Dynamic monitoring-based deaerator feed water control method and system for nuclear power station

Abstract

The invention discloses a dynamic monitoring-based deaerator water supply control method and system for a nuclear power plant, and relates to the field of deaerator water supply control, and the method comprises the steps: firstly, through a hybrid online identification algorithm, analyzing a historical valve position instruction and a water supply flow sequence in real time, and dynamically estimating the local gain and hysteresis width of a valve under the current working condition; then, the parameters identified in real time are utilized to carry out prospective compensation on the expected flow variation calculated by the main controller. In other words, the instruction amplitude is adjusted according to the estimated local gain, and it is ensured that consistent flow response can be obtained under different loads; meanwhile, when the instruction is reverse, the compensation amount is actively applied according to the estimated hysteresis width so as to eliminate the adjustment dead zone and oscillation caused by hysteresis. In this way, the nonlinear object of the valve is equivalent to a linear link with consistent response, and the stability and accuracy of the deaerator water level control system in the full working condition range are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of deaerator feedwater control, and more specifically, to a method and system for deaerator feedwater control in nuclear power plants based on dynamic monitoring. Background Technology

[0002] The deaerator in a nuclear power plant is a critical hub connecting the thermal cycle system and the feedwater system. Stable water level control in the deaerator is essential for ensuring stable feedwater supply to the steam generator, maintaining stable unit power output, and even the safety of the entire nuclear power plant. Excessively high water levels can cause water hammer in the turbine, while excessively low water levels can lead to feedwater pump cavitation, both threatening unit safety. Therefore, achieving high-precision and high-stability control of the deaerator water level is a core technical requirement for nuclear power plant operation.

[0003] To achieve precise control of the deaerator water level, existing technologies commonly employ a three-impulse control scheme that combines water level feedback, steam flow feedforward, and feedwater flow feedback. This scheme uses a PID controller to handle water level deviations and introduces the main steam flow, which is the primary load disturbance, as a feedforward signal to improve the system's response speed and disturbance rejection capability. However, this traditional control strategy typically idealizes the final control actuator—the feedwater regulating valve—as a linear, fast-responding actuator. Once the controller parameters are tuned under a typical operating condition, they remain fixed throughout the entire operating range. The fundamental flaw of this control scheme is that it neglects online monitoring and compensation of the valve's actual dynamic characteristics, and cannot adapt to the complex nonlinear behavior exhibited by the valve in actual operation.

[0004] In reality, the actual characteristics of feedwater regulating valves are far from ideal. First, the valve's flow gain changes significantly with the opening degree (i.e., nonlinear flow characteristics). This causes the performance of PID parameters tuned under a certain operating condition to degrade after changes in unit load (such as low or high load), leading to sluggish regulation or overshoot oscillations. Second, due to the influence of mechanical clearance and friction, valves generally exhibit hysteresis. When the controller issues a small reverse regulation command, the valve may not respond at all until the command accumulates enough to overcome the hysteresis width. This easily leads to continuous small oscillations in the system near the setpoint, which not only reduces control accuracy but also accelerates the mechanical wear of the valve. Traditional control methods cannot identify and compensate for these nonlinear characteristics that change with operating conditions and equipment aging online. Therefore, it is difficult to achieve high-performance control of the deaerator water level across the entire operating range. There is an urgent need for a new control method that can dynamically monitor and adaptively compensate for valve nonlinearity to improve the safety and operational economy of nuclear power plants. Summary of the Invention

[0005] To address the problems in the prior art, according to one aspect of this application, a method for controlling deaerator feedwater in nuclear power plants based on dynamic monitoring is provided, comprising: Based on the actual deaerator water level, water level setpoint, main steam flow rate and actual feedwater flow rate, calculate the expected flow rate change of the main controller to obtain the expected feedwater flow rate change. A hybrid online identification of valve characteristic parameters is performed on historical valve position command sequences and historical water supply flow sequences to obtain estimated valve local gain and estimated valve hysteresis width. Based on the estimated local valve gain and the estimated valve hysteresis width, adaptive compensation is performed on the final valve command and the expected change in feedwater flow rate of the previous cycle to obtain an unconstrained compensated valve command. The unconstrained compensated valve commands are subjected to command amplitude limiting and safety verification to obtain the final issued valve commands.

[0006] According to another aspect of this application, a deaerator feedwater control system for nuclear power plants based on dynamic monitoring is provided, comprising: The expected flow change calculation module is used to calculate the expected flow change of the main controller based on the actual deaerator water level, water level setpoint, main steam flow, and actual feedwater flow to obtain the expected feedwater flow change. The parameter estimation module is used to perform online mixed identification of valve characteristic parameters on historical valve position command sequences and historical water supply flow sequences to obtain estimated valve local gain and estimated valve hysteresis width. The adaptive compensation module is used to adaptively compensate the final valve command and expected change in feedwater flow rate of the previous cycle based on the estimated local valve gain and the estimated valve hysteresis width to obtain an unconstrained compensated valve command. The valve command generation module is used to perform command amplitude limiting and safety verification on the unconstrained compensated valve commands to obtain the final issued valve commands.

[0007] Compared with existing technologies, this application provides a dynamic monitoring-based deaerator feedwater control method and system for nuclear power plants. It constructs an adaptive compensation controller based on online valve characteristic identification between the main controller and the feedwater valve actuator. This aims to solve the control performance degradation problem caused by valve nonlinear characteristics (such as gain variation and hysteresis) in the prior art. The implementation involves first using a hybrid online identification algorithm to analyze historical valve position commands and feedwater flow sequences in real time, dynamically estimating the local gain and hysteresis width of the valve under the current operating conditions. Then, using these real-time identified parameters, the expected flow change calculated by the main controller is proactively compensated. That is, the command amplitude is adjusted according to the estimated local gain to ensure a consistent flow response under different loads, thereby overcoming the influence of nonlinear flow characteristics. Simultaneously, when the command reverses, a compensation amount is actively applied based on the estimated hysteresis width to eliminate the dead zone and oscillation caused by hysteresis. In this way, the nonlinear object of the valve is equivalent to a linear element with a consistent response, significantly improving the stability and accuracy of the deaerator water level control system across the entire operating range. Attached Figure Description

[0008] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings.

[0009] Figure 1 This is a flowchart of a deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of data flow in a deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application.

[0011] Figure 3 This is a flowchart of step 2 in the deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application.

[0012] Figure 4 This is a flowchart of step 3 in the deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application.

[0013] Figure 5 This is a flowchart of step 4 in the dynamic monitoring-based deaerator feedwater control method for nuclear power plants according to an embodiment of this application.

[0014] Figure 6 This is a block diagram of a deaerator feedwater control system for a nuclear power plant based on dynamic monitoring, according to an embodiment of this application. Detailed Implementation

[0015] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. It should be understood that the drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0016] In view of the shortcomings in the above-mentioned technical fields, this application proposes a method for controlling deaerator feedwater in nuclear power plants based on dynamic monitoring. Figure 1 This is a flowchart of a deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in a deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application. Figure 1 and Figure 2 As shown in the embodiment of this application, the deaerator feedwater control method for nuclear power plants based on dynamic monitoring includes: Step 1, calculating the expected flow change of the main controller based on the actual deaerator water level, water level setpoint, main steam flow rate, and actual feedwater flow rate to obtain the expected feedwater flow change; Step 2, performing hybrid online identification of valve characteristic parameters on historical valve position command sequences and historical feedwater flow sequences to obtain estimated valve local gain and estimated valve hysteresis width; Step 3, performing adaptive compensation on the final valve command and expected feedwater flow change of the previous cycle based on the estimated valve local gain and estimated valve hysteresis width to obtain an unconstrained compensated valve command; Step 4, performing command limiting and safety verification on the unconstrained compensated valve command to obtain the final issued valve command.

[0017] In step 1, based on the actual deaerator water level, water level setpoint, main steam flow rate, and actual feedwater flow rate, the expected flow rate change of the main controller is calculated to obtain the expected feedwater flow rate change. It should be understood that in the traditional three-impulse control scheme of a nuclear power plant deaerator, the main controller directly outputs commands acting on the feedwater valve. However, due to the inherent nonlinear flow characteristics and hysteresis behavior of the valve, a fixed controller output change will produce drastically different actual flow responses under different operating conditions, leading to deterioration of control quality. To solve this problem, it is necessary to decouple the task of the main controller from the nonlinear compensation task of the actuator. Therefore, this application calculates the expected flow rate change of the main controller to obtain the expected feedwater flow rate change, transforming the main controller's output from an uncertain valve position command strongly correlated with valve characteristics into a clear, physically meaningful expected flow rate adjustment. This expected flow rate change serves as an intermediate target, precisely expressing the specific value that the current feedwater flow rate needs to increase or decrease to maintain water level stability. Subsequent compensation stages can then focus on how to achieve this clear physical target through precise control of the nonlinear valve, thereby isolating the valve nonlinearity from interfering with the main control loop.

[0018] In one feasible scheme, step 1, based on the actual deaerator water level, water level setpoint, main steam flow rate, and actual feedwater flow rate, calculates the expected flow rate change of the main controller to obtain the expected feedwater flow rate change, including: step 11, calculating the difference between the actual deaerator water level and the water level setpoint to obtain the water level deviation for the current control cycle; step 12, inputting the water level deviation for the current control cycle into the PID controller to obtain the feedback flow component; step 13, combining the feedback flow component and the main steam flow rate with the expected total flow rate from the feedforward signal to obtain the expected total feedwater flow rate; and step 14, subtracting the expected total feedwater flow rate from the actual feedwater flow rate to obtain the expected feedwater flow rate change.

[0019] In the above scheme, step 1 can be implemented as follows: To achieve precise and robust control of the deaerator water level, the control logic needs to fully understand the current state of the object and external disturbances. This process is executed within one control cycle, for example, once per second. First, four key real-time process variables need to be acquired. These variables are measured by sensors installed on-site and provided to the control algorithm through the nuclear power plant's distributed control network. The first variable is the actual deaerator water level, measured by a differential pressure or radar level gauge installed on the deaerator. The second variable is the water level setpoint, which is a target value set by operators or upper-level coordination control logic based on the current unit operating conditions. Acquiring these two values ​​is the basis for achieving closed-loop feedback control; their deviation is the fundamental source driving the adjustment action. The third variable is the main steam flow rate, measured by a flow meter installed on the turbine's main steam pipeline, which is the most direct indicator reflecting the unit load. This variable is introduced to achieve feedforward control, so that when the load changes, the feedwater flow can be adjusted in advance and quickly, rather than passively waiting for the water level to deviate before adjustment, such as 2000 tons / hour. The fourth variable is the actual feedwater flow rate, measured by a flow meter installed on the feedwater pipeline at the deaerator inlet, such as 2285 tons / hour. This value is obtained to construct an inner loop targeting flow rate, ensuring that the main controller's output is a clear flow demand rather than an indirect valve position command, thus creating conditions for subsequent compensation of valve nonlinearity.

[0020] At the start of a specific control cycle, step 11 is executed first. For example, a level gauge has an effective measurement range of 0 mm to 8000 mm. Within this control cycle, the obtained water level setpoint and the actual deaerator water level must first be converted into a percentage form relative to this range. Then, the percentage form of the water level setpoint is subtracted from the percentage form of the actual deaerator water level; the difference is the water level deviation for the current control cycle. For example, if the water level setpoint is 5000 mm, its corresponding percentage value is (5000 / 8000) × 100%, which is 62.5%. If the actual deaerator water level measured by the level gauge is 4920 mm, its corresponding percentage value is (4920 / 8000) × 100%, which is 61.5%. Then, the water level deviation for the current control cycle is 62.5% minus 61.5%, resulting in a positive 1.0%. This +1.0% water level deviation value will be used as the output of this step and as the input of the PID controller in the next step for subsequent feedback flow calculation.

[0021] Next, proceed to step 12. First, a set of control parameters needs to be pre-set for the proportional-integral-derivative (PID) controller, including the proportional gain Kp, integral gain Ki, derivative gain Kd, and control period Ts. These parameters are determined based on the analysis of the deaerator water level dynamic characteristics, obtained through engineering tuning methods or model-based optimization calculations to ensure the speed, accuracy, and stability of the control response. For example, the proportional gain Kp can be set to 5000 (tons / hour) / %, the integral gain Ki to 100 (tons / hour) / (%·second), the derivative gain Kd to 200 (tons / hour)·second / %, and the control period Ts to 1 second. Within one control cycle, the controller receives the current water level deviation from the previous stage and reads the water level deviation and the cumulative integral value of the previous cycle from the memory. Continuing the example from the previous stage, the current water level deviation is +1.0%. Simultaneously, the water level deviation of the previous cycle is set to +0.8%, and the cumulative integral value of the previous cycle is 150 tons / hour. Next, the three calculations are performed in parallel. The first term is proportional calculation, which multiplies the current water level deviation by the proportional gain Kp to produce an instantaneous response proportional to the deviation. The result is 5000 (tons / hour) / % × 1.0% = 50 tons / hour. The second term is integral calculation, which adds the cumulative integral value from the previous cycle to the new integral value obtained by multiplying the current water level deviation, the integral gain Ki, and the control period Ts. This step is used to eliminate long-term static errors. The result is 150 tons / hour + 100 (tons / hour) / (%·second) × 1.0% × 1 second, resulting in a new cumulative integral value of 250 tons / hour. The third term is differential calculation, which multiplies the difference between the current water level deviation and the water level deviation from the previous cycle by the differential gain Kd, and then divides by the control period Ts to predict the deviation trend and provide damping to prevent overshoot. The result is 200 (tons / hour)·second / % × (1.0% - 0.8%) / 1 second, resulting in 40 tons / hour. Finally, the results of the proportional term calculation, the updated integral cumulative value, and the differential term calculation are algebraically summed to obtain the total feedback flow component. In this example, the feedback flow component is 50 tons / hour + 250 tons / hour + 40 tons / hour, with a final result of 340 tons / hour. This value of 340 tons / hour is the output of this stage, which will be passed to the next stage for synthesizing the desired total feedwater flow. Simultaneously, the water level deviation for the current cycle (+1.0%) and the new integral cumulative value (250 tons / hour) will be stored for use in the next control cycle.

[0022] Then, step 13 is executed. In one feasible embodiment, step 13, which involves combining the feedback flow component and the main steam flow with the feedforward signal to obtain the desired total flow rate, includes: combining the feedback flow component and the main steam flow with the feedforward signal to obtain the desired total feedwater flow rate using the following formula: , ;in, This is the feedforward gain coefficient. Main steam flow rate, To provide feedback on flow components, For feedforward flow, The desired total feedwater flow rate is determined. This stage receives two inputs: the feedback flow component calculated in the previous stage and the main steam flow rate collected in real time from the main steam pipeline flow meter. A pre-set parameter, the feedforward gain coefficient, is also required. This coefficient is set based on the principle of mass conservation in the deaerator. Theoretically, to maintain a stable water level, the inflow of feedwater should be approximately equal to the outflow of steam. Therefore, The theoretical value is close to 1.0. It can be determined based on thermodynamic calculations during the design phase and fine-tuned during unit commissioning to most accurately reflect the static relationship between feedwater and steam flow. For example, the feedforward gain coefficient can be set. The value is 0.98. During implementation, first, according to the formula... Calculate the feedforward flow rate. This calculation multiplies the currently measured main steam flow rate by a preset feedforward gain factor. If the currently measured main steam flow rate is 2000 tons / hour, then the calculated feedforward flow rate is 0.98 multiplied by 2000 tons / hour, resulting in 1960 tons / hour. This value represents the basic feedwater flow rate required to meet the current unit load. Subsequently, according to the formula... Total flow rate synthesis is performed. This calculation algebraically sums the feedforward flow rate obtained in the previous step with the feedback flow rate component from the PID controller. The logic behind this synthesis method is that the feedforward flow rate provides most of the rapid adjustment to cope with major load disturbances, while the feedback flow rate component is used for fine-tuning to compensate for the inaccuracies of the feedforward model and the impact of other unmeasured disturbances on the water level. Specifically, the calculated feedforward flow rate of 1960 tons / hour is added to the feedback flow rate component of 340 tons / hour obtained in the previous step. The final expected total feedwater flow rate is 1960 tons / hour plus 340 tons / hour, which is 2300 tons / hour. This value of 2300 tons / hour represents the theoretically achievable target total feedwater flow rate within the current control cycle, taking into account both load feedforward and water level feedback.

[0023] Finally, proceed to step 14. The desired total feedwater flow rate is an absolute quantity, while the subsequent adaptive compensation control requires a specific, incremental adjustment command to calculate the specific valve action. Therefore, it is necessary to calculate the desired change in feedwater flow rate. Specifically, at the current moment, the actual feedwater flow rate measured by the feedwater flow meter is 2285 tons / hour. Subtracting the desired total feedwater flow rate from the actual feedwater flow rate (2300 tons / hour - 2285 tons / hour) yields a result of +15 tons / hour. This +15 tons / hour value is the desired change in feedwater flow rate; a positive value indicates a net increase in feedwater flow rate, while a negative value indicates a net decrease in feedwater flow rate.

[0024] In step 2, a hybrid online identification of valve characteristic parameters is performed on the historical valve position command sequence and the historical feedwater flow sequence to obtain the estimated local valve gain and the estimated valve hysteresis width. Accordingly, the background technology has clearly pointed out that the nonlinear characteristics of feedwater regulating valves, such as flow gain and hysteresis, are dynamically changing, altering with valve opening, media conditions, and equipment aging. Using a fixed set of parameters to compensate for these nonlinearities cannot achieve ideal results across the entire operating range. To achieve truly effective adaptive compensation, the control logic needs to be able to perceive the valve's current dynamic characteristics accurately and in real time. Therefore, hybrid online identification of valve characteristic parameters can establish a continuously operating online detection mechanism. This mechanism dynamically extracts key parameters that quantify the valve's current nonlinear behavior by continuously analyzing historical data between valve commands (cause) and feedwater flow (effect), providing accurate and real-time basis for subsequent compensation algorithms.

[0025] In one feasible solution, Figure 3 This is a flowchart of step 2 in the dynamic monitoring-based deaerator feedwater control method for nuclear power plants according to an embodiment of this application. Figure 3 As shown, step 2, which involves performing hybrid online identification of valve characteristic parameters on the historical valve position command sequence and the historical water supply flow sequence to obtain the estimated valve local gain and the estimated valve hysteresis width, includes: step 21, based on the system pure time delay, performing dynamic data vector construction and time delay alignment on the historical valve position command sequence and the historical water supply flow sequence to obtain the current flow change, the time delay aligned command change, the command change direction vector, and the unaligned original command change; step 22, performing recursive least squares estimation of the valve local gain based on the current flow change and the time delay aligned command change to obtain the estimated valve local gain; and step 23, performing event-driven identification and filtering of the valve hysteresis width on the command change direction vector and the unaligned original command change to obtain the estimated valve hysteresis width.

[0026] In the above scheme, step 2 can be implemented as follows: Before implementing this step, three key inputs need to be defined. First is the historical valve position command sequence, a time-series data buffer consisting of the command values ​​(expressed as percentages) ultimately issued by the control logic to the valve positioner over multiple past control cycles. Second is the historical feedwater flow sequence, a time-series buffer consisting of the actual flow values ​​(expressed as tons per hour) measured by the feedwater flow meter over multiple past cycles. Both sequences are recorded and updated by the control logic at the end of each cycle. Third is the system pure time delay, a predetermined constant representing the time delay between a change in valve command and the start of a feedwater flow response, expressed as an integer multiple of the control cycle. This value is mainly determined by the response speed of the valve actuator and the transport delay of the fluid in the pipeline, and can be accurately measured through step response testing during unit commissioning. For example, the system pure time delay can be preset to 3 control cycles.

[0027] First, step 21 is executed. This step involves three parallel calculations. Within a control cycle, for example, at time t, the pure time delay d is three cycles. The control logic needs to retrieve stored historical data, such as the valve position command sequence from time t-5 to t-1, and the water supply flow rates at times t-1 and t. Let's define a specific scenario: the historical valve position command sequence is [..., 52.0%, 52.2%, 52.3%, 52.3%, 52.5%], corresponding to the commands from time t-5 to t-1; the historical water supply flow rate is 2278 tons / hour at time t-1 and 2285 tons / hour at time t. The first calculation is to obtain the current flow rate change. This calculation subtracts the actual water supply flow rate of the previous cycle (time t-1) from the actual water supply flow rate at the current time t. This difference represents the actual flow response observed within the current cycle. Specifically, the current flow rate change is 2285 tons / hour - 2278 tons / hour, resulting in +7 tons / hour. The second calculation involves constructing the command change after time-delay alignment. To establish the correct causal relationship, the root cause of the aforementioned +7 tons / hour flow rate change needs to be found, namely, the change in valve commands. Considering the pure lag of three cycles, the flow rate change observed in the current cycle is caused by command changes that occurred three cycles prior. Therefore, it is necessary to trace back to the command values ​​at times td-1 and td-2. In this example, this means the commands at times t-4 and t-5. Calculating the command at time t-4 minus the command at time t-5 yields 52.2% - 52.0%, resulting in +0.2%. This +0.2% command change is a causal pair that is correctly aligned in time with the +7 tons / hour flow rate change. The third calculation involves extracting auxiliary features for hysteresis identification. This calculation does not consider time delay but focuses on the most recent command dynamics. First, the misaligned original command change is calculated, i.e., the command at time t-1 minus the command at time t-2. Based on the example data, this value is 52.5%-52.3%, resulting in +0.2%. This value is used to accumulate the width when hysteresis occurs. Next, a command change direction vector is constructed, consisting of the sign of the current unaligned command change and the sign of the unaligned command change from the previous cycle. To calculate this vector, the command changes at times t-2 and t-3 are also needed, i.e., 52.3%-52.3%, which results in 0. Therefore, the sign of the current unaligned command change is positive (+1), and the sign of the previous cycle is zero (0). The final command change direction vector is [+1, 0]. This vector is used to determine whether the valve command has reversed direction, which is a key event triggering hysteresis width identification. The final output includes four quantities: the current flow rate change (+7 tons / hour), the command change after time-delay alignment (+0.2%), the command change direction vector ([+1, 0]), and the unaligned original command change (+0.2%).

[0028] Next, proceed to step 22. Here, the Recursive Least Squares (RLS) algorithm with a forgetting factor is used. The core of this algorithm is to establish a simple linear model to describe the relationship between command changes and flow rate changes, i.e., y(t) = θ * Φ(t), where θ is the local gain of the valve to be estimated. The RLS algorithm recursively corrects the estimate of θ using new measurement data in each control cycle, continuously approximating the true value. Initialization is required before implementation. This includes setting the initial state and parameters of the algorithm. First, the parameter estimation vector θ(t-1) from the previous cycle can be set to an initial guess based on the valve design manual or experience during the initial startup. Second, the covariance matrix P(t-1) from the previous cycle represents the degree of uncertainty regarding the initial parameter estimates; it should be set to a large positive number during the initial startup to allow the algorithm to make rapid adjustments in the early stages. Finally, there is the forgetting factor λ, a constant between 0 and 1, used to adjust the algorithm's memory length of historical data. The closer the value is to 1, the greater the weight given to historical data, resulting in a smoother estimation but a slower response to changes. It is set based on a trade-off between noise level and parameter change rate. For example, the initial parameter estimate θ(0) can be set to 30 (tons / hour) / %, the initial covariance P(0) to 1000, and the forgetting factor λ to 0.98. Within a control cycle, this stage receives the current flow change, denoted as y(t), and the time-delay aligned command change, denoted as Φ(t), from the previous stage. Simultaneously, it reads the updated parameter estimate vector θ(t-1) and covariance matrix P(t-1) from internal storage. Specifically, the input current flow change y(t) is +7 tons / hour, and the time-delay aligned command change Φ(t) is +0.2%. After several cycles of operation, the parameter estimate θ(t-1) from the previous cycle has been updated to 32.5 (tons / hour) / %, and the covariance P(t-1) has been updated to 50. This step involves four calculation steps. The first step is to calculate the prediction error. Using the model parameters θ(t-1) from the previous cycle and the current input, the time-delay aligned command change Φ(t) is used to predict the expected flow rate change, and this is compared with the actual observed flow rate change y(t). The prediction error e(t) is calculated as e(t) = y(t) - Φ(t) * θ(t-1). Based on the example data, the prediction error is 7 - (0.2 * 32.5) = 7 - 6.5 = +0.5 tons / hour. This positive error indicates that the gain estimate from the previous cycle was underestimated. The second step is to calculate the gain vector. The gain vector K(t) determines the extent to which the prediction error is used to correct the parameter estimate. The calculation formula is K(t)=P(t-1)*Φ(t) / (λ+Φ(t)*P(t-1)*Φ(t)). Based on the example data, the gain vector is (50*0.2) / (0.98+0.2*50*0.2)≈3.356. The third step is to update the parameter estimates.The new parameter estimate θ(t) is based on the previous period's estimate, with the addition of a correction term determined by the prediction error and the gain vector. Its calculation formula is θ(t) = θ(t-1) + K(t) * e(t). Specifically, the updated parameter estimate is 32.5 + 3.356 * 0.5 = 34.178 (tons / hour) / %. The fourth step is to update the covariance matrix. This step updates the uncertainty measure of the parameter estimate, preparing for the calculation in the next period. Its calculation formula is P(t) = (1 - K(t) * Φ(t)) * P(t-1) / λ. Specifically, the updated covariance matrix is ​​(1 - 3.356 * 0.2) * 50 / 0.98 ≈ 16.77. Finally, gain extraction is performed. Since the parameter to be estimated in this application is a scalar, namely the valve local gain, the updated parameter estimation vector θ(t) is directly assigned to the output of the current cycle, i.e., the estimated valve local gain. Therefore, the final output estimated valve local gain is 34.178 (tons / hour) / %. At the same time, the updated parameter estimate 34.178 and covariance 16.77 will be stored as θ(t-1) and P(t-1) for the next control cycle.

[0029] In particular, the forgetting factor λ plays a crucial role in the memory length during recursive least squares estimation. Because a fixed forgetting factor λ acts as the memory length in the algorithm, the closer λ is to 1, the longer the algorithm's memory is, and the slower the weighted decay of historical data. This makes the parameter estimation insensitive to measurement noise and exhibits good stability. However, when the valve's local gain changes rapidly due to real variations in operating conditions, a long memory slows down the algorithm's convergence speed, resulting in poor tracking ability, i.e., poor adaptability. Conversely, if λ is small, the algorithm's memory is short, allowing it to quickly forget old data and rapidly track time-varying parameters, exhibiting good adaptability. However, when the system is stable and only noise interference exists, it will overreact to noise, leading to drastic fluctuations in the parameter estimation results and poor stability. Therefore, using a fixed λ is essentially a static trade-off between stability and adaptability, which cannot achieve optimal performance under all operating conditions. Therefore, this application overcomes the inherent limitations of a fixed forgetting factor by using adaptive variable forgetting factor recursive least squares estimation based on prediction error, based on the current flow rate change and the command change after time delay alignment. In other words, by enabling the forgetting factor to be dynamically adjusted according to the algorithm's own prediction error, the estimation algorithm can intelligently switch between stability and adaptability automatically. This allows it to exhibit high stability when valve parameters are stable to suppress noise, and high adaptability when valve parameters change to achieve fast tracking. Ultimately, this yields a more accurate and robust estimate of the valve's local gain across the entire operating range.

[0030] Based on this, in a feasible preferred embodiment, step 22, which involves performing an adaptive variable forgetting factor recursive least squares estimation based on the current flow rate change and the command change after time delay alignment to obtain the estimated valve local gain, includes: Based on the current flow rate change and the command change after time delay alignment, the prediction error is calculated. First, the prediction error is calculated to quantify the current parameter model's ability to interpret new data. The formula is: e(t) = y(t) - Φ(t) * θ(t-1). Here, y(t) is the current flow rate change, Φ(t) is the command change after time delay alignment, and θ(t-1) is the gain estimate from the previous cycle. The calculated prediction error e(t) is the core basis for subsequent adaptive adjustments. When the valve parameters remain unchanged and the model is accurate, e(t) mainly consists of measurement noise, and its amplitude is usually small and random. When the valve parameters undergo abrupt changes or drift, the old model θ(t-1) cannot accurately predict the new behavior, leading to a significant increase in the amplitude of e(t). Specifically, this process is the same as the prediction error calculation process in step 22 of the aforementioned feasible scheme.

[0031] Based on an exponentially decaying function, the prediction error is mapped to an adaptive variable forgetting factor. Then, the prediction error e(t) obtained in the previous step is intelligently converted into a forgetting factor λ(t) that guides the algorithm's behavior. The specific implementation uses the following formula: This formula requires pre-setting two parameters, which are typically tuned jointly based on engineering experience and simulation testing to balance the algorithm's tracking speed and noise immunity. (Forgetting factor lower limit) For example, setting it to 0.95 ensures the algorithm has the fastest tracking speed when parameters change drastically; and a positive tuning parameter. Used for control Sensitivity to prediction error e(t), The larger, From 1 down to The faster the speed, for example, set to 0.1. This way, when the prediction error e(t) is close to 0, the exponential term... Approaching 1, making The exponent term is close to 1, thus enhancing the stability of the algorithm; when e(t) is large, the exponent term is close to 0, making... close to This enhances the algorithm's adaptability and allows for rapid tracking of parameter changes. Therefore, in subsequent recursive least squares iterative updates, the fixed λ is replaced with a dynamically calculated λ. This allows the algorithm to automatically make a dynamic trade-off between stability and adaptability based on real-time data.

[0032] Based on the covariance matrix of the previous cycle, the instruction change after time delay alignment, and the adaptive variable forgetting factor, the gain vector is calculated. Then, the step size and direction of the current parameter update are calculated. The implementation process uses the standard gain vector calculation formula of the recursive least squares algorithm, which is the same as the gain vector calculation process in step 22 of the feasible scheme mentioned above. However, the key difference is that the fixed forgetting factor λ in the gain vector calculation formula of step 22 in the feasible scheme is replaced by the dynamic forgetting factor from the previous step. Calculated adaptive variable forgetting factor That is, K(t) = P(t-1) * Φ(t) / ( +Φ(t)*P(t-1)*Φ(t)). That is, when When the gain is smaller (indicating a need for rapid adaptation), the calculated gain vector will be larger, resulting in a larger subsequent parameter correction; conversely, when... When the value is close to 1 (indicating the need for stability), the gain vector will be smaller, making parameter adjustments smoother.

[0033] The gain estimate from the previous cycle is updated based on the prediction error and the gain vector to obtain the updated parameter estimate, which is then used as the estimated local valve gain. Finally, the gain estimate θ(t-1) from the previous cycle is supplemented with a correction term consisting of the product of the prediction error e(t) and the gain vector K(t) to obtain the updated parameter estimate θ(t) for the current cycle. This updated parameter estimate θ(t) is used as the estimated local valve gain for the final output of this cycle. This adaptive mechanism also brings additional robustness gains. Firstly, during the stable period when the system input is insufficient (i.e., the change in command Φ(t) after delay alignment is small), a fixed λ<1 will cause the covariance matrix to increase unnecessarily, making the algorithm exceptionally sensitive to any small future perturbations. By adjusting the λ<1 during this period... Setting it to 1 freezes the growth of the covariance matrix, preventing its expansion and thus enhancing robustness. Secondly, even if the preset pure time delay d has a small deviation, causing Φ(t) and y(t) to not be perfectly matched, the resulting e(t) will be relatively small during the system's stable period. When the value is close to 1, the algorithm will not overreact to such small model mismatches. Only when e(t) shows a sustained and significant increase (which is a stronger signal of parameter change) will the value decrease significantly. This accelerates tracking, thereby indirectly filtering out some of the noise effects caused by the inaccuracy of time delay d, making it insensitive to errors in time delay d. Ultimately, a more accurate estimate of the valve's local gain is obtained, which effectively suppresses noise and responds quickly to changes. Specifically, this process is identical to the parameter estimation process updated in step 22 of the aforementioned feasible scheme.

[0034] Finally, step 23 is executed. This step is implemented using an event-driven identification logic based on a state machine. It triggers, executes, and terminates a hysteresis width measurement process by monitoring the dynamic behavior of the instruction sequence. Before implementation, two key parameters need to be pre-set. The first is the flow response noise threshold, a small positive value used to distinguish the actual flow response from measurement noise. Its setting should be slightly larger than the fluctuation amplitude of the flow signal during normal operation, and can be determined by analyzing historical data or field testing. The second is the low-pass filter coefficient α, a decimal between 0 and 1, used to smooth each measured hysteresis width value to filter out random errors in single measurements and obtain more stable estimation results. The smaller α is, the stronger the filtering effect and the more stable the estimation value, but the slower the response to actual changes in hysteresis width. For example, the flow response noise threshold can be set to 0.5 tons / hour, and the low-pass filter coefficient α to 0.1. This step also needs to maintain three internal state variables, which transmit information between different control cycles. The first is a detection status flag, used to indicate whether a hysteresis width detection process is currently underway. The second is the instruction accumulation value, used to accumulate the reversed instruction change during the detection process. The third is the hysteresis width estimate from the previous cycle. Upon initial startup, the detection status flag is initialized to "No," the instruction accumulation value is initialized to 0, and the hysteresis width estimate can be set to an empirically based initial value, such as 0.1%. Within a control cycle, this step receives three inputs from step 21: the instruction change direction vector, the current flow change, and the misaligned original instruction change. Its logic is divided into three cases based on the detection status flag. The first case: trigger judgment in the non-detection state. If the detection status of the previous cycle is "No," then it first determines whether the instruction has reversed. This is achieved by checking the instruction change direction vector; a typical reversal event is when the vector changes from [+1,...] to [-1,+1] or from [-1,...] to [+1,-1]. Specifically, at time t, the instruction change direction vector is [+1,0], which does not constitute a reversal, therefore the logic is not triggered. At this point, the detection status flag is still negative, the accumulated command value remains at 0, and the estimated hysteresis width for this cycle is equal to the value of the previous cycle. To demonstrate the complete process, at time t+1, due to changes in control requirements, the valve command begins to decrease, resulting in a misaligned command change of -0.05% at time t+1, compared to +0.2% at time t. At this time, the command change direction vector at time t+1 becomes [-1, +1], satisfying the reverse trigger condition. The logic is triggered, entering the detection process: setting the detection status flag to positive and clearing the accumulated command value to zero. The second case: accumulation and judgment during the detection process. At time t+2, since the detection status of the previous cycle was positive, the logic enters this branch. For example, if the misaligned command change at time t+2 is -0.1%, and the current flow rate change is -0.2 tons / hour.First, the current command change is added to the command accumulation value, changing it from 0 to -0.1%. Then, it is determined whether the absolute value of the current flow rate change (0.2 tons / hour) is greater than the preset noise threshold (0.5 tons / hour). In this example, 0.2 is less than 0.5, indicating that the flow rate has not yet shown an effective response, and the hysteresis has not been completely overcome. Therefore, the detection status remains "yes" for this cycle, the command accumulation value is updated to -0.1%, and the hysteresis width estimate remains unchanged. At time t+3, the detection status is still "yes". If the misaligned command change at time t+3 is -0.15%, and the current flow rate change is -1.2 tons / hour, first, the command change continues to accumulate, and the command accumulation value changes from -0.1% to -0.1% + (-0.15%) = -0.25%. Next, it is determined whether the absolute value of the current flow rate change (1.2 tons / hour) is greater than the noise threshold (0.5 tons / hour). At this time, 1.2 is greater than 0.5, indicating that the hysteresis has been overcome, and the valve has begun to operate effectively. At this moment, a successful measurement is completed. The measured original hysteresis width is the absolute value of the current instruction accumulation value, which is 0.25%. The third case: Filtering update and reset after measurement. After the measurement is completed at time t+3, the final hysteresis width estimate is immediately updated by filtering. If the hysteresis width estimate of the previous cycle is 0.1%, then the new estimate is calculated using the first-order low-pass filter formula: New estimate = α * current measurement value + (1-α) * previous cycle estimate. Substituting the data, we get: 0.1 * 0.25% + (1-0.1) * 0.1% = 0.115%. Finally, the detection status flag is reset to "No" to prepare for the capture of the next instruction reverse event. Ultimately, in cycle t+3, the estimated valve hysteresis width output by this step is 0.115%, and the internal state is updated: the detection status flag is "No," and the instruction accumulation value is logically cleared to zero.

[0035] In step 3, based on the estimated valve local gain and estimated valve hysteresis width, adaptive compensation is performed on the final valve command and expected feedwater flow change of the previous cycle to obtain an unconstrained compensated valve command. It is understandable that the expected feedwater flow change calculated by the main controller is a purely process objective, while the response of the valve actuator is a complex nonlinear process. If this flow target is directly and linearly mapped to a valve position command, deviations will occur due to the varying gain and inherent hysteresis characteristics of the valve under different operating conditions, leading to a discrepancy between the actual flow response and the expected result. To accurately translate the idealized adjustment intention of the main controller into actual commands that can drive the non-ideal actuator to produce the expected physical effect, an intermediate intelligent conversion layer needs to be established. Therefore, adaptive compensation is performed on the final valve command and expected feedwater flow change of the previous cycle. This utilizes the real-time dynamic characteristics of the valve identified online to proactively pre-distort the original control requirements, thereby actively offsetting the nonlinear effects of the valve and ensuring that the final command enables the valve to produce an actual flow response that matches the expected feedwater flow change.

[0036] In one feasible solution, Figure 4 This is a flowchart of step 3 in the deaerator feedwater control method for nuclear power plants based on dynamic monitoring, according to an embodiment of this application. Figure 4 As shown, step 3, based on the estimated valve local gain and the estimated valve hysteresis width, adaptively compensates the final valve command and the expected change in feedwater flow rate of the previous cycle to obtain an unconstrained compensated valve command, including: step 31, calculating the basic change based on the estimated valve local gain and the expected change in feedwater flow rate; step 32, calculating the hysteresis compensation based on the estimated valve hysteresis width and the basic change; step 33, adding the basic change, the hysteresis compensation, and the final valve command of the previous cycle to obtain the unconstrained compensated valve command.

[0037] In the above scheme, step 3 can be implemented as follows: First, step 31 is executed to address the issue of nonlinear flow characteristics of the valve, i.e., the problem of valve gain varying with opening degree. The core of this is performing an inverse gain calculation, based on the formula ΔCmd_base(t) = desired change in feedwater flow / estimated local valve gain. The physical meaning of this formula is that it answers the question of how much valve position command change is needed to achieve the target flow change under the current valve sensitivity (local gain). By dividing the desired flow change by the real-time identified local gain, a theoretically sufficient basic valve position command change to produce that flow change can be calculated. Specifically, the input desired feedwater flow change is +15 tons / hour, and the input estimated local valve gain is 34.178 (tons / hour) / %. The calculated basic change is +15 / 34.178 ≈ +0.439%. This +0.439% value is the core command change required to achieve the target flow after gain compensation.

[0038] Next, proceed to step 32. This step is specifically designed to address valve hysteresis. In one feasible solution, step 32 involves calculating the hysteresis compensation based on the estimated valve hysteresis width and the basic change, including: determining whether the direction of the basic change is opposite to the direction of the command change in the previous cycle; if so, calculating the hysteresis compensation based on the estimated valve hysteresis width and the basic change using the following formula: ;in, For symbolic functions, Based on the change, To estimate the valve hysteresis width, Is it the hysteresis compensation amount? If not, set the hysteresis compensation amount to zero.

[0039] The core of this step is an event-driven judgment logic: First, it determines whether the direction of the basic change is opposite to the direction of the command change in the previous cycle. If the directions are the same or the command in the previous cycle did not change, it means that the valve will continue to move in the original direction and will not encounter any hysteresis or invalid stroke, so the hysteresis compensation is set to zero. Compensation is only required when the command reverses. Specifically, the currently calculated basic change is +0.439%, and the direction is positive. The command change in the previous cycle (time t-1 relative to time t-2) is +0.2%, and the direction is also positive. Since the directions are the same and no reversal has occurred, the hysteresis compensation is calculated to be 0 in this example. To fully illustrate this logic, consider a different scenario: if the command change in the previous cycle is negative (e.g., -0.1%), while the currently calculated basic change is positive (+0.439%), then a command reversal is detected. At this time, hysteresis compensation calculation will be initiated, based on the formula... .in, It represents the sign of the basic change, which is +1 in this scenario. The estimated valve hysteresis width is calculated to be 0.115%. Therefore, the calculated hysteresis compensation will be +1 * 0.115% = +0.115%. The significance of this compensation is that it proactively provides an additional push command equal to the hysteresis width at the instant the command reverses, quickly eliminating the mechanical backlash and allowing the valve to immediately begin an effective response, thereby eliminating the regulation delay and oscillations near the setpoint caused by hysteresis. In the main example of this process, since no reversal occurs, the hysteresis compensation is 0.

[0040] Finally, step 33 is executed. This is the final instruction synthesis step. It uses the final valve instruction of the previous cycle as a reference point, and then superimposes the base change amount after gain compensation and the compensation amount used to overcome hysteresis. The calculation formula is: Unconstrained compensated valve instruction = Final instruction of the previous cycle + Base change amount + Hysteresis compensation amount. The final valve instruction of the previous cycle (at time t-1) is 52.5%. Substituting all values ​​into the formula, the calculation is: Unconstrained compensated valve instruction = 52.5% + 0.439% + 0% = 52.939%. This value of 52.939% is the valve instruction target value that, theoretically, can accurately achieve a flow rate increment of +15 tons / hour after comprehensively considering gain compensation and hysteresis compensation.

[0041] In step 4, the unconstrained compensated valve commands are subjected to command limiting and safety verification to obtain the final issued valve commands. It should be understood that the unconstrained compensated valve commands are an idealized result derived from algorithms and real-time identification parameters. However, in actual industrial control environments, any command directly acting on physical equipment must be strictly constrained to ensure equipment safety and process stability. On the one hand, valve actuators themselves have physical stroke limitations and mechanical speed limitations; any command exceeding these limitations is invalid and may cause damage. On the other hand, the reliability of the estimation results of the online identification algorithm, which is the cornerstone of the entire adaptive compensation logic, needs to be continuously monitored. Under abnormal operating conditions or sensor noise interference, the identification results may deviate unreasonably. Directly adopting such erroneous parameters for compensation would severely undermine control stability. Therefore, command limiting and safety verification of the unconstrained compensated valve commands ensure that every command output to the field is physically feasible, process-smooth, and that the adaptive logic it relies on is in a reliable and reasonable operating state, thereby guaranteeing the safety and robustness of the entire advanced control strategy.

[0042] In one feasible solution, Figure 5 This is a flowchart of step 4 in the dynamic monitoring-based deaerator feedwater control method for nuclear power plants according to an embodiment of this application. Figure 5 As shown, step 4, performing instruction amplitude limiting and safety verification on the unconstrained compensated valve instruction to obtain the final issued valve instruction, includes: step 41, performing amplitude and rate limiting on the unconstrained compensated valve instruction to obtain the valve instruction to be issued; step 42, determining whether the estimated valve local gain and the estimated valve hysteresis width are within a reasonable range, if so, determining the valve instruction to be issued as the final issued valve instruction.

[0043] In the above scheme, step 4 can be implemented as follows: Before implementation, four boundary parameters need to be pre-set. These parameters are all determined based on the valve's design specifications and process safety requirements. They include: the upper limit of the command amplitude (100%, representing the valve fully open), the lower limit of the amplitude (0%, representing the valve fully closed), and the command change rate limit (for example, set to no more than 2.0% per second), to prevent excessively rapid valve action from causing pipeline pressure surges or mechanical stress damage to the valve itself.

[0044] First, step 41 is executed. This step receives the unconstrained, compensated valve command output from the previous step as input. The input unconstrained command is 52.939%. The implementation process includes two verification steps. The first step is amplitude verification. The input 52.939% is compared with the upper and lower amplitude limits (0% and 100%). Since 52.939% is within this range, the amplitude verification passes. If the calculated command exceeds this range, for example, 102%, it will be forcibly limited to the upper limit of 100%. The second step is rate verification. The difference between the current command and the final command issued in the previous cycle is calculated. Specifically, the final command in the previous cycle is 52.5%. Therefore, the command change is 52.939% - 52.5% = +0.439%. The absolute value of this change (0.439%) is compared with the preset rate limit (2.0%). Since 0.439% is less than 2.0%, the rate verification passes. If the calculated change exceeds the limit, for example, +3.0%, then the current instruction will be limited to the instruction of the previous cycle plus the rate limit, i.e., 52.5% + 2.0% = 54.5%. In this example, since both the amplitude and rate checks pass, the valve instruction to be issued output in this step is 52.939%.

[0045] Next, proceed to step 42. This step is a health check of the adaptive algorithm itself, designed to prevent erroneous compensation based on unreliable identification results. Before implementation, a reasonable range of values ​​needs to be set for the estimated parameters based on valve engineering knowledge, design data, and historical operating experience. For example, the reasonable range for valve local gain can be set to [5 (tons / hour) / %, 100 (tons / hour) / %], which excludes physically impossible negative gains or excessively small or large gain values. The reasonable range for estimated valve hysteresis width can be set to [0%, 1.0%], which excludes negative values ​​and excessive hysteresis for a well-maintained valve. Receive the valve command to be issued from the previous step and retrieve the estimated valve local gain and estimated valve hysteresis width for the current period from step 2. The command to be issued is 52.939%, the estimated local gain is 34.178 (tons / hour) / %, and the estimated hysteresis width is 0.115%. During implementation, two judgments are performed. First, it is determined whether the estimated local gain of 34.178 is within the range of [5, 100]. The result is yes. Second, it is determined whether the estimated hysteresis width of 0.115% is within the range of [0%, 1.0%]. The result is also yes. Since all identification parameters are within their reasonable ranges, it indicates that the adaptive identification logic is working normally and its compensation result is reliable. Therefore, the safety check passes. The logic then confirms that the valve command to be issued, 52.939%, will be the final valve command issued. If any estimated parameter exceeds its reasonable range in this step, for example, if the identified gain is -10, the safety check fails. This indicates that the identification algorithm may have been interfered with, and its output compensation result is unreliable. In this case, the valve command to be issued will be rejected, and the control logic will execute a preset safety mode action, such as keeping the valve command of the previous cycle unchanged, or temporarily disabling the adaptive compensation function to ensure the stability and safety of the control process. In this example, since the safety check passes, the final command issued to the valve actuator is determined to be 52.939%. This final valve command, as a precise and safety-verified digital signal, is sent to the positioner or actuator of the feedwater regulating valve. It directly drives the valve to perform physical actions, adjusting its actual opening degree, thereby achieving precise regulation of the feedwater flow rate and ultimately maintaining the deaerator water level stably near the set value.

[0046] In summary, a dynamic monitoring-based feedwater control method for deaerators in nuclear power plants, based on embodiments of this application, is explained. It constructs an adaptive compensation controller based on online valve characteristic identification between the main controller and the feedwater valve actuator. This aims to solve the control performance degradation problem caused by valve nonlinear characteristics (such as gain variation and hysteresis) in the prior art. The implementation involves first using a hybrid online identification algorithm to analyze historical valve position commands and feedwater flow sequences in real time, dynamically estimating the local gain and hysteresis width of the valve under the current operating conditions. Then, using these real-time identified parameters, the expected flow change calculated by the main controller is proactively compensated. That is, the command amplitude is adjusted according to the estimated local gain to ensure a consistent flow response under different loads, thereby overcoming the influence of nonlinear flow characteristics; simultaneously, when the command reverses, a compensation amount is actively applied based on the estimated hysteresis width to eliminate the dead zone and oscillation caused by hysteresis. In this way, the nonlinear object of the valve is equivalent to a linear element with a consistent response, significantly improving the stability and accuracy of the deaerator water level control system across the entire operating range.

[0047] Figure 6 This is a block diagram of a deaerator feedwater control system for a nuclear power plant based on dynamic monitoring, according to an embodiment of this application. Figure 6 As shown, the nuclear power plant deaerator feedwater control system 100 based on dynamic monitoring according to an embodiment of this application includes: a desired flow change calculation module 110, used to calculate the desired flow change of the main controller based on the actual deaerator water level, water level setpoint, main steam flow, and actual feedwater flow to obtain the desired feedwater flow change; a parameter estimation module 120, used to perform hybrid online identification of valve characteristic parameters on historical valve position command sequences and historical feedwater flow sequences to obtain estimated valve local gain and estimated valve hysteresis width; an adaptive compensation module 130, used to perform adaptive compensation on the final valve command and desired feedwater flow change of the previous cycle based on the estimated valve local gain and estimated valve hysteresis width to obtain an unconstrained compensated valve command; and a valve command generation module 140, used to perform command limiting and safety verification on the unconstrained compensated valve command to obtain the finally issued valve command.

[0048] Here, those skilled in the art will understand that the specific operations of each step in the above-described dynamic monitoring-based deaerator feedwater control system for nuclear power plants have been referenced above. Figures 1 to 5 The description of the dynamic monitoring-based deaerator feedwater control method for nuclear power plants is detailed here, and therefore, its repeated description will be omitted.

Claims

1. A method for controlling deaerator feedwater in a nuclear power plant based on dynamic monitoring, characterized in that, include: Based on the actual deaerator water level, water level setpoint, main steam flow rate and actual feedwater flow rate, calculate the expected flow rate change of the main controller to obtain the expected feedwater flow rate change. A hybrid online identification of valve characteristic parameters is performed on historical valve position command sequences and historical water supply flow sequences to obtain estimated valve local gain and estimated valve hysteresis width. Based on the estimated local valve gain and the estimated valve hysteresis width, adaptive compensation is performed on the final valve command and the expected change in feedwater flow rate of the previous cycle to obtain an unconstrained compensated valve command. The unconstrained compensated valve commands are subjected to command amplitude limiting and safety verification to obtain the final issued valve commands.

2. The method for controlling deaerator feedwater in nuclear power plants based on dynamic monitoring according to claim 1, characterized in that, Based on the actual deaerator water level, water level setpoint, main steam flow rate, and actual feedwater flow rate, the expected flow rate change of the main controller is calculated to obtain the expected feedwater flow rate change, including: Calculate the difference between the actual deaerator water level and the water level setpoint to obtain the water level deviation for the current control cycle; The water level deviation of the current control cycle is input into the PID controller to obtain the feedback flow component; The expected total feedwater flow rate is obtained by combining the feedback flow component and the main steam flow rate with the feedforward signal. The expected total water supply flow rate and the actual water supply flow rate are subtracted to obtain the expected change in water supply flow rate.

3. The method for controlling deaerator feedwater in nuclear power plants based on dynamic monitoring according to claim 2, characterized in that, The desired total feedwater flow rate is obtained by combining the feedback flow component and the main steam flow rate with the feedforward signal. This includes: combining the feedback flow component and the main steam flow rate with the feedforward signal to obtain the desired total flow rate using the following formula: , ;in, This is the feedforward gain coefficient. Main steam flow rate, To provide feedback on flow components, For feedforward flow, This represents the expected total water supply flow rate.

4. The method for controlling deaerator feedwater in nuclear power plants based on dynamic monitoring according to claim 1, characterized in that, Online identification of valve characteristic parameters is performed on historical valve position command sequences and historical feedwater flow sequences to obtain estimated valve local gain and estimated valve hysteresis width, including: Based on the system's pure time delay, dynamic data vectors are constructed and time-delay aligned with the historical valve position command sequence and historical water supply flow sequence to obtain the current flow change, the command change after time delay alignment, the command change direction vector, and the unaligned original command change. The estimated valve local gain is obtained by performing a recursive least squares estimation based on the current flow rate change and the command change after time delay alignment. Event-driven identification and filtering of valve hysteresis width are performed on the command change direction vector and the misaligned original command change amount to obtain the estimated valve hysteresis width.

5. The method for controlling deaerator feedwater in a nuclear power plant based on dynamic monitoring according to claim 1, characterized in that, Based on the estimated local valve gain and estimated valve hysteresis width, adaptive compensation is performed on the final valve command and expected feedwater flow rate change of the previous cycle to obtain an unconstrained compensated valve command, including: The basic change is calculated based on the estimated local valve gain and the expected change in feedwater flow rate; Based on the estimated valve hysteresis width and the basic change, the hysteresis compensation is calculated. The unconstrained compensated valve command is obtained by adding the basic change, the hysteresis compensation, and the final valve command of the previous cycle.

6. The method for controlling deaerator feedwater in a nuclear power plant based on dynamic monitoring according to claim 5, characterized in that, Based on the estimated valve hysteresis width and the basic change, the hysteresis compensation is calculated, including: Determine whether the direction of the basic change is opposite to the direction of the instruction change in the previous cycle; Therefore, based on the estimated valve hysteresis width and the basic change, the hysteresis compensation is calculated using the following formula: ;in, For symbolic functions, Based on the change, To estimate the valve hysteresis width, It is the amount of hysteresis compensation; If not, set the hysteresis compensation to zero.

7. The method for controlling deaerator feedwater in a nuclear power plant based on dynamic monitoring according to claim 1, characterized in that, The unconstrained, compensated valve commands are subjected to command amplitude limiting and safety verification to obtain the final issued valve commands, including: The amplitude and rate of the unconstrained compensated valve commands are limited to obtain the valve commands to be issued. Determine whether the estimated valve local gain and estimated valve hysteresis width are within a reasonable range. If so, determine the valve command to be issued as the final valve command to be issued.

8. A deaerator feedwater control system for nuclear power plants based on dynamic monitoring, characterized in that, include: The expected flow change calculation module is used to calculate the expected flow change of the main controller based on the actual deaerator water level, water level setpoint, main steam flow, and actual feedwater flow to obtain the expected feedwater flow change. The parameter estimation module is used to perform online mixed identification of valve characteristic parameters on historical valve position command sequences and historical water supply flow sequences to obtain estimated valve local gain and estimated valve hysteresis width. The adaptive compensation module is used to adaptively compensate the final valve command and expected change in feedwater flow rate of the previous cycle based on the estimated local valve gain and the estimated valve hysteresis width to obtain an unconstrained compensated valve command. The valve command generation module is used to perform command amplitude limiting and safety verification on the unconstrained compensated valve commands to obtain the final issued valve commands.

Images