A method for rapid response and efficiency assurance of deep peak shaving in double reheat units

Abstract

This invention relates to the field of thermal power generation process control technology, and discloses a method for rapid response and efficiency assurance in deep peak shaving of double reheat units. First, it calculates the model deviation integral term in real time to construct an augmented state space vector. Then, it calculates the virtual heat capacity potential energy index based on the wall temperature deviation of the heated surface and performs adaptive correction, while simultaneously calculating the dynamic loss gradient vector using a local linearization method. Subsequently, it dynamically reconstructs the benchmark setpoint based on the aforementioned index and gradient. Next, it establishes a hierarchical safety constraint system and constructs a multi-objective optimization model including an economic penalty term. Finally, it solves the quadratic programming model, and generates the final control command after dual verification of mathematical solution validity and physical boundaries. This invention effectively utilizes thermal potential energy and suppresses high-energy-consumption regulation actions by integrating the unit's thermal storage state and energy efficiency evaluation indicators into the control loop in real time, achieving coordinated optimization of unit load response speed and operational economy under deep peak shaving conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power generation process control, and specifically to a method for deep peak shaving, rapid response and efficiency guarantee of a secondary reheat unit. Background Art

[0002] By increasing the initial steam parameters and adding a primary reheat process, the secondary reheat coal-fired generating unit significantly improves the overall plant thermal efficiency and is the main development direction of current large-capacity, high-parameter thermal power technology. Such units usually have a multi-stage complex heating surface system and long-distance steam connection pipelines. With the advancement of the construction of the new power system, the grid connection proportion of new energy sources such as wind power and photovoltaic power, which are random and volatile, continues to increase, and the grid peak shaving pressure is increasing day by day. In order to accommodate new energy and maintain the stability of the grid frequency, the functional positioning of thermal power units is changing from base-load power sources to regulating power sources, forcing secondary reheat units to frequently participate in deep peak shaving operations. This means that the unit not only needs to maintain a low-load condition for a long time, but also must be able to follow the Automatic Generation Control (AGC) command and perform frequent and large-scale power regulation within a wide load range to balance the real-time supply and demand fluctuations of the grid.

[0003] Compared with conventional primary reheat units, the secondary reheat unit increases the complexity of the thermal system and the length of the steam flow path, resulting in a significant increase in the metal heat storage of the heating surface and the steam-water volume, and has great thermal inertia and pure time-delay characteristics. Under the deep peak shaving condition, the existing coordinated control strategies are usually based on fixed parameter settings and are difficult to accurately evaluate and utilize the real-time thermal potential energy inside the unit, resulting in a slow tracking response of the unit to the grid load command. If the adjustment strength of the controller is simply increased in order to pursue the response speed, it is very easy to cause overshoot and oscillation of the main steam pressure or reheat steam temperature, and even lead to system parameter overlimit triggering of the trip protection, and it is impossible to meet the grid's requirements for rapid load change under the premise of ensuring operation safety.

[0004] The existing automatic generation control strategies mainly focus on eliminating the tracking error of the controlled variable and generally lack a real-time quantification and optimization mechanism for the economic cost of the adjustment process. During the rapid load change process, the control system often tends to adopt energy-consuming adjustment means, such as forcibly pulling the load by spraying excessive desuperheated water or rapidly operating the valve greatly, which will generate obvious throttling losses and irreversible thermal exergy losses. This adjustment method lacking an energy efficiency gradient can shorten the adjustment time to a certain extent, but at the cost of sacrificing the instantaneous operating efficiency of the unit, and it is difficult to achieve the coordinated optimization of the adjustment quality and operating energy consumption, which does not meet the current operating goal of energy conservation and consumption reduction of thermal power units under the background of deep peak shaving normalization. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for rapid response and efficiency assurance in deep peak shaving of double reheat units. This method solves the problems of low load response rate in existing double reheat units under deep peak shaving conditions due to the large thermal inertia and long time delay characteristics of multi-stage reheat systems, and the lack of real-time quantitative assessment of the unit's heat storage status and regulation energy consumption costs in existing control strategies.

[0006] To achieve the above objectives, this invention provides a method for rapid response and efficiency assurance of deep peak shaving in double reheat units, comprising the following steps: Step S1: Collect unit operation data in real time and preprocess it, calculate the model deviation integral term of the key controlled variable, and concatenate the model deviation integral term with the basic state variable to construct an augmented state space vector. Step S2: Calculate the virtual heat capacity potential energy index based on the wall temperature deviation of the unit's heated surface, and adaptively correct the virtual heat capacity potential energy index according to the power response deviation; calculate the dynamic loss gradient vector reflecting the adjustment cost under the current operating condition based on the local linearization method. Step S3: Based on the virtual thermal capacity potential energy index and the dynamic loss gradient vector, dynamically reconstruct the benchmark setpoint corresponding to the power grid load command to generate a reconstructed setpoint trajectory that takes into account both response speed and operating economy. Step S4: Establish a hierarchical security constraint system including a physical protection layer, an absolute security layer, and a dynamic optimization layer. Use the augmented state space vector as the initial state input of the state space prediction model, and combine the reconstructed setpoint trajectory and the dynamic loss gradient vector to construct a multi-objective optimization function and a quadratic programming model including an economic penalty term. Step S5: Solve the quadratic programming model to obtain the optimal control increment sequence, perform a dual safety check on the mathematical solution validity and physical boundary of the optimal control increment sequence, and generate the final control command based on the verified effective control increment and the actual measured value of the actuator position.

[0007] Optionally, constructing the augmented state space vector in step S1 includes calculating the model deviation integral term: the current value of this integral term equals its previous value plus the difference between the sensor-measured value and the model-predicted value of the controlled variable at the previous time step. Subsequently, the model deviation integral term is concatenated with the collected unit basic state variables along the vector dimension to form the augmented state space vector. This vector serves as the unified input for subsequent control algorithms, used to compensate for steady-state control deviations caused by model mismatch.

[0008] Optionally, the preprocessing process in step S1 includes: for the heated surface area where no physical measuring points are arranged, estimating the real-time wall temperature using a soft measurement model based on the energy balance principle between the flue gas side and the working fluid side; at the same time, performing time alignment processing on the collected coal feed signal, and correcting the time deviation of the coal feed signal caused by the transmission and actuator action delay by introducing a pure time delay parameter that matches the physical characteristics of the equipment.

[0009] Optionally, calculating the virtual heat capacity potential energy index in step S2 includes: obtaining the measured wall temperature of each heating surface and the steady-state design wall temperature corresponding to the current load command, calculating the weighted deviation between the two and summing them. The adaptive correction is performed using an iterative update law: if the actual power change rate of the unit in the previous control cycle is less than the power change rate predicted by the model, the adaptive correction factor is decreased; otherwise, the adaptive correction factor is increased. This process allows the amplitude of the virtual heat capacity potential energy index to be dynamically adjusted according to the actual response capability.

[0010] Optionally, calculating the dynamic var gradient vector in step S2 includes: constructing a real-time energy efficiency evaluation model that includes the var loss from the cooling water mixing process and the var loss from boiler flue gas heat. At the current unit operating point, a small perturbation is applied to each manipulated variable, and the resulting change in the total var loss rate is calculated. The limit of the ratio of the change in the total var loss rate to the perturbation is the partial derivative of the total var loss rate with respect to that manipulated variable. The partial derivatives corresponding to all manipulated variables together constitute the dynamic var gradient vector.

[0011] Optionally, the dynamic reconstruction in step S3 specifically involves calculating the setpoint correction increment. This increment is a function of the load command change, the virtual thermal capacity potential energy exponent, and the dynamic loss gradient vector. Specifically, the virtual thermal capacity potential energy exponent positively adjusts the magnitude of the correction increment; a larger exponent value increases the rate of change of the setpoint. The norm of the dynamic loss gradient vector negatively adjusts the magnitude of the correction increment; a larger norm value suppresses the rate of change of the setpoint.

[0012] Optionally, the hierarchical safety constraint system in step S4 classifies the constraint conditions: the physical protection layer corresponds to the action threshold of the unit's emergency trip system, serving as the final physical boundary of the system; the absolute safety layer sets the allowable fluctuation range of the controlled variable and the physical travel limit of the manipulated variable, serving as hard constraints in the optimization process; the dynamic optimization layer sets the rate of change limit of the state variable, employing soft constraint processing, that is, introducing slack variables into the optimization model, allowing the rate limit to be exceeded within a preset range in the initial stage of the prediction time domain.

[0013] Optionally, the multi-objective optimization function in step S4 consists of a weighted sum of three parts: the first part is the tracking error term between the predicted value of the controlled variable and the reconstructed setpoint trajectory; the second part is the sum of squares of the changes in the manipulated variable in the prediction time domain, i.e., the control stationarity term; and the third part is an economic penalty term, the value of which is equal to the dot product of the dynamic loss gradient vector and the control increment vector to be solved. When the direction of the control increment is consistent with the gradient direction that leads to an increase in loss, the economic penalty term increases the objective function value, thereby suppressing high-energy-consuming control actions during the optimization process.

[0014] Optionally, the dual safety checks in step S5 include: the first safety check is a mathematical solution validity check, which monitors the status code returned by the quadratic programming solver to determine whether the calculation result is an infeasible or unbounded solution, and checks whether the calculation time exceeds the preset control cycle duration; if the check fails, the current calculation result is determined to be invalid. The second safety check is a physical boundary and rate limit check, which superimposes the solved control increment with the real-time position feedback value of the actuator to determine whether the synthesized command exceeds its physical travel upper and lower limits or maximum action rate limit, and truncates the excess portion.

[0015] Optionally, the generation of the final control command in step S5 adopts an incremental update law based on measured feedback. Specifically, the final control command issued to the actuator at the current moment is equal to the sum of the actual position measurement value of the actuator collected by the sensor feedback at the previous moment and the effective control increment obtained at the current moment after double safety verification. This calculation method is used to eliminate the cumulative error caused by command execution deviation or external interference.

[0016] The above-mentioned method for rapid response and efficiency assurance of deep peak shaving in double reheat units provided by the present invention has the following beneficial technical effects: This invention constructs a virtual thermal capacity potential energy index and implements dynamic reconfiguration of setpoints, utilizing the heat storage state of the heated surface to provide feedforward compensation for load commands. When the unit is detected to have the potential to release thermal potential energy, the algorithm automatically adjusts the rate of change of the setpoint trajectory, transforming the large thermal capacity inertia of the double reheat unit into a regulating assist, effectively improving the load response rate under deep peak shaving conditions, and solving the problem of the lag in tracking grid commands by large-delay thermal systems.

[0017] This invention integrates real-time energy efficiency evaluation directly into the underlying control loop by introducing a dynamic loss gradient vector and a multi-objective optimization control strategy. When solving for the optimal control increment, an economic penalty term is used to suppress the direction of change of the manipulated variable that leads to high losses, achieving a dynamic balance between regulation quality and operating energy consumption. This avoids unnecessary throttling losses or flue gas heat losses caused by the control system simply pursuing regulation speed.

[0018] This invention ensures the dual security of optimized control commands at both the physical and mathematical levels by establishing a hierarchical safety constraint system and executing closed-loop dual verification logic. Combined with an incremental update law based on the measured position feedback of the actuator, it can eliminate accumulated errors caused by model mismatch or equipment nonlinearity in real time, ensuring the numerical stability of the control system during long-term continuous operation and preventing the risk of unplanned unit shutdowns due to command divergence or exceeding limits. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the data acquisition and preprocessing process according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the logic of dual evaluation of unit thermal potential and energy efficiency cost in an embodiment of the present invention; Figure 4 This is a schematic diagram of the dynamic reconfiguration logic of the set value in an embodiment of the present invention; Figure 5 This is a diagram illustrating the hierarchical security constraints and multi-objective optimization control logic of an embodiment of the present invention. Figure 6 This is a flowchart illustrating the closed-loop execution and dual security verification logic of an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] See Figure 1 , Figure 1 This is a schematic flowchart illustrating a method for rapid response and efficiency assurance in deep peak shaving of a double reheat unit according to an embodiment of the present invention. The embodiment of the present invention provides a method for rapid response and efficiency assurance in deep peak shaving of a double reheat unit, comprising the following steps: Step S1: Collect unit operation data in real time and preprocess it, calculate the model deviation integral term of key controlled variables, and construct an augmented state space vector containing basic state variables and integral deviation terms. Step S2: Calculate the virtual heat capacity potential energy index based on the wall temperature deviation of the unit's heated surface, and adaptively correct the virtual heat capacity potential energy index according to the power response deviation; calculate the dynamic loss gradient vector reflecting the adjustment cost under the current operating condition based on the local linearization method. Step S3: Based on the virtual thermal capacity potential energy index and the dynamic loss gradient vector, dynamically reconstruct the set value corresponding to the power grid load command to generate a reconstructed set value trajectory that takes into account both response speed and operating economy. Step S4: Establish a hierarchical security constraint system including a physical protection layer, an absolute security layer, and a dynamic optimization layer. Use the augmented state space vector as the initial state input of the state space prediction model, and combine the reconstructed setpoint trajectory and the dynamic loss gradient vector to construct a multi-objective optimization function and a quadratic programming model including an economic penalty term. Step S5: Solve the quadratic programming model to obtain the optimal control increment sequence, perform a dual safety check on the mathematical solution validity and physical boundary of the optimal control increment sequence, and generate the final control command based on the verified effective control increment and the actual measured value of the actuator position.

[0022] The technical details of each of the above steps will be explained in detail below with reference to specific implementation methods.

[0023] See Figure 2 As shown, the data acquisition and augmented state space construction process described in step S1 includes the following implementation details: During step S1, data is exchanged with the distributed control system via a communication interface program configured on the control server at a preset sampling period. This sampling period is set to the second level to match the thermal inertia characteristics of the unit. A distributed control system (DCS) is a computer-based networked industrial control system. Its core architecture adopts the principle of centralized management and decentralized control. Multiple controllers distributed across the production site control different processes or equipment, while a higher-level operator station performs centralized monitoring and data management, thereby achieving highly reliable automated control of industrial processes.

[0024] Step S11 involves real-time acquisition and verification of multi-source operating data. The data acquisition targets cover all dimensions of unit operation parameters, specifically including: load command signals characterizing grid demand; controlled variable signals characterizing unit response status, such as main steam pressure, main steam temperature, reheat steam temperature, and unit active power; manipulated variable signals characterizing the degree of control intervention, such as coal feed command, water feed command, and desuperheating water regulating valve opening command; and monitoring variable signals characterizing equipment health and thermal status, namely, the metal wall temperature of each stage of the boiler's heating surfaces. For acquiring heating surface wall temperature data, in areas where thermocouple sensors are installed, the measured values ​​are directly read; for heating surface areas without physical measuring points or where the measuring points are damaged, a soft measurement model based on the energy balance principle between the flue gas and working fluid sides is used to calculate the real-time estimated wall temperature by inputting the flue gas flow rate, flue gas temperature, and working fluid flow rate parameters under the current load. During the data acquisition process, the data acquisition system verifies the quality code of all input signals. If a poor signal quality is detected, it automatically retains the valid value from the previous moment or triggers a data anomaly alarm to ensure the validity of the data. The working fluid flow parameter refers to parameters characterizing the mass flow state of water and steam in the boiler and turbine system, specifically including feedwater flow, main steam flow, and reheat steam flow. For example, measuring feedwater flow can reflect the boiler's heat absorption intensity, while main steam flow characterizes the unit's current load level. The quality code is a status identifier corresponding to each data point, used to characterize the real-time performance, validity, communication status, and source reliability of the collected data, thereby determining whether the data can be used as reliable input. The soft measurement model specifically involves: using the known steam temperature, pressure, and mass flow rate data at the inlet and outlet of the heated surface pipe section as input, calculating the enthalpy increase and heat absorption on the working fluid side; establishing a steady-state or dynamic heat balance equation based on the convective and radiative heat transfer mechanisms, where the heat released by the flue gas side to the pipe wall is equal to the heat released by the pipe wall to the working fluid side; and, under the known or estimated temperature distribution boundary conditions on the flue gas side, combining the thermal conductivity and geometric dimensions of the pipe wall material, using a numerical iteration method to solve inversely for the intermediate point pipe wall metal temperature value that satisfies the above heat balance equation.

[0025] Step S12 involves preprocessing the collected raw data to eliminate interference. To address electromagnetic interference and measurement noise present in the industrial environment, an amplitude-limiting filter algorithm is used to remove abnormal jump values ​​exceeding the physical range. Subsequently, a moving average filter algorithm is used to smooth the signal. For signals with large delay characteristics, such as coal feed rate, the preprocessing step also includes time alignment. This aims to compensate for the pure time lag during signal transmission and actuator operation, ensuring that all data input to the control algorithm remains consistent on the time base.

[0026] Step S13: Construct an augmented state space containing the model deviation integral term. To address the steady-state error problem caused by model mismatch in conventional model predictive control, this embodiment introduces an integral state variable that reflects the accumulated error. Specifically, for key controlled variables such as main steam pressure and main steam temperature, their corresponding model deviation integral terms are calculated. The update logic of this integral term is defined by the following formula: ; in: This is the discrete-time index of the current control cycle; Represents any key controlled variable (e.g., main steam pressure or main steam temperature). Controlled variable At the present moment The model bias integral term; Controlled variable In the previous control cycle The cumulative integral value of the model bias at each time step; Controlled variable In the previous moment The actual measured value of the sensor; Controlled variable In the previous moment The predicted value calculated by the state-space prediction model.

[0027] Step S14 generates the final augmented state space vector. The preprocessed basic state variables are concatenated with the model deviation integral term calculated in step S13 to form an augmented state vector containing both the physical state and the deviation state. This augmented state vector includes not only pressure, temperature, and flow data describing the current physical state of the double reheat unit, but also integral data describing the system's historical prediction deviations. This augmented state vector is transmitted as a unified standard input data format to the subsequent virtual thermal potential energy observer and multi-objective optimization controller. In this way, the control system can sense and remember the deviation between historical control effects and actual responses, thereby automatically compensating in subsequent optimization calculations and achieving zero steady-state error tracking of the setpoint. The data communication protocol configuration and the specific implementation of the general filtering algorithm involved in the above steps are well-known to those skilled in the art and will not be elaborated upon here.

[0028] See Figure 3 The dual evaluation process of unit thermal potential and energy efficiency cost described in step S2 specifically includes the following implementation steps: Step S2 consists of two parallel sub-processes: adaptive virtual thermal capacity potential energy observation and dynamic loss gradient calculation. The purpose of this step is to transform the complex nonlinear thermodynamic state of the unit into a quantitative index that can be directly used for a linear controller, thereby solving the problem that traditional control strategies struggle to balance speed and economy.

[0029] Step S21: Perform adaptive virtual heat capacity potential energy observation. The secondary reheat unit control system first calls the unit design data or historical baseline operating data stored in the database, and obtains the steady-state design wall temperature that each stage of the heating surface should reach under the current load command through table lookup or polynomial fitting. The heating surface area covers key heat exchange components such as economizer, water-cooled wall, low-temperature superheater, screen superheater, final stage superheater, and each stage of reheater. The metal heat storage of these components dominates the overall heat storage of the boiler.

[0030] The control system of a double reheat unit calculates the virtual heat capacity potential energy index based on the wall temperature deviation and thermal inertia characteristics of each heated surface. The calculation process for this index is defined by the following formula: ; in: For a moment The virtual thermal capacity potential energy index, the magnitude of which represents the degree and direction of the unit's current deviation from steady-state energy balance; The total number of monitored heat transfer surface nodes, covering all heat transfer surface zones from the furnace outlet to the tail flue; For the first Each heated surface node at time... The measured wall temperature (or the wall temperature estimated by soft measurement) reflects the current actual thermal state; Current load command The corresponding number The steady-state design wall temperature of each heated surface node is used as an energy reference line. For the first The lumped mass heat capacity of each heated surface node is determined by the mass of the heated surface metal and the average specific heat capacity of the working fluid inside the tube. For the first The time weighting coefficient of each heated surface node is determined based on the working fluid flow distance from the node to the turbine inlet. The shorter the flow distance, the more direct the energy release at that point has an impact on the main steam parameters, and therefore the larger the weighting coefficient. For a moment An adaptive correction factor is used to correct theoretical calculations online.

[0031] To compensate for the impact of equipment aging, dust accumulation, and changes in heat dissipation from the insulation layer on model accuracy, the adaptive correction factor... An iterative update law based on power response deviation is adopted: ; in: This is the correction factor from the previous time step; The preset iteration step size gain is used to adjust the balance between correction speed and system stability; and These are the actual power change and the model-predicted power change in the previous control cycle, respectively. The sampling period interval of the control system is used to normalize the power change to the rate of change. The mechanism of this formula is as follows: when the actual power response is slower than the model prediction, it indicates that the actual effective heat storage capacity of the unit is lower than the theoretical value, and the correction factor is automatically reduced, thus making a conservative reduction when calculating the potential energy, and vice versa.

[0032] Step S22: Perform dynamic heat loss gradient calculation based on local linearization. Construct a real-time energy efficiency evaluation model that includes cooling water mixing heat loss and flue gas heat loss. Specifically, the construction method of this real-time energy efficiency evaluation model is as follows: First, set the environmental reference conditions (ambient temperature and ambient pressure) for unit operation.

[0033] To address the mixing entropy loss of the desuperheating water, based on the principle of irreversible processes according to the second law of thermodynamics, the entropy increase during the mixing process of the steam stream and the desuperheating water stream at the desuperheater inlet is calculated. The calculation process is as follows: input the steam flow rate, temperature, and pressure before the desuperheater, and the desuperheating water flow rate and temperature, and calculate the total input entropy rate before mixing; input the steam parameters after mixing, and calculate the output entropy rate; the difference between the two is the irreversible mixing entropy loss caused by the desuperheating water regulation.

[0034] To address the heat loss (e) in flue gas, the physical energy carried by the flue gas is calculated based on the sensible heat of the flue gas. The calculation process is as follows: real-time acquisition of the flue gas temperature and oxygen content at the air preheater outlet (to estimate the flue gas flow rate), and calculation of the flue gas enthalpy and entropy values ​​relative to the environmental reference state based on the specific heat capacity characteristics of the flue gas components, thereby obtaining the effective energy (e) carried away by the flue gas. This portion of energy that is not utilized is considered a loss.

[0035] Finally, the losses from both parts are normalized to form a total energy loss index reflecting the current economic efficiency of the unit. The desuperheating water mixing process is considered a typical irreversible thermodynamic process, and its energy loss directly leads to the loss of work potential. Flue gas heat directly represents the unutilized effective energy emitted into the environment. To meet the computational timeliness requirements of real-time control and avoid complex nonlinear thermodynamic iterative solutions in each control cycle, this embodiment uses a local linearization method to calculate the dynamic energy loss gradient vector at the current operating point. : ; in, for dimensional gradient vector, corresponding to One channel for manipulating variables; The transpose operator for a vector; Let the total entropy loss rate function of the system be defined as follows: That is, the sum of the cooling water mixing loss rate and the flue gas loss rate; For the first A manipulated variable (e.g., the opening degree of a certain stage desuperheating water valve or the amount of fuel); The total entropy loss rate for the first Partial derivatives of each manipulated variable.

[0036] The system utilizes a fast query interface of the unit's thermodynamic property library (e.g., the IAPWS-IF97 standard) to perform numerical differentiation near the current state point and calculate the aforementioned partial derivatives. Specifically, a small virtual perturbation is applied to each manipulated variable, and the resulting change in the total energy loss rate is calculated, thus yielding the partial derivative value. Each element in this gradient vector indicates the direction and magnitude of the additional thermodynamic energy loss resulting from increasing the control input by a unit in the corresponding control channel. For example, a large gradient component value for the secondary desuperheating water valve indicates that opening the valve will cause a significant energy loss. This gradient vector provides direct, linear economic gradient information for subsequent multi-objective optimization, enabling the controller to automatically avoid high-energy-consuming operation paths during the optimization process.

[0037] See Figure 4 , Figure 4 This is a schematic diagram of the setpoint dynamic reconfiguration logic according to an embodiment of the present invention. The setpoint dynamic reconfiguration process based on energy efficiency and potential, described in step S3, transforms rigid load commands into flexible control objectives that balance response speed and operational economy by introducing a correction term reflecting the unit's thermal state. The process specifically includes the following implementation steps: Step S31: Obtain the baseline setpoint trajectory and state boundary. Receive real-time load commands from the power grid dispatch center, as well as the virtual thermal capacity potential index and dynamic efficiency gradient vector output in step S2. Based on the unit's design sliding pressure operating curve and rated design parameters, the system maps the real-time commands to the baseline setpoints of the controlled variables. These controlled variables include at least the main steam pressure and main steam temperature. These baseline setpoints represent the operating parameters that the unit should achieve under ideal conditions to meet load requirements.

[0038] Step S32: Calculate the setpoint correction increment. The system dynamically corrects the rate of change of the baseline setpoint based on the current heat storage state and energy efficiency gradient. The system defines a reconfiguration correction algorithm, which linearly superimposes a potential energy utilization term and a loss suppression term on the baseline load change rate.

[0039] Specifically, the rate of change of the reconstructed setpoint is calculated using the following formula: ; in, For a moment The increment of the setpoint of the reconstructed controlled variable; For a moment Changes in power grid load commands; The load following coefficient converts the load change into the corresponding change in pressure or temperature physical quantities. This load following coefficient is determined by the slope of the tangent line of the unit's sliding pressure operation curve at the current load point. The virtual heat capacity potential energy index is calculated in step S2. When... When the time is right, it indicates sufficient heat storage; this value is positive, causing the setpoint increment to increase, allowing the control system to increase parameters at a greater rate. When the value is negative, it indicates a heat storage deficit. This value is negative, and the setpoint increment is reduced to limit the response rate in order to accumulate energy. The potential energy utilization weighting coefficient is used to adjust the correction strength of the heat storage state to the control target; The Euclidean norm of the dynamic loss gradient vector calculated in step S2 represents the overall energy efficiency cost of the adjustment operation under the current operating conditions. This is the entropy loss suppression weight coefficient, used to suppress changes in the setpoint when the operation cost is too high; This is a sign function, taking values ​​of 1, 0, or -1. This term ensures that the direction of entropy loss suppression is always opposite to the direction of load change; that is, it decreases the setpoint increment when the load is increased and increases the setpoint increment (making its absolute value smaller) when the load is decreased, thereby playing a damping role.

[0040] Step S33: Generate the reconstructed setpoint trajectory within the prediction time domain. The setpoint increment calculated in step S32 is accumulated onto the current actual measurement value to obtain the reconstructed setpoint point at the current moment. Then, using this point as the starting point, extrapolation is performed towards the future prediction time domain. The extrapolation strategy employs the gradient preservation method, assuming that the rate of change of the setpoint within the future prediction time domain remains constant. This results in a continuous reconfiguration setpoint curve on the time axis. This setpoint curve not only reflects load demand but also includes constraints related to the unit's own thermal state. If the unit is currently in a state of high heat storage and low loss gradient, the slope of this curve is greater than the slope of the baseline command mapping, guiding the controller to quickly release heat. If the unit is in a state of low heat storage or high loss gradient, the slope of this curve becomes gentler, guiding the controller to adopt a smooth transition strategy, prioritizing parameter stabilization rather than forcibly tracking the load.

[0041] Step S34: Trajectory Smoothing and Boundary Limiting. To prevent non-physical abrupt changes from being introduced during the reconstruction process, the system performs spline interpolation smoothing on the generated reconstruction setpoint trajectory and saturates and truncates the trajectory amplitude according to the equipment's operational safety regulations (e.g., upper limit of main steam pressure, lower limit of temperature). The processed trajectory is confirmed as the final reference input signal and transmitted to the multi-objective optimization function and quadratic programming model construction stage described in step S4. Through the above steps, this embodiment of the invention transforms heat storage fluctuations and energy efficiency losses into feedforward information for proactive adjustment and control objectives, realizing proactive planning of the control strategy.

[0042] See Figure 5 , Figure 5 This is a hierarchical security constraint and multi-objective optimization control logic diagram according to an embodiment of the present invention. A hierarchical security constraint system including a physical protection layer, an absolute security layer, and a dynamic optimization layer is established. The augmented state space vector is used as the initial state input of the state space prediction model. Combined with the reconstructed setpoint trajectory and the dynamic loss gradient vector, a multi-objective optimization function and a quadratic programming model including an economic penalty term are constructed.

[0043] In practice, a state-space prediction model is first constructed to describe the dynamic characteristics of the double reheat unit. This state-space prediction model is derived from the unit transfer function matrix obtained through system identification, or it can be obtained directly by linearizing the physical mechanism model.

[0044] The state-space prediction model adopts a discrete state-space equation of the following form: ; in, Indicates the next sampling time The augmented state space vector prediction; This refers to the augmented state space vector constructed in step S1, which serves as the initial state input for each control cycle. The control increment sequence to be solved; The predicted output value of the key controlled variable; These represent the corresponding system state matrix, input matrix, and output matrix. The purpose of this predictive model is to iteratively calculate the evolution trajectory of the controlled variable at future time points within the prediction time domain based on the current augmented state space vector, thereby providing a numerical basis for subsequent hierarchical constraint checks and objective function optimization.

[0045] Based on this, perform the following hierarchical constraint settings and optimization construction steps: Step S41: Construct a hierarchical safety constraint system. This hierarchical safety constraint system divides the unit operation constraints into three levels with decreasing priority: the physical protection layer, the absolute safety layer, and the dynamic optimization layer.

[0046] The physical protection layer corresponds to the operating thresholds of the unit protection system (ETS), such as the high main steam pressure protection value and the reheat steam temperature protection value. These values ​​serve as the physical boundaries of the system and are used as the final safety verification benchmark.

[0047] The absolute safety layer defines the operating space that the controller must strictly adhere to. Based on the current operating conditions, the allowable fluctuation range of the controlled variable and the physical travel limits of the manipulated variable are set. For example, the commanded upper limit of the feedwater pump speed is set to 100% of the rated speed; the upper limit of the main steam pressure is set to a safety value lower than the preset margin of the ETS action value. These constraints are treated as hard constraints during the optimization process.

[0048] The dynamic optimization layer is used to handle indicators affecting equipment lifespan or economic efficiency, such as the rate of change of main steam temperature and the operating speed of desuperheating water valves. To prevent the optimization problem from becoming unsolvable due to instantaneous disturbances, this embodiment employs a soft constraint mechanism for this layer. Specifically, slack variables are introduced. Allow the controlled variable to be in the early part of the prediction time domain If the constraint exceeds the soft constraint boundary within a step, a penalty is imposed on the excess amount in the objective function. Furthermore, a constraint shifting strategy is employed, i.e., setting the constraint beforehand... The constraint range of the first step is wider than that of the second step. The scope of constraints extends to the next step.

[0049] Step S42: Construct a multi-objective optimization function based on economic penalties. In this embodiment, the dynamic loss gradient vector calculated in step S2 is introduced into the objective function to construct a comprehensive evaluation index that includes tracking error, control stability, and economic penalties.

[0050] The multi-objective optimization objective function Defined by the following formula: ; in, To predict the length of the time domain; To control the length of the time domain; For a moment Predicted future The vector of controlled variables at each step; For the future generation generated in step S3 The reconstructed setpoint vector of the step; The tracking error weight matrix is ​​used to adjust the control accuracy requirements for different controlled variables; For a moment The future of computation The increment vector of the manipulated variable at each step; The stationarity control weight matrix is ​​used to limit the magnitude of change in the manipulated variables; The dynamic entropy loss gradient vector calculated in step S2; This is an economic penalty term, which is the dot product of two vectors. If the control increment... Direction and entropy loss gradient If the direction is the same (i.e. leading to a decrease in energy efficiency), then the term is positive, increasing the objective function value; if the direction is opposite (i.e. improving energy efficiency), then the term is negative, decreasing the objective function value, thereby driving the optimizer to select the control action with better energy efficiency.

[0051] Step S43: Generate the standard quadratic programming model. The system combines the augmented state-space model constructed in step S1, the hierarchical constraints determined in step S41, and the multi-objective optimization function defined in step S42 to transform the aforementioned control problem into a standard quadratic programming (QP) mathematical model. This transformation process utilizes the recursive relationship of the augmented state equations to predict the output... It is expressed as a linear function of the current state and the future control increment sequence, and then substituted into the objective function. In the process, we obtain information about the control increment sequence. Standard quadratic form: ; in, It is a positive definite Hessian matrix. It is a vector of linear terms containing gradient information. superscript This represents the vector transpose operation. After being encapsulated, the QP model enters the rolling time-domain optimization solution process described in step S5. Through this hierarchical modeling approach, the control system seeks a balance between stable control and optimal energy efficiency within a safe range, while ensuring that safety limits are not violated.

[0052] See Figure 6 , Figure 6 This is a flowchart of closed-loop execution and dual security verification logic according to an embodiment of the present invention. The process described in step S5 transforms the mathematically optimal solution calculated in step S4 into engineering-safe and executable physical control instructions, and completes the closed-loop iteration of the control cycle.

[0053] In step S51, the control system first invokes the embedded numerical optimization solver to iteratively solve the quadratic programming model generated in step S4 using either the effective set method or the interior point method. Under the premise of satisfying all inequality constraints, the numerical optimization solver searches for a sequence of control increments that minimizes the objective function value. If the solution is successful, the system obtains a series of optimal control increments in the future control time domain. Based on the rolling time domain control principle, the system extracts only the first element of this sequence as the control increment to be executed at the current moment. The so-called rolling time-domain control principle refers to the fact that in each control cycle, although the controller calculates the complete optimal control sequence for a future predicted time domain, only the first component of the sequence is implemented in order to compensate for model mismatch and environmental disturbances. At the next sampling time, the initial point of the optimization problem is reset according to the latest measured state, the predicted time domain is shifted one step backward, and the optimization solution is performed again. Through this repeated closed-loop process, the control accuracy is continuously corrected.

[0054] Step S52: Before execution, a first safety check is performed, namely, a mathematical solution validity check. The system monitors the solver's return status code and computation time in real time. If the numerical optimization solver returns an infeasible or unbounded status code, or if the computation time exceeds a preset control cycle threshold (e.g., 1 second), the optimization calculation is deemed to have failed. At this time, the system automatically triggers fault-tolerant protection logic, discards the current calculation result, and invokes a preset fault-safe strategy. The fault-safe strategy includes: maintaining the effective control increment from the previous moment, or forcibly resetting the control increment to zero, i.e., maintaining the current actuator opening unchanged, to prevent abnormal control actions due to numerical calculation divergence.

[0055] In step S53, the system performs a second layer of safety verification, namely, physical boundary and rate limit verification. For the control increment to be executed that has passed the first layer of verification, the system combines the real-time status of the actuator fed back from the DCS system to perform a final physical constraint check. The system calculates the absolute value of the command after applying the increment and determines whether it exceeds the upper and lower limits of the device's physical travel; at the same time, it determines whether the rate of change corresponding to the increment exceeds the mechanical response limit of the actuator. If it exceeds the limit, the system will truncate the increment, forcing it to fall within the allowable safe range boundary, thus obtaining the final effective control increment.

[0056] In step S54, the system generates and issues the absolute control command for the current moment based on the final effective control increment. To eliminate the accumulated errors caused by mechanical dead zones and hysteresis in the actuators, the final command adopts an incremental update law based on measured feedback, the calculation formula of which is as follows: ; in, For a moment The final absolute position command is issued to the execution end of the distributed control system (DCS); For a moment The actual position of the actuator is collected by sensor feedback. Explicitly, measured values ​​are used here, rather than command values ​​or model calculations from the previous moment, to ensure that the control baseline is always calibrated to the true state of the physical equipment. This is the optimal control increment at the current moment after double security checks and amplitude limiting / truncation.

[0057] After issuing the command, the system updates the current measured values ​​of the actual state variables and the final control command to the new historical state, and shifts the prediction time window forward by one control step, entering the next control cycle and repeating the above optimization and control process. For the specific communication protocols and solver underlying code involved in the above process, those skilled in the art can refer to relevant industry standards for implementation, and will not be elaborated upon here.

[0058] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for rapid response and efficiency assurance in deep peak shaving of a double reheat unit, characterized in that, Includes the following steps: Step S1: Collect unit operation data in real time and preprocess it, calculate the model deviation integral term of the key controlled variable, and concatenate the model deviation integral term with the basic state variable to construct an augmented state space vector. Step S2: Calculate the virtual heat capacity potential energy index based on the wall temperature deviation of the unit's heated surface, and adaptively correct the virtual heat capacity potential energy index according to the power response deviation; calculate the dynamic loss gradient vector reflecting the adjustment cost under the current operating condition based on the local linearization method. Step S3: Based on the virtual thermal capacity potential energy index and the dynamic loss gradient vector, dynamically reconstruct the benchmark setpoint corresponding to the power grid load command to generate a reconstructed setpoint trajectory that takes into account both response speed and operating economy. Step S4: Establish a hierarchical security constraint system including a physical protection layer, an absolute security layer, and a dynamic optimization layer. Use the augmented state space vector as the initial state input of the state space prediction model, and combine the reconstructed setpoint trajectory and the dynamic loss gradient vector to construct a multi-objective optimization function and a quadratic programming model including an economic penalty term. Step S5: Solve the quadratic programming model to obtain the optimal control increment sequence, perform a dual safety check on the mathematical solution validity and physical boundary of the optimal control increment sequence, and generate the final control command based on the verified effective control increment and the actual measured value of the actuator position.

2. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S1, the calculation process of the model deviation integral term is as follows: the integral term at the current moment is equal to the integral term at the previous moment plus the difference between the sensor measured value and the predicted value of the controlled variable at the previous moment. The augmented state space vector serves as a unified standard input data format for subsequent state observation and multi-objective optimization calculations.

3. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S2, the virtual heat capacity potential energy index is obtained by calculating the weighted deviation between the measured wall temperature of each stage of the heating surface and the steady-state design wall temperature corresponding to the current load command; the adaptive correction adopts an iterative update law, which includes: when the actual power change rate of the previous control cycle is less than the power change rate predicted by the state space prediction model, the adaptive correction factor is reduced, and vice versa.

4. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S2, the method for calculating the dynamic loss gradient vector includes: constructing a real-time energy efficiency evaluation model that includes cooling water mixing loss and flue gas heat loss; at the current operating point, applying a perturbation of a preset numerical calculation step size to all manipulated variables respectively, calculating the partial derivative of the total loss rate with respect to each manipulated variable through numerical differentiation, and constructing the dynamic loss gradient vector from the partial derivatives.

5. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S3, the dynamic reconstruction process includes: calculating the setpoint correction increment, which is jointly determined by the load command change, the virtual thermal capacity potential energy index, and the dynamic loss gradient vector; wherein, the virtual thermal capacity potential energy index is used to increase the setpoint change rate when the heat storage is sufficient, and the norm of the dynamic loss gradient vector is used to have a reverse inhibition effect on the setpoint change rate when the adjustment cost is too high.

6. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S4, the specific settings of the hierarchical safety constraint system include: the physical protection layer corresponds to the action threshold of the unit protection system, which serves as the final physical boundary of the unit protection system; the absolute safety layer sets the allowable fluctuation range of the controlled variable and the physical travel limit of the manipulated variable, which serve as hard constraints for optimization; the dynamic optimization layer sets the change rate limit of the state variable and adopts a soft constraint processing mechanism, allowing the variable to introduce slack variables in the first few steps of the prediction time domain.

7. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S4, the multi-objective optimization function is composed of a weighted sum of a tracking error term, a control stationarity term, and an economic penalty term; the economic penalty term is the dot product of the dynamic loss gradient vector and the control increment vector; when the direction of the control increment sequence is consistent with the gradient direction that leads to an increase in loss, the economic penalty term increases the objective function value.

8. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S5, the dual safety verification includes: the first safety verification is a mathematical solution validity verification, which monitors whether the solver returns an infeasible solution or an unbounded solution status code, and whether the calculation time exceeds a preset period; the second safety verification is a physical boundary and rate limit verification, which determines whether the control increment to be executed causes the instruction to exceed the upper and lower limits of the physical travel or the action rate limit based on the real-time status of the actuator.

9. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S5, the final control command is generated using an incremental update law based on measured feedback. The calculation formula is: the final control command at the current moment is equal to the measured value of the actual position of the actuator collected by the sensor feedback at the previous moment plus the optimal control increment at the current moment after double safety verification.

10. The method for rapid response and efficiency assurance of deep peak shaving in a double reheat unit according to claim 1, characterized in that, In step S1, the real-time acquisition and preprocessing of unit operation data specifically includes: for the heated surface area where no physical measuring points are arranged, estimating the real-time wall temperature using a soft measurement model based on the energy balance principle between the flue gas side and the working fluid side; and for the coal feed signal in the acquired signals, performing timing alignment processing to compensate for the pure lag time in the signal transmission and actuator operation process.

Images