism model of the working fluid flow and heat transfer process in the boiler-turbine system based on the lumped parameter method. The error correction module is used to construct an error correction model based on the Long Short-Term Memory (LSTM) network, combined with the Convolutional Neural Network (CNN) and residual connections. The model fusion module is used to fuse the mechanistic model and the error correction model through parallel computing to construct a dynamic hybrid model; The validation module is used to validate the dynamic hybrid model using a test dataset that was not used for training. By calculating the error index between the model's predicted values ​​and the actual operating values, it evaluates the model's generalization ability and accuracy, and completes the data and mechanism hybrid modeling of the coordinated control of thermal power units.

9. A computer device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the data and mechanism hybrid modeling method for coordinated control of thermal power units as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the data and mechanism hybrid modeling method for coordinated control of thermal power units as described in any one of claims 1-7.

Images