A main steam temperature model predictive control method and system based on preset performance constraints

Abstract

The application discloses a kind of based on preset performance constraint's main steam temperature model predictive control method and system, it is related to thermal automatic control technical field.The method includes: offline pre-training neural network prediction model using historical operation data, to approximate the dynamic and output mapping of controlled object;Online at each sampling time based on the neural network model is constructed and solved input constraint model predictive control optimization problem, wherein the cost function introduces the barrier term formed by preset performance function, limit the error envelope that converges with time in predicted tracking error;According to the deviation and error margin of actual output and model output, construct compensation control amount, and with predicted control amount superposition form final control rate;Through online parameter update, correct model error to adapt to working condition change.The application does not need accurate mechanism model, can realize main steam temperature fast, small deviation tracking, effectively suppresses overtemperature and frequent manual intervention.

Description

Technical Field

[0001] This invention relates to the field of thermal automatic control technology, and in particular to a main steam temperature model predictive control method and system based on preset performance constraints. Background Technology

[0002] Main steam temperature is a core indicator for the safe and economical operation of thermal power boiler turbine units, directly affecting unit thermal efficiency and equipment lifespan. In engineering, it is typically regulated through a superheater spray desuperheating system. However, this controlled object exhibits significant large delay and inertia characteristics, and the desuperheating valves display nonlinear characteristics. Under deep peak shaving and variable load operation conditions, frequent disturbances such as fuel quality fluctuations, changes in combustion state, and flue gas distribution adjustments lead to increased dynamic deviations in main steam temperature, decreased steady-state accuracy, and even triggering over-temperature or under-temperature alarms. This necessitates frequent manual intervention by operators, severely impacting the unit's automation level and operational safety.

[0003] Traditional main steam temperature control often employs PID cascade control or model predictive control based on step response models. However, PID control has limited regulation quality for objects with large time delays and insufficient adaptability to varying operating conditions. Conventional model predictive control relies on an accurate mathematical model of the controlled object, but the main steam temperature channel mechanism is complex, involving multiple heat transfer processes such as radiation, convection, and conduction, making it difficult to establish a high-precision mechanism model covering all operating conditions. Step test modeling methods are constrained by operating conditions, resulting in high testing costs and long cycles, and the obtained models exhibit poor transferability under changes in coal quality or large-scale load adjustments, leading to significant degradation in control performance.

[0004] In recent years, data-driven modeling methods such as neural networks have provided new approaches to the control of complex thermal objects. However, existing methods mostly focus on open-loop modeling accuracy and rarely address the dynamic convergence process of explicitly constraining tracking errors in closed-loop control. Transient error problems remain prominent when model mismatches or operating condition drift occur. Furthermore, the comprehensive integration of robust performance constraints and online adaptive updates of the controller in existing control methods is insufficient, making it difficult to achieve rapid, low-overshoot, end-to-end constraint tracking of main steam temperature under conditions of unknown models and frequent disturbances.

[0005] Therefore, there is an urgent need for a main steam temperature optimization control method that does not rely on an accurate mechanism model and takes into account both transient performance constraints and online adaptive capabilities. Summary of the Invention

[0006] To address this, embodiments of the present invention provide a main steam temperature model prediction control method and system based on preset performance constraints, which solves the problems in the prior art where main steam temperature control relies on precise mechanism models, suffers from severe dynamic deviations under variable load conditions, and requires frequent manual intervention.

[0007] To address the aforementioned technical problems, this invention provides a main steam temperature model predictive control method based on preset performance constraints, applied to a boiler main steam temperature control system, wherein the system output is the main steam temperature. The reference input is the set value. The control input is the opening degree of the water spray desuperheating valve. The method includes:

[0008] Step S1: Collect historical operating data of the unit, and train the preset neural network prediction model offline to obtain a model for approximating the dynamics of the controlled object and the mapping between the output. The trained network parameters are used as initial values ​​for online updates, where... This is the output mapping function of the neural network prediction model. For neural network prediction models in The internal state vector at time t;

[0009] Step S2: At each sampling time A prediction model is established based on the neural network prediction model. And within the prediction time domain, using a preset performance function To solve the model predictive control optimization problem under constraints, the predictive control input is obtained. ,in Internal state vector of the neural network prediction model Regarding time The derivative, Let be the state self-evolution function. For the input gain function;

[0010] Step S3: Based on the optimal prediction output of the neural network prediction model With the preset performance function Calculate the error margin Then based on the actual output With the optimal prediction output deviation and the error margin Constructing compensation control quantity ;

[0011] Step S4: The predicted control quantity With the compensation control quantity By superimposing these values, the final control rate can be obtained. The output is sent to the actuator of the water spray desuperheating valve;

[0012] Step S5: Based on actual output With model output The deviation is used to construct a loss function, and the parameters of the neural network prediction model are recursively updated to correct the model error. The updated parameters are used in steps S2 and S3 at the next sampling time.

[0013] Preferably, in step S1, the controlled object is an unknown nonlinear system, whose dynamics are described by the following formula:

[0014] ;

[0015] in For system status, System status Regarding time The derivative, For disturbance, , , It is an unknown nonlinear function.

[0016] Preferably, in step S2, the model predictive control optimization problem is:

[0017] ;

[0018] ;

[0019] in, To control input To optimize the variables, we need to minimize them. At the current sampling time, To predict the length of the time domain, Let be the instantaneous cost function of the integrand. For the derivative with respect to time, To control the set of input constraints.

[0020] Preferably, in step S2, the control input constraint set This includes upper and lower limits for the opening of the water spray desuperheating valve and constraints on the rate of change of valve opening.

[0021] Preferably, in step S2, the cost function of the model predictive control optimization problem includes a barrier term, specifically:

[0022] when When, cost function ;

[0023] when When, cost function ;

[0024] in, For the input penalty weight, To determine the norm of the tracking error between the predicted output and the setpoint, the barrier term limits the predicted tracking error to the preset performance function. Within the defined error envelope.

[0025] Preferably, the preset performance function Let f be a monotonically decreasing function with a positive lower bound, and its form be: ,in , , These are adjustable parameters that determine the initial contraction amplitude, convergence rate, and steady-state tolerance of the error envelope, respectively.

[0026] Preferably, in step S3, the error margin The compensation control quantity based on The model is constructed with its gain modulated by the ReLU activation function to enhance compensation when the model bias approaches the error envelope boundary.

[0027] Preferably, in step S1, the offline pre-training adopts a supervised learning method, using the mean square error between the output of the neural network prediction model and the historical measurement output as the loss function for training, and the obtained parameters are used as the initial values ​​for online recursive updates in step S5.

[0028] Preferably, in step S5, the recursive update of the parameters of the neural network prediction model adopts gradient descent, recursive least squares, or a combination thereof, and the neural network parameters are iteratively updated once or multiple times based on the loss function at each sampling time to adapt to the operating condition drift caused by coal quality changes, load adjustments, and combustion fluctuations.

[0029] This invention also provides a main steam temperature model predictive control system based on preset performance constraints. This system is used to implement the aforementioned main steam temperature model predictive control method based on preset performance constraints, specifically including:

[0030] The data acquisition module is used to acquire unit operating data;

[0031] The neural network prediction model module is used for offline pre-training and online recursive updating to provide output predictions for the controlled object.

[0032] The model predictive controller module incorporates a pre-defined performance constraint optimization problem with a barrier term, and outputs predictive control variables. ;

[0033] The preset performance compensation controller module calculates the compensation control quantity based on the error margin and model deviation. ;

[0034] The control quantity synthesis module will and The final control command is formed by superimposing these commands.

[0035] The actuator module applies the final control command to the water spray desuperheating valve;

[0036] The online parameter correction module updates the parameters of the neural network prediction model module online based on the deviation between the actual output and the model output.

[0037] As can be seen from the above technical solutions, this invention application has the following beneficial effects:

[0038] First, there is no need to establish a precise mechanism model for all operating conditions. Instead, a neural network prediction model that combines offline pre-training with online recursive updates is used to directly form a learnable prediction model from the operating data, which solves the problem of difficulty in obtaining the main steam temperature object model.

[0039] Second, by introducing a barrier term into the model prediction control cost function, the prediction tracking error is explicitly constrained within the error envelope that converges over time as defined by the preset performance function, thereby improving the transient tracking quality and effectively suppressing over-temperature and under-temperature deviations.

[0040] Third, by constructing a preset performance compensation control quantity based on error margin and model deviation, and superimposing it with the predictive control quantity to form the final control law, the influence of model mismatch and unmeasurable disturbances is effectively suppressed, and the robustness under variable load conditions is improved.

[0041] Fourth, by recursively updating the neural network model through online parameter correction, the control system can adapt to changes in coal quality, combustion, load, and other operating conditions, reducing the number of manual interventions by operators and the frequency of alarms, thereby improving the economy and safety of unit operation. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Referring to the drawings will make the features and advantages of the present invention clearer. The drawings are illustrative and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0043] Figure 1 This is a flowchart of a main steam temperature model predictive control method based on preset performance constraints provided by the present invention;

[0044] Figure 2 This is a block diagram of a main steam temperature model predictive control system based on preset performance constraints provided by the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1:

[0047] To address the problems in existing main steam temperature control technologies, such as reliance on precise mechanistic models, severe dynamic deviations under varying load conditions, and the need for frequent manual intervention, this invention proposes a main steam temperature model predictive control method based on preset performance constraints. This method is applied to a boiler main steam temperature control system, where the system's output is the main steam temperature. The reference input is the set value. The control input is the opening degree of the water spray desuperheating valve. .

[0048] like Figure 1 As shown, the method includes the following steps S1 to S5:

[0049] Step S1: Collect historical operating data of the unit, and train the preset neural network prediction model offline to obtain a model for approximating the dynamics of the controlled object and the mapping between the output. The trained network parameters are then used as initial values ​​for online updates.

[0050] in, This is the output mapping function of the neural network prediction model. For neural network prediction models in The internal state vector at time step S5. The offline pre-training adopts a supervised learning method, using the mean square error between the output of the neural network prediction model and the historical measurement output as the loss function for training. The obtained parameters are used as the initial values ​​for online recursive updates in the subsequent step S5.

[0051] In this embodiment, the controlled object is an unknown nonlinear system, whose dynamics are described by the following equation:

[0052] ;

[0053] in, For system status, System status Regarding time The derivative, For disturbances (including fluctuations in coal quality and combustion, and adjustments in flue gas distribution, etc.). , , It is an unknown nonlinear function.

[0054] Traditional methods struggle to establish accurate mechanistic models for this unknown nonlinear system under all operating conditions. This invention uses a neural network prediction model to perform data-driven approximation of the aforementioned unknown function, effectively solving the problem of difficulty in obtaining models.

[0055] The collected historical operating data sequence includes at least the main steam temperature. Valve opening The data includes unit load and available auxiliary variables. These data are used to construct supervised learning training samples. After offline training, the parameters are stored in the controller as the starting point for online updates.

[0056] Step S2: At each sampling time A prediction model is established based on the neural network prediction model. And within the prediction time domain, using a preset performance function To solve the model predictive control optimization problem under constraints, the predictive control input is obtained. .

[0057] in, Internal state vector of the neural network prediction model Regarding time The derivative, The state self-evolution function is obtained by parameterization learning from a neural network and represents the dynamic change of the system state itself when the control input is zero. The input gain function, also learned through neural network parameterization, represents how the control input affects the state change.

[0058] The above prediction model is for the unknown nonlinear system in step S1. , , Online learnable approximation of functions, where For the internal state of the model, , , Let be the function to be updated, represented by a neural network.

[0059] The specific model predictive control optimization problem is as follows:

[0060] ;

[0061] ;

[0062] in, To control input To optimize the variables, we need to minimize them. At the current sampling time, To predict the length of the time domain, Let be the instantaneous cost function of the integrand. For the derivative with respect to time, To control the set of input constraints.

[0063] The set of control input constraints This includes upper and lower limit constraints on the opening of the water spray desuperheating valve and constraints on the valve opening change rate, such as physical limits on the valve opening from 0% to 100% and actuator constraints on the opening change rate not exceeding a certain limit.

[0064] The cost function of the model predictive control optimization problem includes a barrier term, specifically:

[0065] when When, cost function ;

[0066] when When, cost function ;

[0067] in, The input penalty weight is used to balance tracking accuracy and control energy, and to suppress excessive or frequent adjustment actions. This is the norm (typically the Euclidean norm, representing the magnitude of the deviation) of the tracking error between the predicted output and the setpoint. This barrier term forces the predicted tracking error to be limited to a preset performance function. Within the defined error envelope: when the tracking error approaches When, the denominator term As the cost function approaches zero, its value increases dramatically, creating a barrier effect that causes the optimization solver to automatically avoid error boundaries, ensuring that the tracking error in the prediction time domain is always less than zero. .

[0068] The preset performance function Let f be a monotonically decreasing function with a positive lower bound, and its form be: ,in , , These are adjustable parameters that determine the initial contraction magnitude, convergence rate, and steady-state tolerance of the error envelope. For example, they can be set to... That is, the initial maximum allowable deviation is approximately 24℃, which converges exponentially to a steady-state allowable deviation of 4℃ at a rate of approximately 2. The parameters can be tuned in engineering according to the object scale and control indicators.

[0069] Through the aforementioned barrier cost function, this invention achieves explicit constraints on the transient and steady-state performance of the main steam temperature tracking error, fundamentally improving the dynamic quality of closed-loop control.

[0070] Step S3: Based on the optimal prediction output of the neural network prediction model With the preset performance function Calculate the error margin Then based on the actual output With the optimal prediction output deviation and the error margin Constructing compensation control quantity .

[0071] Specifically, the error margin is defined as This represents the remaining margin between the current predicted tracking error and the error envelope boundary. Model bias is defined as... This refers to the mismatch between the actual output and the model's predicted output.

[0072] The compensation control quantity based on The model is constructed such that its gain is modulated by the ReLU (Linear Rectified Unit) activation function. The ReLU activation function is characterized by an output of zero when the input is negative and a linear output when the input is positive. In this invention, when the model biases... relative to error margin When the error is small, the compensation effect is mild; as the model bias increases and approaches the error envelope boundary (i.e., ... When it approaches zero, The value increases sharply, and after ReLU modulation, it produces a strong compensation control effect to suppress the adverse effects of model mismatch and external disturbances on actual tracking performance.

[0073] The introduction of this compensation mechanism enables the system to effectively maintain the actual tracking error within the preset performance envelope even when there is a model mismatch or changes in operating conditions, significantly improving robustness under variable load conditions.

[0074] Step S4: The predicted control quantity With the compensation control quantity By superimposing these values, the final control rate can be obtained. The output is sent to the actuator of the water spray desuperheating valve.

[0075] In this dual-channel control architecture By utilizing neural network prediction models and preset performance barriers, a feedforward-feedback optimization control function is provided to satisfy input constraints and performance constraints. Based on real-time model bias and error margin, online compensation and correction are performed for model mismatch and unmeasurable disturbances. The two work together to achieve high-quality tracking control of the main steam temperature.

[0076] Step S5: Based on actual output With model output The deviation is used to construct a loss function, and the parameters of the neural network prediction model are recursively updated to correct the model error. The updated parameters are used in steps S2 and S3 at the next sampling time.

[0077] Specifically, at each sampling time, based on the output error Construct a loss function (such as mean squared error loss), and use gradient descent, recursive least squares, or a combination thereof to iteratively update the neural network parameters once or multiple times. After the update is completed, proceed to the next sampling period to form a complete closed-loop adaptive control.

[0078] This online parameter correction mechanism enables the neural network prediction model to adapt to the operating condition drift caused by changes in coal quality, load adjustment, and combustion fluctuations, maintaining prediction accuracy and thus ensuring the continued effectiveness of preset performance constraints in actual operation, reducing the frequency of over-temperature and under-temperature alarms and the need for manual intervention.

[0079] Example 2:

[0080] like Figure 2 As shown, this invention provides a main steam temperature model predictive control system based on preset performance constraints. This system is used to implement the main steam temperature model predictive control method based on preset performance constraints described in Embodiment 1 above, and specifically includes:

[0081] The data acquisition module is used to acquire unit operating data, including main steam temperature. Valve opening Unit load and auxiliary variables, etc.;

[0082] The neural network prediction model module is used for offline pre-training and online recursive updates, providing output predictions for the controlled object. Internally, it contains a state self-evolution function represented by a neural network parameterization. Input gain function and output mapping function ;

[0083] The model predictive controller module incorporates a pre-defined performance constraint optimization problem with a barrier term. It solves this optimization problem at each sampling time and outputs the predictive control quantity. ;

[0084] Preset performance compensation controller module, based on error margin Deviation from the model Calculate the compensation control quantity Its gain is modulated by the ReLU activation function;

[0085] The control quantity synthesis module will and The final control command is formed by superposition. ;

[0086] The actuator module transmits the final control command. Acting on the water spray desuperheating valve;

[0087] Online parameter calibration module, based on actual output With model output The deviation is used to recursively update the parameters of the neural network prediction model module online.

[0088] The main steam temperature model predictive control system based on preset performance constraints in this embodiment is used to implement the aforementioned main steam temperature model predictive control method based on preset performance constraints. Therefore, the specific implementation method of this system can be found in the previous method embodiment section, and will not be repeated here to avoid redundancy.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0092] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A predictive control method for main steam temperature model based on preset performance constraints, characterized in that, It is applied to the main steam temperature control system of a boiler, and the output of the system is the main steam temperature. The reference input is the set value. The control input is the opening degree of the water spray desuperheating valve. The method includes: Step S1: Collect historical operating data of the unit, and train the preset neural network prediction model offline to obtain a model for approximating the dynamics of the controlled object and the mapping between the output. The trained network parameters are used as initial values ​​for online updates, where... This is the output mapping function of the neural network prediction model. For neural network prediction models in The internal state vector at time t; Step S2: At each sampling time A prediction model is established based on the neural network prediction model. And within the prediction time domain, using a preset performance function To solve the model predictive control optimization problem under constraints, the predictive control quantity is obtained. ,in Internal state vector of the neural network prediction model Regarding time The derivative, Let be the state self-evolution function. For the input gain function; Step S3: Based on the optimal prediction output of the neural network prediction model With the preset performance function Calculate the error margin Then based on the actual output With the optimal prediction output deviation and the error margin Constructing compensation control quantity ; Step S4: The predicted control quantity With the compensation control quantity By superimposing these values, the final control rate can be obtained. The output is sent to the actuator of the water spray desuperheating valve; Step S5: Based on actual output With model output The deviation is used to construct a loss function, and the parameters of the neural network prediction model are recursively updated to correct the model error. The updated parameters are used in steps S2 and S3 at the next sampling time.

2. The main steam temperature model predictive control method based on preset performance constraints according to claim 1, characterized in that, In step S1, the controlled object is an unknown nonlinear system, whose dynamics are described by the following equation: ； in For system status, System status Regarding time The derivative, For disturbance, , , It is an unknown nonlinear function.

3. The main steam temperature model predictive control method based on preset performance constraints according to claim 2, characterized in that, In step S2, the model predictive control optimization problem is: ； ； in, To control input To optimize the variables, we need to minimize them. At the current sampling time, To predict the length of the time domain, Let be the instantaneous cost function of the integrand. For the derivative with respect to time, To control the set of input constraints.

4. The main steam temperature model predictive control method based on preset performance constraints according to claim 3, characterized in that, In step S2, the set of control input constraints This includes upper and lower limits for the opening of the water spray desuperheating valve and constraints on the rate of change of valve opening.

5. The main steam temperature model predictive control method based on preset performance constraints according to claim 3, characterized in that, In step S2, the cost function of the model predictive control optimization problem includes a barrier term, specifically: when When, cost function ; when When, cost function ; in, For the input penalty weight, To determine the norm of the tracking error between the predicted output and the setpoint, the barrier term limits the predicted tracking error to the preset performance function. Within the defined error envelope.

6. The main steam temperature model predictive control method based on preset performance constraints according to claim 1 or 5, characterized in that, The preset performance function Let f be a monotonically decreasing function with a positive lower bound, and its form be: ,in , , These are adjustable parameters that determine the initial contraction amplitude, convergence rate, and steady-state tolerance of the error envelope, respectively.

7. The main steam temperature model predictive control method based on preset performance constraints according to claim 1, characterized in that, In step S3, the error margin The compensation control quantity based on The model is constructed with its gain modulated by the ReLU activation function to enhance compensation when the model bias approaches the error envelope boundary.

8. The main steam temperature model predictive control method based on preset performance constraints according to claim 1, characterized in that, In step S1, the offline pre-training adopts a supervised learning method, using the mean square error between the output of the neural network prediction model and the historical measurement output as the loss function for training, and the obtained parameters are used as the initial values ​​for online recursive updates in step S5.

9. The main steam temperature model predictive control method based on preset performance constraints according to claim 1, characterized in that, In step S5, the parameters of the neural network prediction model are updated recursively using gradient descent, recursive least squares, or a combination thereof. At each sampling time, the neural network parameters are updated iteratively once or multiple times based on the loss function to adapt to the operating condition drift caused by changes in coal quality, load adjustment, and combustion fluctuations.

10. A main steam temperature model predictive control system based on preset performance constraints, characterized in that, The system is used to implement the main steam temperature model predictive control method based on preset performance constraints as described in any one of claims 1 to 9, specifically including: The data acquisition module is used to acquire unit operating data; The neural network prediction model module is used for offline pre-training and online recursive updating to provide output predictions for the controlled object. The model predictive controller module incorporates a pre-defined performance constraint optimization problem with a barrier term, and outputs predictive control variables. ; The preset performance compensation controller module calculates the compensation control quantity based on the error margin and model deviation. ; The control quantity synthesis module will and The final control command is formed by superimposing these commands. The actuator module applies the final control command to the water spray desuperheating valve; The online parameter correction module updates the parameters of the neural network prediction model module online based on the deviation between the actual output and the model output.

Images