Coal-fired power unit wet state control method based on physical information neural network predictive control, storage medium and equipment

Abstract

The invention discloses a coal-fired power generation unit wet state control method based on physical information neural network prediction control, a storage medium and equipment, and the method comprises the steps: building a coal-fired power generation unit wet state control prediction model based on a physical information neural network based on a mechanism model of a coal-fired power generation unit; acquiring real-time state data of the coal-fired power generation unit, wherein the state data comprises wet state operation data and wet state control data; according to the difference between the wet state control data in the state data and the set value of the wet state control data, the wet state operation data in the state data are dynamically adjusted, the adjusted wet state operation data are input into the wet state control prediction model of the coal-fired power generation unit, and the wet state control data at the next moment are predicted. And updating the wet state control data in the state data to ensure that the predicted wet state control data of the coal-fired power generation unit not only conforms to a physical rule, but also can adapt to the dynamic change of actual operation data, so that the precision and robustness of the wet state control of the coal-fired power generation unit are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal automatic control, and in particular relates to a wet state control method, storage medium and equipment for a coal-fired power generation unit based on physical information neural network predictive control. Background Art

[0002] As the proportion of renewable energy in the power grid continues to rise, grid load characteristics are becoming increasingly complex. The intermittent and unpredictable nature of renewable energy significantly increases the peak-to-valley variation in grid load, requiring generators to possess stronger regulation capabilities. Deep peak shaving involves reducing generator output to a lower level of rated capacity during periods of low grid load to meet fluctuating grid load demands. Deep peak shaving allows for better integration of renewable energy, reduces reliance on traditional fossil fuels, thereby lowering carbon emissions and promoting an optimized energy mix.

[0003] During deep peak-shaving operation, the combustion system of coal-fired power generation units will change. When the unit load drops to a certain level, the combustion stability of the burner will decrease, and even incomplete combustion may occur. To ensure combustion stability, coal-fired power generation units need to switch to wet operation mode. In dry operation, the water level in the water storage tank remains unchanged. In wet operation, the working fluid in the steam-water separator consists of steam and water. Steam enters the superheater and water flows into the water storage tank. The water level in the water storage tank has also become an important safety indicator. As a result, wet control of coal-fired power generation units faces difficulties such as large load fluctuations, difficult water flow regulation, complex steam temperature control, fluctuating water level in the steam-water separator, and poor combustion stability.

[0004] Therefore, to improve the stability and economic efficiency of coal-fired power generation units, it is necessary to establish models that more accurately characterize system characteristics, design control algorithms that are more adaptable to operating mode switching, optimize operating parameters, and promote the safe operation of coal-fired power generation units in the power grid. Existing modeling methods mainly include mechanism modeling and data-driven modeling. Mechanism modeling is more difficult, with complex model calculations and difficulty in real-time optimization. Data-driven modeling uses black box models, which have low reliability.

[0005] To improve the stability, accuracy, and speed of wet-state control for coal-fired power generation units, model predictive control (MPC) can be employed. MPC is a model-based control strategy whose core concept is to estimate the future behavior of the system through a predictive model and calculate the optimal control input based on an optimization algorithm. It is suitable for multivariable, strongly coupled, nonlinear, and complexly constrained systems. However, traditional MPC prediction models typically employ transfer functions or state-space equations, which are difficult to fully characterize the complex physical characteristics of coal-fired power generation units. This is especially true under wet-state operating conditions, where dynamic characteristics vary dramatically. Traditional MPC prediction accuracy is insufficient, resulting in reduced control performance. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a wet state control method, storage medium and equipment for coal-fired power generation units based on physical information neural network predictive control. The combustion kinetics equation, mass conservation equation and energy conservation equation are embedded in the physical information neural network to ensure that the predicted wet state control data of the coal-fired power generation unit conforms to the physical laws and can adapt to the dynamic changes of actual operating data, thereby greatly improving the accuracy and robustness of the wet state control of the coal-fired power generation unit.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solution: a wet state control method for a coal-fired power generation unit based on physical information neural network predictive control, comprising the following steps: Step 1: Based on the mechanism model of coal-fired power generation units, a prediction model for wet state control of coal-fired power generation units based on physical information neural network is established; Step 2: Acquire real-time status data of the coal-fired power generation unit, wherein the status data includes wet-state operation data and wet-state control data; Step 3: Based on the difference between the wet state control data in the state data and the set value of the wet state control data, the wet state operation data in the state data is dynamically adjusted, the adjusted wet state operation data is input into the wet state control prediction model of the coal-fired power generation unit, the wet state control data at the next moment is predicted, and the wet state control data in the state data is updated; Step 4: Repeat step 3 to achieve wet state control of the coal-fired power generation unit until it shuts down.

[0008] Furthermore, the wet operation data include: turbine main steam valve opening, coal consumption, feed water flow and recirculation water flow; the wet control data include: main steam pressure, output power, main steam temperature and water tank water level.

[0009] Furthermore, the mechanism model of the coal-fired power generation unit includes:

[0010] in, is the amount of coal burned, is the amount of coal entering the coal mill, is the dynamic coefficient of coal burning, is the function fitted for coal mill; is the water flow rate, is the main steam temperature, is the function fitted to the boiler system, is the main steam temperature dynamic coefficient; is the opening of the main steam valve of the steam turbine, is the main steam pressure, is the main steam pressure dynamic coefficient, is the water supply dynamic coefficient, Function fitted to the boiler system; is the output power, is the power dynamic coefficient, Function fitted for the steam turbine system; is the recirculating water flow rate, is the water level in the water tank, is the dynamic coefficient of water level in the water tank, Function fitted to the water storage tank system.

[0011] Furthermore, the objective function of the wet state control prediction model of the coal power generation unit is for:

[0012] in, N represents the time increment, i express N The index of Indicates the current moment, represents the predicted wet control data, Indicates the set wet state control data, represents the two-norm, Indicates the increment of wet operation data.

[0013] Furthermore, the constraints of the wet state control prediction model of the coal-fired power generation unit include:

[0014] in, Indicates wet running data, Indicates the minimum value of wet operation data, Indicates the maximum value of wet operation data, Indicates the minimum value of the wet operation data increment, Indicates the maximum value of the wet operation data increment, Indicates the minimum value of the predicted wet control data, Indicates the maximum value of the predicted wet control data.

[0015] Furthermore, it is necessary to train the wet state control prediction model of coal-fired power generation units: i. Collect historical state data of coal-fired power generation units in wet operation mode, use the wet operation data in the historical state data as input to the wet control prediction model of the coal-fired power generation units, and use the corresponding wet control data in the historical state data as labels; ii. Input the collected wet-state operation data into the wet-state control prediction model of the coal-fired power generation unit, and predict the wet-state control data with the goal of minimizing the objective function while satisfying the constraints; iii. Calculate the loss function including data error term and physical information error term based on the predicted wet control data and corresponding labels, and use the gradient descent method to update the wet control prediction model of the coal-fired power generation unit; iv. Repeat steps ii-iii until the loss function converges, completing the training of the wet state control prediction model for the coal-fired power generation unit.

[0016] Furthermore, the loss function for:

[0017] in, represents the sample size of the collected historical state dataset, j express The index of Indicates that through j Wet control data for sample prediction, Indicates the j wet control data in samples; represents the sample size involved in the construction of the mechanism model of coal-fired power generation units, l express The index of Indicates the l The wet control data calculated by the mechanism model of coal-fired power generation units, Indicates that through l Wet control data for sample prediction, Represents the weight coefficient.

[0018] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the wet state control method of a coal-fired power generation unit based on physical information neural network predictive control.

[0019] Furthermore, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the wet state control method of the coal-fired power generation unit based on physical information neural network predictive control is implemented.

[0020] Compared with the prior art, the present invention has the following beneficial effects: the wet state control method of the coal-fired power generation unit based on the physical information neural network predictive control of the present invention is based on the mechanism model of the coal-fired power generation unit, establishes a wet state control prediction model of the coal-fired power generation unit based on the physical information neural network, integrates the physical laws of the mass conservation equation, the combustion kinetics equation and the energy conservation equation with the real-time wet state operation data, generates high-precision wet state control data, and can significantly reduce the solution dimension of the nonlinear optimization problem; at the same time, physical constraints are introduced in the optimization of the objective function of the wet state control prediction model of the coal-fired power generation unit, and the physical constraints are used to dynamically eliminate control decisions that violate the physical laws, avoid the conservatism or mismatch problems caused by the fixed constraint boundaries in the traditional MPC, narrow the search space, and combine the highly parallel computing power of the physical information neural network to significantly improve the solution efficiency. The wet state control method of the coal-fired power generation unit of the present invention ensures that the predicted wet state control data of the coal-fired power generation unit not only conforms to the physical laws, but also can adapt to the dynamic changes of the actual operation data, greatly improving the accuracy and robustness of the wet state control of the coal-fired power generation unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be further explained below with reference to the accompanying drawings.

[0023] like Figure 1 Schematic diagram of the wet state control method of a coal-fired power generation unit based on physical information neural network predictive control of the present invention. The wet state control method of a coal-fired power generation unit includes the following steps: Step 1: Based on the mechanism model of coal-fired power generation units, a wet state control prediction model of coal-fired power generation units based on physical information neural network is established.

[0024] The mechanism model of the coal-fired power generation unit in the present invention is as follows. Simplified mechanism models of the coal mill, boiler system, water supply system, steam turbine, and water storage tank are constructed respectively, and the relationship between the main parameters is established:

[0025] in, is the amount of coal burned, is the amount of coal entering the coal mill, is the dynamic coefficient of coal burning, is the function fitted for coal mill; is the water flow rate, is the main steam temperature, is the function fitted to the boiler system, is the main steam temperature dynamic coefficient; is the opening of the main steam valve of the steam turbine, is the main steam pressure, is the main steam pressure dynamic coefficient, is the water supply dynamic coefficient, Function fitted to the boiler system; is the output power, is the power dynamic coefficient, Function fitted for the steam turbine system; is the recirculating water flow rate, is the water level in the water tank, is the dynamic coefficient of water level in the water tank, Function fitted to the water storage tank system.

[0026] Step 2: Obtain real-time status data of the coal-fired power generation unit, which includes wet operation data and wet control data. The wet operation data includes: turbine main steam valve opening, coal consumption, feed water flow rate, and recirculation water flow rate; the wet control data includes: main steam pressure, output power, main steam temperature, and water tank water level.

[0027] Step 3: According to the difference between the wet control data in the status data and the set value of the wet control data, the wet operation data in the status data is dynamically adjusted, and the adjusted wet operation data is input into the wet control prediction model of the coal-fired power generation unit to predict the wet control data at the next moment, and the wet control data in the status data is updated. The wet control process of the coal-fired power generation unit is analyzed through the wet control prediction model of the coal-fired power generation unit, and the future operating conditions of the coal-fired power generation unit can be predicted. According to the predicted change trend, control measures can be taken in advance to adapt to the complex mechanism of the coal-fired power generation unit, and the wet operation data of the coal-fired power generation unit can be fully utilized to make decisions, which has strong flexibility and adaptability.

[0028] Objective function of the wet state control prediction model of the coal-fired power generation unit of the present invention for:

[0029] in, N represents the time increment, i express N The index of Indicates the current moment, represents the predicted wet control data, Indicates the set wet state control data, represents the two-norm, Indicates the increment of wet operation data.

[0030] The constraints of the wet state control prediction model for coal-fired power generation units include:

[0031] in, Indicates wet running data, Indicates the minimum value of wet operation data, Indicates the maximum value of wet operation data, Indicates the minimum value of the wet operation data increment, Indicates the maximum value of the wet operation data increment, Indicates the minimum value of the predicted wet control data, Indicates the maximum value of the predicted wet control data.

[0032] Step 4: Repeat step 3 to achieve wet state control of the coal-fired generator set until it shuts down. The wet state control method of the coal-fired generator set of the present invention ensures that the predicted wet state control data of the coal-fired generator set not only conforms to physical laws, but also can adapt to the dynamic changes of actual operating data, thereby greatly improving the accuracy and robustness of the wet state control of the coal-fired generator set.

[0033] In one technical solution of the present invention, it is necessary to train a wet state control prediction model for a coal-fired power generation unit: i. Collect historical status data of coal-fired power generation units in wet operation mode, use the wet operation data in the historical status data as the input of the wet control prediction model of the coal-fired power generation units, and use the corresponding wet control data in the historical status data as labels.

[0034] ii. The collected wet operating data is input into the wet control prediction model of the coal-fired power generation unit. Under the premise of satisfying the constraints, the wet control data is predicted with the goal of minimizing the objective function. Physical constraints are used to dynamically eliminate control decisions that violate physical laws, avoiding the conservatism or mismatch problems caused by fixed constraint boundaries in traditional MPC, narrowing the search space, and combining the highly parallel computing power of physical information neural networks to significantly improve the solution efficiency.

[0035] iii. Based on the predicted wet-state control data and corresponding labels, a loss function consisting of a data error term and a physical information error term is calculated. The data error term measures the difference between the predicted output of the wet-state control prediction model for coal-fired power plants and the actual data, using the mean square error as the data error term. The physical information error term is calculated by substituting the physical quantities predicted by the wet-state control prediction model for the coal-fired power plant into the mechanism model of the coal-fired power plant and calculating the residual. The coal-fired power plant wet-state control prediction model is then updated using a gradient descent method. The physical laws of the mass conservation equation, combustion kinetics equation, and energy conservation equation are integrated with real-time wet-state operation data to generate high-precision wet-state control data, significantly reducing the dimensionality of the nonlinear optimization problem.

[0036] The loss function in this invention for:

[0037] in, represents the sample size of the collected historical state dataset, j express The index of Indicates that through j Wet control data for sample prediction, Indicates the j wet control data in samples; represents the sample size involved in the construction of the mechanism model of coal-fired power generation units, l express The index of Indicates the l The wet control data calculated by the mechanism model of coal-fired power generation units, Indicates that through l Wet control data for sample prediction, Represents the weight coefficient.

[0038] iv. Repeat steps ii-iii until the loss function converges, completing the training of the wet state control prediction model for the coal-fired power generation unit.

[0039] In one technical solution of the present invention, a computer-readable storage medium is also provided, which stores a computer program, and the computer program enables a computer to execute a wet state control method for a coal-fired power generation unit based on physical information neural network predictive control.

[0040] In one technical solution of the present invention, an electronic device is also provided, comprising: a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, a wet state control method for a coal-fired power generation unit based on physical information neural network predictive control is implemented.

[0041] In the embodiments disclosed herein, computer storage media may be tangible media that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. Computer storage media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media may include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0042] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0043] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A wet state control method for coal-fired power generation units based on physical information neural network predictive control, characterized in that: The steps include: Step 1: Based on the mechanism model of coal-fired power generation units, a prediction model for wet state control of coal-fired power generation units based on physical information neural network is established; Step 2: Acquire real-time status data of the coal-fired power generation unit, wherein the status data includes wet-state operation data and wet-state control data; Step 3: Based on the difference between the wet state control data in the state data and the set value of the wet state control data, the wet state operation data in the state data is dynamically adjusted, the adjusted wet state operation data is input into the wet state control prediction model of the coal-fired power generation unit, the wet state control data at the next moment is predicted, and the wet state control data in the state data is updated; Step 4: Repeat step 3 to achieve wet state control of the coal-fired power generation unit until it shuts down.

2. The wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to claim 1 is characterized in that: The wet operation data include: the opening of the main steam valve of the turbine, the amount of coal burned, the feed water flow rate and the recirculation water flow rate; the wet control data include: the main steam pressure, the output power, the main steam temperature and the water level of the water storage tank.

3. The wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to claim 1 is characterized in that: The mechanism model of the coal-fired power generation unit includes: in, is the amount of coal burned, is the amount of coal entering the coal mill, is the dynamic coefficient of coal burning, is the function fitted for coal mill; is the water flow rate, is the main steam temperature, is the function fitted to the boiler system, is the main steam temperature dynamic coefficient; is the opening of the main steam valve of the steam turbine, is the main steam pressure, is the main steam pressure dynamic coefficient, is the water supply dynamic coefficient, Function fitted to the boiler system; is the output power, is the power dynamic coefficient, Function fitted for the steam turbine system; is the recirculating water flow rate, is the water level in the water tank, is the dynamic coefficient of water level in the water tank, Function fitted to the water storage tank system.

4. The wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to claim 3 is characterized in that: The objective function of the wet state control prediction model of the coal-fired power generation unit for: in, N represents the time increment, i express N The index of Indicates the current moment, represents the predicted wet control data, Indicates the set wet state control data, represents the two-norm, Indicates the increment of wet operation data.

5. The wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to claim 4 is characterized in that: The constraints of the wet state control prediction model for coal-fired power generation units include: in, Indicates wet running data, Indicates the minimum value of wet operation data, Indicates the maximum value of wet operation data, Indicates the minimum value of the wet operation data increment, Indicates the maximum value of the wet operation data increment, Indicates the minimum value of the predicted wet control data, Indicates the maximum value of the predicted wet control data.

6. The wet state control method of a coal-fired power generation unit based on physical information neural network predictive control according to claim 5, characterized in that: It is necessary to train the wet state control prediction model for coal-fired power generation units: i. Collect historical state data of coal-fired power generation units in wet operation mode, use the wet operation data in the historical state data as input to the wet control prediction model of the coal-fired power generation units, and use the corresponding wet control data in the historical state data as labels; ii. Input the collected wet-state operation data into the wet-state control prediction model of the coal-fired power generation unit, and predict the wet-state control data with the goal of minimizing the objective function while satisfying the constraints; iii. Calculate the loss function including data error term and physical information error term based on the predicted wet control data and corresponding labels, and use the gradient descent method to update the wet control prediction model of the coal-fired power generation unit; iv. Repeat steps ii-iii until the loss function converges, completing the training of the wet state control prediction model for the coal-fired power generation unit.

7. A wet state control method for coal-fired power generation units based on physical information neural network predictive control according to claim 6, characterized in that: The loss function for: in, represents the sample size of the collected historical state dataset, j express The index of Indicates that through j Wet control data for sample prediction, Indicates the j wet control data in samples; represents the sample size involved in the construction of the mechanism model of coal-fired power generation units, l express The index of Indicates the l The wet control data calculated by the mechanism model of coal-fired power generation units, Indicates that through l Wet control data for sample prediction, Represents the weight coefficient.

8. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the wet state control method of a coal-fired power generation unit based on physical information neural network predictive control as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for wet state control of a coal-fired power generation unit based on physical information neural network predictive control as described in any one of claims 1 to 7 is implemented.

Images