A machine learning-based nonlinear predictive control method and system for a coal-fired unit coordination system

Abstract

The application discloses a kind of nonlinear prediction control method and system of coal-fired unit coordination system based on machine learning, the method first obtains coal-fired unit historical operation data, establishes nonlinear prediction model based on convolutional neural network and gated recurrent unit;Secondly, a rolling optimization mechanism is constructed to minimize output deviation and control increment;Then construct the deviation prediction model based on Gaussian regression process, update using real-time data and feedback correction to predicted output;Grey wolf optimization algorithm is used to solve the rolling optimization problem online, and the optimal control instruction sequence is obtained and issued to the actuator.The application improves the prediction accuracy, dynamic response speed and anti-interference ability of the unit coordination system under wide load and variable working conditions through deep learning modeling and online deviation compensation, significantly reduces the fluctuation amplitude of main steam pressure and steam-water separator outlet enthalpy, realizes high-precision coordinated control of fuel quantity, main steam valve opening and feedwater flow, and has strong engineering practicability.

Description

Technical Field

[0001] This invention relates to the field of coal-fired power unit control technology, specifically to a nonlinear predictive control method and system for a coal-fired power unit coordination system based on machine learning. Background Technology

[0002] Coal-fired power generating units are the cornerstone of my country's energy supply system, serving as crucial support for power grid peak and frequency regulation, compensation for renewable energy, and load fluctuations. The boiler and turbine coordinated control system, as a core component of coal-fired power generating units, directly determines the unit's ability to respond quickly and accurately to grid load commands while maintaining its own safe and efficient operation. However, the strong nonlinearity, large inertia, and multivariable coupling characteristics of coal-fired units under peak and deep-range operating conditions, coupled with frequent fluctuations in fuel quality, changes in equipment status, and environmental disturbances, pose severe challenges to efficient and reliable operation control.

[0003] Traditional coordinated control strategies typically employ a control structure based on the PID method. However, when dealing with multivariable coupled systems and strongly nonlinear dynamic processes, the control quality of PID-based control methods often deteriorates significantly, making it difficult to meet the control requirements for flexible operation of the unit under a wide range of operating conditions.

[0004] Model predictive control (MRC) algorithms have demonstrated great potential in the coordinated control of coal-fired power units due to their advantages in predicting future performance, explicitly handling constraints, and multi-objective optimization. However, MRC algorithms rely on the effective construction of predictive models. Existing MRC methods are mostly based on linear models identified under given operating conditions or their weighted combinations, which are difficult to describe well the strongly nonlinear dynamic characteristics of the unit under a wide range of operating condition transitions. Summary of the Invention

[0005] The purpose of this invention is to provide a nonlinear predictive control method and system for a coal-fired power unit coordination system based on machine learning, in order to solve the technical problems of existing coal-fired power unit coordination control under wide load operating conditions, which is difficult to cope with the strong nonlinearity, large inertia and strong coupling characteristics of the system, resulting in slow load response speed and large fluctuations of key parameters.

[0006] To achieve the above objectives, the technical solution provided by this invention is: a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, comprising the following steps: S1: Obtain historical operating data of the coal-fired power unit and establish a nonlinear prediction model for the coal-fired power unit coordination system; the input of the nonlinear prediction model is the current fuel quantity, main steam valve opening and feedwater flow rate, as well as the power generation, main steam pressure and steam-water separator outlet enthalpy value of the previous two time points; its output is the predicted value of the power generation, main steam pressure and steam-water separator outlet enthalpy value of the current time. S2: Construct a rolling optimization mechanism, the objective function of which is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command; the constraints include the amplitude constraint and the rate of change constraint of the control command. S3: Construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation, obtain the corrected future output prediction value, and substitute the corrected prediction value into the rolling optimization mechanism in step S2. S4: The gray wolf optimization algorithm is used to solve the rolling optimization mechanism online to obtain the optimal control command sequence in the future control time domain; S5: The first element of the optimal control command sequence is the control command at the current moment, which is then sent to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit to perform nonlinear predictive control of the coal-fired unit coordination system.

[0007] To optimize the above technical solution, the specific measures also include: In step S1, the nonlinear prediction model of the coal-fired power unit coordination system is established based on a convolutional neural network and a gated recurrent unit algorithm, and the nonlinear mapping relationship is as follows:

[0008] in, , and The coordination system for coal-fired power units is currently t The power generation capacity, main steam pressure, and steam enthalpy at the outlet of the steam-water separator at any given time; , and The coordination system for coal-fired power units is currently t Fuel quantity, main steam valve opening, and feedwater flow rate at any given time; subscript t-1 and t-2 Represent t-1 Time and t-2 time; This represents the nonlinear mapping function that couples the convolutional neural network with the gated recurrent unit.

[0009] Furthermore, the coupling method of the convolutional neural network and the gated recurrent unit algorithm is a serial hierarchical fusion, specifically: the one-dimensional convolutional neural network module performs convolution operations on the multi-dimensional input parameters at each time step to extract the spatial coupling features between different operating parameters at the same time step; the feature-enhanced time-series parameters output by the convolutional neural network are input to the gated recurrent unit module, and the temporal dependencies across time steps are captured through the gating mechanism of reset gate and update gate; finally, the spatial features and temporal features are integrated through a fully connected layer to output the predicted values ​​of power generation, main steam pressure and steam-water separator outlet enthalpy at the current time step.

[0010] In step S2, the objective function of the rolling optimization mechanism is specifically expressed as follows:

[0011] in, N p For prediction in the time domain; N c To control the time domain; k This is the current sampling time; for k Predicting time domain The output of the coal-fired power unit coordination system at all times; for The output setpoint of the coal-fired power unit coordination system at all times; for k Time calculation Constantly control the increment; Q c This is the error weight matrix; R c This is the control weight matrix.

[0012] In step S3, the step of feedback correction of the output of the nonlinear prediction model based on the prediction deviation specifically involves: A deviation prediction model based on Gaussian processes is constructed, with the following expression:

[0013] The inputs to the deviation prediction model include: from the previous n The historical deviation sequence from time 1 to the previous time and the historical control input from time n to the previous time; the output of the deviation prediction model is the prediction time domain. Np Deviation sequence within; Represents the Gaussian process model function; Indicates the deviation amount; This indicates the control quantity.

[0014] Furthermore, the deviation between the actual output of the coal-fired unit coordination system at historical moments and the predicted output of the nonlinear prediction model is calculated, expressed as:

[0015] in, for k The actual system output measurement value at any given time; for k-1 Time Prediction k The predicted output value of the system at time t.

[0016] Furthermore, the output of the nonlinear prediction model is corrected based on the prediction bias, specifically as follows:

[0017] in, express k Time prediction N p The output of the prediction model for each step; express k Predicted output at time step; express k Time prediction N p The output of the Gaussian process deviation prediction model for the step; k This is the current sampling time; N p For prediction in the time domain.

[0018] In step S4, the Grey Wolf optimization algorithm solves the rolling optimization mechanism online. The specific process is as follows: The wolf pack size, maximum number of iterations, and search space dimension are set, and the position of each wolf is randomly initialized; the wolf pack size represents the total number of candidate control instruction sequences; the search space dimension represents the total number of decision variables contained in each candidate control sequence; and the position of each wolf represents a candidate control instruction sequence. Furthermore, the position vector of each wolf is substituted into the objective function of the rolling optimization mechanism to calculate its fitness value, and the fitness value is used to determine the current population's... α Wolf, β wolves and δ Wolf; Based on the gray wolf optimization algorithm, update the position of each ordinary wolf, causing it to move towards... α Wolf, β wolves and δ The wolf is located nearby.

[0019] Furthermore, the fitness is iteratively calculated and the wolf pack positions are updated. After each iteration, the wolf pack's quality is reassessed based on the fitness function value, and the results are updated accordingly. α Wolf, β wolves and δThe wolf's position is determined until the maximum number of iterations is reached, and the final... α The wolf position vector is output as the optimal control command sequence; the quality level of the wolf pack represents the objective function value corresponding to the selected control sequence.

[0020] As another important technical solution, the present invention also provides a nonlinear predictive control system for a coal-fired power unit coordination system based on machine learning, comprising: The nonlinear dynamic modeling module is used to acquire historical operating data of coal-fired power units and establish a nonlinear prediction model for the coal-fired power unit coordination system. The inputs of the nonlinear prediction model are the fuel quantity, main steam valve opening and feedwater flow rate at the current moment, as well as the power generation, main steam pressure and steam-water separator outlet enthalpy at the previous two moments. Its output is the predicted values ​​of the power generation, main steam pressure and steam-water separator outlet enthalpy at the current moment. The deviation modeling and output correction module is used to construct a rolling optimization mechanism. The objective function of the rolling optimization mechanism is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command. The constraints include the amplitude constraint and the rate of change constraint of the control command. The multi-objective rolling optimization module is used to construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, and perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation to obtain the corrected future output prediction value. The corrected prediction value is then substituted into the rolling optimization mechanism of the deviation modeling and output correction module. The intelligent optimization solution module is used to solve the rolling optimization mechanism online using the gray wolf optimization algorithm to obtain the optimal control command sequence in the future control time domain; The real-time control execution module is used to take the first element in the optimal control command sequence as the control command at the current moment and send it to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit, respectively, for nonlinear predictive control of the coal-fired unit coordination system.

[0021] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning as described above.

[0022] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, as described above.

[0023] Compared with the prior art, the beneficial effects of the present invention are: This invention employs a nonlinear prediction model constructed by serially fusing convolutional neural networks and gated recurrent units. This model can simultaneously extract the spatial coupling characteristics between operating parameters and the temporal dependencies across time periods, providing a more accurate characterization of the dynamic characteristics of the unit under a wide range of variable load conditions. Compared with traditional linear models or weighted combination models, the prediction error is significantly reduced, providing a more reliable basis for subsequent rolling optimization.

[0024] This invention, through the solution capabilities of the predictive model and the Grey Wolf optimization algorithm, enables the generating power of the unit to approach the set value more quickly after receiving the AGC command, significantly shortening the response delay. In scenarios with short-range rapid load fluctuations, the adjustment time is greatly reduced compared to the comparative methods, effectively improving the problems of slow response and lag in traditional methods.

[0025] The method of this invention can significantly suppress overshoot and oscillation, especially during long-term large-scale load increases and decreases. The peak fluctuation and recovery time of the main steam pressure are significantly better than those of the comparative method. The fluctuation range of the enthalpy value at the outlet of the steam-water separator is greatly reduced, which helps to reduce the risk of overheating of the steam-water separator and improve operational safety and equipment life.

[0026] This invention introduces a deviation prediction model based on Gaussian regression and uses real-time data for online rolling updates, enabling the system to automatically compensate for model mismatch caused by factors such as coal quality fluctuations, equipment aging, and environmental changes. Even under continuous load changes or external disturbances, the control quality remains stable, and the robustness is significantly improved compared to fixed parameter models or PID controllers. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0028] Figure 2 This is a control scheme strategy diagram of the present invention.

[0029] Figure 3 The control method proposed in this invention, along with the PID control method and the conventional model predictive control method, is compared with the controlled variable in a short-range fluctuation scenario of AGC commands.

[0030] Figure 4 The diagram shows a comparison of the control quantities of the control method proposed in this invention with the PID control method and the conventional model predictive control method under the short-range fluctuation scenario of AGC commands.

[0031] Figure 5 The control method proposed in this invention, along with the PID control method and the conventional model predictive control method, is compared in terms of the controlled variable under long-range fluctuation scenarios of AGC commands.

[0032] Figure 6 The diagram shows a comparison of the control quantities of the proposed control method, PID control method, and conventional model predictive control method under long-range fluctuation scenarios of AGC commands, as presented in a specific example of the present invention. Detailed Implementation

[0033] The present invention will be further described in detail below through specific embodiments, but it should not be construed as limiting the scope of the subject matter of the present invention to the following embodiments. All technologies implemented based on the above content of the present invention fall within the scope of the present invention.

[0034] In some implementations, such as Figure 1 As shown, this invention provides a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, comprising the following steps: S1: Obtain historical operating data of the coal-fired power unit and establish a nonlinear prediction model for the coal-fired power unit coordination system; the input of the nonlinear prediction model is the current fuel quantity, main steam valve opening and feedwater flow rate, as well as the power generation, main steam pressure and steam-water separator outlet enthalpy value of the previous two time points; its output is the predicted value of the power generation, main steam pressure and steam-water separator outlet enthalpy value of the current time. The nonlinear prediction model for the coordination system of coal-fired power units is established based on convolutional neural networks and gated recurrent unit algorithms. The specific nonlinear mapping relationship is as follows:

[0035] in, , and The coordination system for coal-fired power units is currently t The power generation capacity, main steam pressure, and steam enthalpy at the outlet of the steam-water separator at any given time; , and The coordination system for coal-fired power units is currently t Fuel quantity, main steam valve opening, and feedwater flow rate at any given time; subscript t-1 and t-2 Represent t-1 Time and t-2 time; This represents the nonlinear mapping function that couples the convolutional neural network with the gated recurrent unit.

[0036] In some implementations, the coupling method between the convolutional neural network and the gated recurrent unit algorithm is a serial hierarchical fusion, specifically: through a one-dimensional convolutional neural network module... t , t -1、 t-2. Convolution operation is performed on the multi-dimensional input parameters (fuel quantity, main steam valve opening, feedwater flow rate, power generation, main steam pressure, steam enthalpy at the steam-water separator outlet) at each time point to extract the spatial coupling characteristics between different operating parameters at the same time point. Specifically, this includes the strong coupling relationship between fuel quantity and power generation, the correlation characteristics between feedwater flow rate and steam enthalpy at the steam-water separator outlet, and the nonlinear mapping law between main steam valve opening and main steam pressure. The feature-enhanced temporal parameters output from the convolutional neural network are input into the gated recurrent unit module. Through a gating mechanism involving reset and update gates, temporal dependencies across time steps are captured. Specifically, this includes... t The change in fuel quantity at time -2 t The hysteresis effect of main steam pressure at any given time t -1 moment main steam valve opening adjustment t The dynamic response law of power generation at any time is obtained; finally, the spatial and temporal characteristics are integrated through the fully connected layer to output the predicted values ​​of power generation, main steam pressure and steam-water separator outlet enthalpy at the current time, so as to achieve accurate modeling of the nonlinear dynamic characteristics of the unit coordination system.

[0037] S2: Construct a rolling optimization mechanism, the objective function of which is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command; the constraints include the amplitude constraint and the rate of change constraint of the control command. The objective function of the rolling optimization mechanism is specifically expressed as:

[0038] in, N p For prediction in the time domain; N c To control the time domain; k This is the current sampling time; for k Predicting time domain The output of the coal-fired power unit coordination system at all times includes power generation, main steam pressure, and steam enthalpy at the steam-water separator outlet; for The output setpoint of the coal-fired power unit coordination system at all times; for k Time calculation Constantly control the increment; Q c This is the error weight matrix; the larger the value, the better the tracking performance of the setpoint. R c This is the control weight matrix, used to limit drastic changes in control increments.

[0039] In some implementations, the constraints of the rolling optimization mechanism include the magnitude constraints of the control command and its rate of change constraints, specifically:

[0040] in, Indicates the upper limit of the control quantity; Indicates the lower limit of the control quantity; This indicates the upper limit of the control variation; Indicates the lower limit of the control change; S3: Construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation, obtain the corrected future output prediction value, and substitute the corrected prediction value into the rolling optimization mechanism in step S2. The output of the nonlinear prediction model is corrected based on the prediction deviation. The specific process is as follows: A deviation prediction model based on Gaussian processes is constructed, with the following expression:

[0041] The inputs to the deviation prediction model include: from the previous n Historical deviation sequence from time to the previous time and the previous time n Historical control inputs from time point A to the previous time point; the output of the deviation prediction model is the prediction time domain. Np Deviation sequence within; Represents the Gaussian process model function; Indicates the deviation amount; This indicates the control quantity.

[0042] The deviation between the actual output of the coal-fired power unit coordination system at historical moments and the predicted output of the nonlinear prediction model is calculated using the following expression:

[0043] in, for k The actual system output measurement value at any given time; for k-1 Time Prediction k The predicted output value of the system at time t.

[0044] In some implementations, during use, the deviation prediction model will be updated at intervals of... m The frequency of each sampling time is used to update the rolling window with the latest collected data, and the Gaussian regression process model is retrained to achieve online rolling updates of the prediction model.

[0045] The output of the nonlinear prediction model is corrected based on the prediction bias, specifically as follows:

[0046] in, express k Time prediction N p The output of the prediction model for each step; express k Predicted output at time step; express k Time prediction N p The output of the Gaussian process deviation prediction model for the step; k This is the current sampling time; N p For prediction in the time domain.

[0047] S4: The gray wolf optimization algorithm is used to solve the rolling optimization mechanism online to obtain the optimal control command sequence in the future control time domain; In some implementations, wolf populations are divided into four levels: α Wolf, β Wolf, δ wolves and ω Wolf. α The wolf's position is X α The fitness value is optimal; β The wolf's position is X β The fitness value is optimal; δ The wolf's position is X δ The fitness value is optimal; ω The wolf refers to the remaining wolf members.

[0048] Set wolf pack size N wolf The maximum number of iterations and the search space dimension are defined. The position of each wolf is randomly initialized. The wolf pack size represents the total number of candidate control command sequences. The search space dimension represents the total number of decision variables contained in each candidate control sequence. The position of each wolf represents a candidate control command sequence. U i :

[0049] Substitute the position vector of each wolf into the objective function of the rolling optimization mechanism to calculate its fitness value, and determine the current population based on the fitness value. α Wolf, β wolves and δ Wolf; the fitness function is the objective function of the rolling optimization mechanism.

[0050] The position of each ordinary wolf is updated based on the gray wolf optimization algorithm. X i Location range [ X min， X max ], so that it moves towards α Wolf, β wolves and δ The wolf's location is close, and its position update equation is:

[0051] and The relevant regulating factor is represented by the following update equation:

[0052] in, a It is a linearly decreasing coefficient, decreasing linearly from 2 to 0 during the iteration process; r i and It is a random number in [0,1].

[0053] The location of a typical wolf pack is determined by α Wolf, β wolves and δ The wolves' positions are determined by their shared location:

[0054] in, n This represents the current iteration number. For the first n+1 The wolf pack's position in the next iteration. For the first n+1 During the next iteration α Wolf pack location; For the first t+1 During the next iteration β Wolf pack location; For the first n+1 During the next iteration δ Location of the wolf pack.

[0055] The fitness of the wolf pack is iteratively calculated and its position is updated. After each iteration, the pack's quality is reassessed based on the fitness function value, and the results are updated accordingly. α Wolf, β wolves and δ The wolf's position is determined until the maximum number of iterations is reached, and the final... α The wolf position vector is output as the optimal control command sequence; the quality level of the wolf pack represents the objective function value corresponding to the selected control sequence.

[0056] In some implementations, after each iteration, the wolf pack's quality is reassessed based on the fitness function value, and the results are updated. α , β , δ The wolf's location, and the corresponding... A and C Coefficients. Parameters a The update equation is:

[0057] in, a init =2, T max This represents the maximum number of iterations.

[0058] S5: The first element of the optimal control command sequence is the control command at the current moment, which is then sent to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit to perform nonlinear predictive control of the coal-fired unit coordination system.

[0059] Example 1 To verify the superiority of the nonlinear predictive control method for coal-fired power unit coordination system based on machine learning proposed in this invention, two simulation scenarios were designed: 1. Short-range fluctuations in AGC commands, i.e., rapid and small fluctuations in the power generation setpoint. 2. Long-range fluctuations in AGC commands, i.e., large fluctuations in the power generation setpoint. Sampling time T s =10s, prediction time domain N p =30, control time domain N c =10, the weight matrix is ​​set to... Q c = diag(100,50,10); R c = diag(10,10,5). The amplitude and rate constraints of the control variables are as follows: ; ; .

[0060] Two conventional control schemes were compared: 1. Conventional PI control: The coal-fired unit coordination system adopts a boiler-following control mode, and each loop uses a PI controller. 2. Linear model predictive control: The predictive model adopts a linear state-space model, and the parameter settings are consistent with the predictive control method proposed in this paper.

[0061] The simulation results of short-range fluctuations in AGC instructions are as follows: Figure 3 and Figure 4As shown in the figure, under the short-range fluctuation scenario of AGC commands, the PI control method exhibits a certain overshoot, resulting in significant dynamic fluctuations in power generation and the enthalpy at the steam-water separator outlet, making it difficult for the system to stabilize quickly. While the linear model predictive control algorithm can improve some control performance, it still suffers from some adjustment deviation under multiple fluctuations in the power generation setpoint. In contrast, the proposed nonlinear model predictive control method has a faster output response, significantly reduced oscillations, and smoother adjustment.

[0062] Table 1 shows the IAE control performance indices for each output variable under different control methods: Table 1 Summary of IAE Indicators for Various Control Methods under AGC Short-Range Fluctuation Simulation

[0063] The results show that the IAE indices of this invention for the three output variables of power generation, main steam pressure and outlet steam enthalpy are 473.26, 43.51 and 127.33, respectively, which are significantly better than PI control (2303.31, 24.33 and 766.25) and linear MPC control (555.70, 48.03 and 291.36), verifying its effectiveness in the coordinated control system of coal-fired power generating units.

[0064] The simulation results of long-range fluctuations in AGC instructions are as follows: Figure 5 and Figure 6 As shown in the figure, under the long-range fluctuation condition of AGC commands, the control performance of the three control methods on the system output variables differs significantly during the process of a large load decrease. The PI control method exhibits severe lag and large deviation after the setpoint disturbance, especially with drastic fluctuations in the steam enthalpy value at the steam-water separator outlet, resulting in the worst control performance. Although linear model predictive control is superior to PI in response speed, it still suffers from slow regulation and large regulation deviation. In contrast, the proposed nonlinear model predictive control method has a more stable response on all output variables, and the system can track the setpoint more quickly.

[0065] Table 2 shows the IAE control performance indices for each output variable under different control methods: Table 2 Summary of IAE Indicators for Various Control Methods under AGC Long-Range Fluctuation Simulation

[0066] The results show that the proposed invention achieves better performance in power generation (10026.82), main steam pressure (154.88), and steam enthalpy (3698.20), respectively, than PI control (13454.30, 314.66, 29850.81) and linear model predictive control (11331.66, 303.41, 4002.89), with a particularly significant improvement in the control accuracy of the steam enthalpy at the steam-water separator outlet. These results further validate the superiority of the proposed control strategy in handling long-range fluctuations in AGC commands.

[0067] In another embodiment of the present invention, a nonlinear predictive control system for a coal-fired power unit coordination system based on machine learning is proposed, comprising: The nonlinear dynamic modeling module is used to acquire historical operating data of coal-fired power units and establish a nonlinear prediction model for the coal-fired power unit coordination system. The inputs of the nonlinear prediction model are the fuel quantity, main steam valve opening and feedwater flow rate at the current moment, as well as the power generation, main steam pressure and steam-water separator outlet enthalpy at the previous two moments. Its output is the predicted values ​​of the power generation, main steam pressure and steam-water separator outlet enthalpy at the current moment. The deviation modeling and output correction module is used to construct a rolling optimization mechanism. The objective function of the rolling optimization mechanism is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command. The constraints include the amplitude constraint and the rate of change constraint of the control command. The multi-objective rolling optimization module is used to construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, and perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation to obtain the corrected future output prediction value. The corrected prediction value is then substituted into the rolling optimization mechanism of the deviation modeling and output correction module. The intelligent optimization solution module is used to solve the rolling optimization mechanism online using the gray wolf optimization algorithm to obtain the optimal control command sequence in the future control time domain; The real-time control execution module is used to take the first element in the optimal control command sequence as the control command at the current moment and send it to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit, respectively, for nonlinear predictive control of the coal-fired unit coordination system.

[0068] In another embodiment of the present invention, an electronic device is proposed, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning as described above.

[0069] In another embodiment of the present invention, a computer-readable storage medium is provided storing a computer program that causes a computer to execute a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, as described above.

[0070] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent substitutions, and improvements made by those skilled in the art to the above embodiments without departing from the scope of the technical solution of the present invention, based on the technical essence of the present invention, shall still fall within the protection scope of the technical solution of the present invention.

Claims

1. A nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, characterized in that, Includes the following steps: S1: Obtain historical operating data of the coal-fired power unit and establish a nonlinear prediction model for the coal-fired power unit coordination system; the input of the nonlinear prediction model is the current fuel quantity, main steam valve opening and feedwater flow rate, as well as the power generation, main steam pressure and steam-water separator outlet enthalpy value of the previous two time points; its output is the predicted value of the power generation, main steam pressure and steam-water separator outlet enthalpy value of the current time. S2: Construct a rolling optimization mechanism, the objective function of which is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command; the constraints include the amplitude constraint and the rate of change constraint of the control command. S3: Construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation, obtain the corrected future output prediction value, and substitute the corrected prediction value into the rolling optimization mechanism in step S2. S4: The gray wolf optimization algorithm is used to solve the rolling optimization mechanism online to obtain the optimal control command sequence in the future control time domain; S5: The first element of the optimal control command sequence is the control command at the current moment, which is then sent to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit to perform nonlinear predictive control of the coal-fired unit coordination system.

2. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 1, characterized in that: In step S1, the nonlinear prediction model of the coal-fired power unit coordination system is established based on a convolutional neural network and a gated recurrent unit algorithm, and the nonlinear mapping relationship is as follows: in, , and The coordination system for coal-fired power units is currently t The power generation capacity, main steam pressure, and steam enthalpy at the outlet of the steam-water separator at any given time; , and The coordination system for coal-fired power units is currently t Fuel quantity, main steam valve opening, and feedwater flow rate at any given time; subscript t-1 and t-2 Represent t-1 Time and t-2 time; This represents the nonlinear mapping function that couples the convolutional neural network with the gated recurrent unit.

3. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 2, characterized in that: The coupling method of the convolutional neural network and the gated recurrent unit algorithm is a serial hierarchical fusion. Specifically, the one-dimensional convolutional neural network module performs convolution operations on the multi-dimensional input parameters at each time step to extract the spatial coupling features between different operating parameters at the same time step; the feature-enhanced time-series parameters output by the convolutional neural network are input to the gated recurrent unit module, and the temporal dependencies across time steps are captured through the gating mechanism of reset gate and update gate; finally, the spatial features and temporal features are integrated through a fully connected layer to output the predicted values ​​of power generation, main steam pressure and steam-water separator outlet enthalpy at the current time step.

4. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 1, characterized in that: In step S2, the objective function of the rolling optimization mechanism is specifically expressed as follows: in, N p For prediction in the time domain; N c To control the time domain; k This is the current sampling time; for k Predicting time domain The output of the coal-fired power unit coordination system at all times; for The output setpoint of the coal-fired power unit coordination system at all times; for k Time calculation Constantly control the increment; Q c This is the error weight matrix; R c This is the control weight matrix.

5. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 1, characterized in that: In step S3, the step of feedback correction of the output of the nonlinear prediction model based on the prediction deviation specifically involves: in, express k Time prediction N p The output of the prediction model for each step; express k Predicted output at time step; express k Time prediction N p Output of the Gaussian process deviation prediction model; k This is the current sampling time; N p For prediction in the time domain.

6. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 1, characterized in that: In step S4, the gray wolf optimization algorithm is used to solve the rolling optimization mechanism online. The specific process is as follows: The wolf pack size, maximum number of iterations, and search space dimension are set, and the position of each wolf is randomly initialized; the wolf pack size represents the total number of candidate control instruction sequences; the search space dimension represents the total number of decision variables contained in each candidate control sequence; and the position of each wolf represents a candidate control instruction sequence. Substitute the position vector of each wolf into the objective function of the rolling optimization mechanism to calculate its fitness value, and determine the current population based on the fitness value. α Wolf, β wolves and δ Wolf; The position of each ordinary wolf is updated according to the gray wolf optimization algorithm, causing it to move towards... α Wolf, β wolves and δ The wolf is located nearby.

7. The nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning according to claim 6, characterized in that: The solution process of the gray wolf optimization algorithm also includes: iteratively calculating the fitness and updating the wolf pack position. After each iteration, the quality level of the wolf pack is re-evaluated based on the fitness function value, and the position is updated. α Wolf, β wolves and δ The wolf's position is determined until the maximum number of iterations is reached, and the final... α The wolf position vector is output as the optimal control command sequence; the quality level of the wolf pack represents the objective function value corresponding to the selected control sequence.

8. A nonlinear predictive control system for a coal-fired power unit coordination system based on machine learning, characterized in that, include: The nonlinear dynamic modeling module is used to acquire historical operating data of coal-fired power units and establish a nonlinear prediction model for the coal-fired power unit coordination system. The inputs to the nonlinear prediction model are the current fuel quantity, main steam valve opening, and feedwater flow rate, as well as the power generation, main steam pressure, and steam-water separator outlet enthalpy from the previous two time points; its outputs are the predicted values ​​of the current power generation, main steam pressure, and steam-water separator outlet enthalpy. The deviation modeling and output correction module is used to construct a rolling optimization mechanism. The objective function of the rolling optimization mechanism is to minimize the deviation between the system output variable and the desired setpoint, and to minimize the change in the control command. The constraints include the amplitude constraint and the rate of change constraint of the control command. The multi-objective rolling optimization module is used to construct a deviation prediction model based on Gaussian regression process, calculate the deviation between the actual output of the coal-fired unit coordination system at historical time and the predicted output of the nonlinear prediction model, use the latest collected data to perform online rolling updates of the Gaussian regression process model, and perform feedback correction on the output of the nonlinear prediction model according to the prediction deviation to obtain the corrected future output prediction value. The corrected prediction value is then substituted into the rolling optimization mechanism of the deviation modeling and output correction module. The intelligent optimization solution module is used to solve the rolling optimization mechanism online using the gray wolf optimization algorithm to obtain the optimal control command sequence in the future control time domain; The real-time control execution module is used to take the first element in the optimal control command sequence as the control command at the current moment and send it to the fuel quantity actuator, main steam valve opening actuator, and feedwater flow actuator of the coal-fired unit, respectively, for nonlinear predictive control of the coal-fired unit coordination system.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program causes the computer to execute a nonlinear predictive control method for a coal-fired power unit coordination system based on machine learning, as described in any one of claims 1 to 7.

Images