Generalized predictive control method for once-through steam generator of nuclear power plant

Abstract

The invention discloses a generalized predictive control method for a once-through steam generator of a nuclear power plant, and the method comprises the steps: inputting a working condition change plan into a once-through steam generator simulation calculation model based on numerical calculation, predicting the dynamic behavior of the once-through steam generator in a future period of time, and obtaining a simulation result, a generalized predictive control algorithm is adopted to optimize the control vector of the once-through steam generator; the optimized control vector acts on the once-through steam generator, iterative optimization of the control vector is achieved, and finally working condition adjustment of the once-through steam generator is completed safely, economically and stably according to a working condition change plan. The method can improve the response speed, reduce the adjustment time, reduce the pressure fluctuation of the once-through steam generator, indirectly improve the safety of the device, prolong the service life of the heat transfer pipe, achieve the automatic control of the once-through steam generator under the low-power working condition, and quantitatively bring various demands in the operation process of the once-through steam generator into control constraints.

Description

Technical Field

[0001] The invention relates to the technical field of direct current steam generator control, and in particular to a generalized predictive control method for a direct current steam generator of a nuclear power plant. Background Art

[0002] Steam generator is an important device for transferring energy between the primary and secondary circuits of the reactor. According to the flow mode of the secondary working fluid, it can be mainly divided into natural circulation steam generator and direct current steam generator. The main characteristics of the direct current steam generator are: the secondary circuit working fluid is driven by the feedwater pump pressure head through the steam generator, and once flows through the heat transfer surface to transform into slightly superheated steam, and the circulation ratio is 1. Compared with the vertical U-tube natural circulation steam generator widely used in pressurized water reactor nuclear power plants, the direct current steam generator has significant advantages such as simple structure, compact size, good static performance, fast power increase and decrease speed, and the ability to improve the thermal efficiency of the system. The low circulation ratio of the direct current steam generator makes its heat storage and buffering capacity poor, and the secondary side pressure is very easy to fluctuate. Too high pressure will endanger the safety of the system, and too low pressure will not meet the steam demand of downstream devices. Therefore, maintaining the stability of the secondary side pressure is the main goal of the automatic control of the direct current steam generator, that is, by controlling the feedwater flow rate to stabilize the steam pressure at the secondary side outlet of the steam generator, ensure the safe and stable operation of the direct current steam generator and provide high-quality steam that meets the operating requirements for steam-using equipment.

[0003] The automatic control of direct current steam generators generally has the following problems: 1. The nonlinearity between the control quantity and the controlled quantity is strong. The dynamic characteristics of the direct current steam generator will undergo complex changes with different power levels, which further increases the difficulty of secondary side pressure control and makes it difficult to obtain excellent results through simple control models; 2. Simple control methods are difficult to meet the various requirements and restrictions of factors such as outlet pressure, fluctuation of evaporation point, and change speed of feed water flow during device operation; 3. Traditional methods are difficult to identify and avoid unstable conditions of two-phase flow. These problems together make it difficult to determine the optimal control strategy of direct current steam generators. At the same time, due to the reduction of the length of the two-phase section under low power conditions, the steam pressure is very sensitive to changes in feed water flow. At present, the control of direct current steam generators adopts a method that combines automatic control under high power conditions and manual control under low power conditions. Although this method can meet the current operating needs of direct current steam generators, it is difficult to adapt to the development direction of miniaturization and unmanned operation of future nuclear power plants.

[0004] For the automatic control of direct current steam generators, the classic PID method was used in the early days. This method has been widely verified in various fields, has a simple principle and strong applicability, and can obtain relatively reliable results for various systems. The literature "Zhang Yusheng, Guo Lifeng, Cai Meng. Research on PID control system of direct current steam generator based on fuzzy adaptive parameter setting [J]. Nuclear Power Engineering, 2008 (04): 93-96" uses the PID method as the basis to design a three-impulse fuzzy adaptive PID control system for direct current steam generators. The primary circuit feed water temperature, steam flow rate, and feed water flow rate are fed into the control system through a proportional link, and the three parameters of the PID system are adjusted in real time by the fuzzy method. Although the control effect of the PID method is improved to a certain extent, the PID method still has the problems of large pressure fluctuations and long adjustment time. In order to further improve the performance of the direct current steam generator control system, neural network methods and fuzzy methods have gradually become the main development direction of direct current steam generator automatic control in recent years. These two methods have been tried in many literatures to improve the automatic control system of direct current steam generators. However, the lack of interpretability of the neural network method and the poor stability of fuzzy control have always made these two methods difficult to apply on a large scale. The concept of digital nuclear power provides a new solution to the control problem of direct current steam generators, that is, the controller relies on the digital nuclear power system to obtain a large amount of data support, and uses this as a basis to improve the automatic control effect of the direct current steam generator. Summary of the invention

[0005] The purpose of the present invention is to provide a generalized predictive control method for a direct-current steam generator of a nuclear power plant, which achieves the effect of improving response speed, reducing adjustment time, reducing pressure fluctuation of the direct-current steam generator, quantitatively considering the operating requirements of the direct-current steam generator and automatically controlling the direct-current steam generator under low power conditions through generalized predictive control, synchronous disturbance stochastic approximation, numerical calculation and other technical means.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A generalized predictive control method for a nuclear power plant once-through steam generator, the method comprising the following steps:

[0008] S1, inputting the operating condition change plan into the simulation calculation model of the once-through steam generator based on numerical calculation, predicting the dynamic behavior of the once-through steam generator in the future, and obtaining the simulation results;

[0009] S2, based on the simulation results, the control vector of the once-through steam generator is optimized using the generalized predictive control algorithm;

[0010] S3, applying the optimized control vector to the once-through steam generator, and using sensors to perform real-time measurements on the once-through steam generator in online operation to obtain thermal hydraulic data;

[0011] S4, based on the thermal hydraulic data, the simulation results in S1 are corrected, and steps S2-S4 are repeated to achieve iterative optimization of the control vector, and finally the operating condition change plan is implemented on the once-through steam generator.

[0012] Further, in said S1, said direct current steam generator simulation calculation model includes an operation simulation model and a fast solution model of the direct current steam generator;

[0013] The construction method of the running simulation model is as follows:

[0014] According to the structural parameters and static and dynamic operating characteristics of the once-through steam generator, a one-dimensional thermal analysis model of the full-scale once-through steam generator is constructed as an operating simulation model of the once-through steam generator.

[0015] The method for constructing the fast solution model is as follows:

[0016] According to the actual situation of the once-through steam generator, a single heat transfer tube is taken according to the arrangement of the heat transfer tubes to establish a one-dimensional thermal analysis model of the once-through steam generator as a fast solution model, and the fast solution model is used to solve the secondary side pressure drop;

[0017] The fast solution model interacts with the running simulation model in data, and the primary side inlet and outlet flow rates and the secondary side inlet and outlet flow rates are magnified by corresponding multiples.

[0018] Furthermore, before S1, the operation simulation model is trained offline to construct an operation database, and the specific method is as follows:

[0019] According to the operation requirements of the once-through steam generator, a large number of examples with different load change rates and different start and stop conditions are set up and input into the operation simulation model and the fast solution model. Offline training is performed using the synchronous perturbation random approximation method. Through the offline training process, simulation calculation and control vector solution are performed to obtain the operation data and corresponding control vectors under different load change rates and different start and stop conditions.

[0020] The operation data and the corresponding control vectors are divided into several groups based on the load change rate. The operation data and the corresponding control vectors are arranged and stored in each group based on the initial operating conditions and the target operating conditions to build an operation database.

[0021] Furthermore, in S1, the operating condition change plan is input into a simulation calculation model of a once-through steam generator based on numerical calculation, and the dynamic behavior of the once-through steam generator in the future is predicted to obtain simulation results, which specifically include:

[0022] Based on the operating condition change plan, run the simulation model to perform simulation calculations to obtain the expected load change rate and start and end conditions;

[0023] Based on the constructed operation database, the selection range of the control vector is first determined according to the expected load change rate, and then the required control vector is determined according to the expected start and end conditions. The feedwater flow in the control vector is fitted according to the change trend of the steam flow, and at the same time, the fitting curve is ensured to be equal to the integral of the feedwater flow over time. Finally, this fitting curve is used as the initial value of the control vector during online operation.

[0024] After obtaining the initial value of the control vector, the control vectors of the forward and reverse disturbances are simultaneously input into the fast solution model to predict the dynamic behavior of the DC evaporator in the future and obtain the simulation results.

[0025] Furthermore, the S2, based on the simulation results, uses a generalized predictive control algorithm to optimize the control vector of the once-through steam generator, specifically including:

[0026] The simulation results are processed to obtain the gradient of the loss function under the current control vector, and the control vector is optimized based on this gradient. This process is repeated until the termination criterion is met, and then the first item of the control vector is passed to the running simulation model and the inlet flow is advanced to the next control step.

[0027] Furthermore, the simulation results are processed to obtain the gradient of the loss function under the current control vector, and the control vector is optimized according to the gradient, and this process is repeated until the termination criterion is met, specifically including:

[0028] Under the operation scheme with constant secondary side steam pressure, the square of the difference between the steam pressure per unit time and the rated value is selected as the loss function, and its expression is:

[0029]

[0030] Where t′ is the current time, t″ is the simulation time required to quickly solve the model, p is the steam pressure, p0 is the rated steam pressure or target steam pressure, and θ is the current control vector;

[0031] The current control vector is judged according to the obtained loss function value. When the loss function value is within an acceptable range, the direct-flow steam generator is determined to be in a stable state and the current control vector is output. Otherwise, the current control vector is optimized and this process is repeated until the termination criterion is met.

[0032] Furthermore, the optimization of the current control vector is repeated until the termination criterion is met, including:

[0033] Based on the current control vector, the simulation model is run to update the parameters, and the synchronous perturbation random approximation method is used to generate the perturbation vector, including the forward perturbation vector and the reverse perturbation vector. Through the gradient approximation, the control vector is optimized and iterated.

[0034] When the maximum number of iterations is reached and the loss function value corresponding to the current control vector is less than the minimum value of the loss function that occurs during the iteration, the iteration ends;

[0035] If the maximum number of iterations is reached, but the loss function value corresponding to the current control vector is not less than the minimum value of the loss function that occurs during the iteration, the control vector corresponding to the minimum value of the loss function during the iteration is used as the result to end the iteration.

[0036] Furthermore, in S4, steps S2-S4 are repeated to achieve the online running iteration termination criterion of iterative optimization of the control vector:

[0037] The control vector is continuously optimized and iterated. After each iteration, the control vector with the smallest corresponding loss function generated during the iteration process is updated. When the real time passes a control step, the optimization is stopped and the first item of the control vector is directly output. At the same time, the time domain is moved backward to start solving the next control step.

[0038] Furthermore, the method also includes updating the running database, as follows:

[0039] After the operating condition change plan is implemented on the once-through steam generator, the simulation results of the once-through steam generator are compared and sorted with the measurement data, and the erroneous data are corrected to form a control case and store it in the operation database;

[0040] After the control case storage is completed, the update of the operation database begins. According to the recording time of a case and the data density in a certain area around the case, when establishing the operation database update mechanism, experts set different data densities in different areas based on experience. Frequently occurring working conditions or working conditions with more complex changes require a larger data density, and vice versa.

[0041] When updating the running database, first traverse the historical data around the new case in chronological order from old to new. If it is detected that the data density around a case exceeds the local standard, the group of data will be deleted.

[0042] The present invention also provides a control device, including a memory, a processor, and a program stored in the memory, wherein when the processor executes the program, the generalized predictive control method of the direct current steam generator of a nuclear power plant as described above is implemented.

[0043] According to the specific embodiment provided by the present invention, the generalized predictive control method of a nuclear power plant once-through steam generator provided by the present invention discloses the following technical effects:

[0044] (1) The present invention realizes the generalized predictive control algorithm through numerical calculation and gradient optimization of the control vector. On the one hand, it realizes the advance response of the water supply flow rate. On the other hand, it can also better complete the task of automatic control of the direct current steam generator under low power conditions. At the same time, the control method combined with numerical calculation not only ensures the strong interpretability of the control scheme, but also improves the control accuracy and robustness to a certain extent. It can effectively deal with the simultaneous disturbance of multiple parameters and provide a reliable control scheme to achieve stable and efficient control effects. It can also effectively improve the safety of the device, reduce parameter fluctuations and extend the life of the heat transfer tube.

[0045] (2) The present invention incorporates the use of an operation database, which significantly improves computational efficiency while giving the control method the ability to learn and optimize autonomously. By collecting and storing operating data during operation in real time, the operation database implicitly contains the dynamic characteristics of the system, which not only provides rich data support for the formulation of control strategies, but also becomes the cornerstone of the self-evolution and optimization of the control method.

[0046] (3) The present invention combines the gradient optimization algorithm with the rapid solution model, so that the control system indirectly considers the internal relationship between the thermal and hydraulic parameters of the direct current steam generator. At the same time, by linking the controller coefficient and the power level, the dynamic characteristics of the direct current steam generator are better considered, and the changes in the direct current steam generator can be intuitively and accurately reflected through the rapid solution model; the direct current steam generator data provided by the digital nuclear power system is effectively utilized, and the accuracy of the results is enhanced by correcting the rapid solution model.

[0047] (4) The present invention achieves a streamlined algorithm structure and a high degree of robustness in coefficient setting by applying the synchronous disturbance stochastic approximation algorithm, which makes the coefficient setting simpler and enhances the fault tolerance during the operation of the control method. Moreover, the synchronous disturbance stochastic approximation algorithm cleverly simplifies the feedback correction link of predictive control through its unique mechanism, making the entire control method more concise and clear, and avoiding the increase in system complexity and failure risk caused by the combination of multiple control algorithms in complex control. At the same time, through the setting of the penalty function, various constraints in the operation of the direct current steam generator can be quantitatively considered and a control scheme with the best comprehensive performance can be given. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 It is a schematic diagram of a generalized predictive control method for a nuclear power plant once-through steam generator according to an embodiment of the present invention;

[0050] Figure 2 It is a flow chart of a generalized predictive control method for a nuclear power plant once-through steam generator according to an embodiment of the present invention;

[0051] Figure 3 This is a flow chart of a synchronous perturbation random approximation algorithm according to an embodiment of the present invention;

[0052] Figure 4 Schematic diagram of the Relap5 model of a once-through steam generator according to an embodiment of the present invention, wherein 101-105 respectively represent an inlet control body, an inlet pipe, a tube bundle, an outlet pipe and an outlet control body on the primary side, 201-205 respectively represent an inlet control body, an inlet pipe, a tube bundle, an outlet pipe and an outlet control body on the secondary side, and the heat transfer solution domain of the heat exchange tube wall is between the primary side and the secondary side;

[0053] Figure 5 A schematic diagram of a database operation according to an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of updating a database according to an embodiment of the present invention;

[0055] Figure 7 It is a schematic diagram showing the comparison of the effects of the generalized predictive control method and the PID control method under the secondary frequency modulation and power reduction condition of a nuclear power plant according to an embodiment of the present invention;

[0056] Figure 8It is a schematic diagram comparing the effects of the generalized predictive control method and the PID control method under the emergency power reduction condition of an embodiment of the present invention;

[0057] Fig. 9 Schematic diagram of coefficient perturbation calculation example of an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The purpose of the present invention is to provide a generalized predictive control method for a nuclear power plant direct current steam generator, combining numerical calculation and gradient optimization, and combining thermal hydraulic data to improve the direct current steam generator control effect. The present invention establishes an effective control scheme through technical means such as generalized predictive control, synchronous disturbance stochastic approximation, and numerical calculation, and achieves the purpose of improving response speed, reducing adjustment time, reducing direct current steam generator pressure fluctuation, quantitatively considering the constraints of direct current steam generator operation requirements, and automatically controlling the direct current steam generator under low power conditions.

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0061] like Figures 1 to 5 As shown, the generalized predictive control method for a nuclear power plant once-through steam generator provided by the present invention is performed by a generalized predictive control system of the nuclear power plant once-through steam generator, and the method comprises the following steps:

[0062] S1, inputting the operating condition change plan into the simulation calculation model of the once-through steam generator based on numerical calculation, predicting the dynamic behavior of the once-through steam generator in the future, and obtaining the simulation results;

[0063] S2, based on the simulation results, the control vector of the once-through steam generator is optimized using the generalized predictive control algorithm;

[0064] S3, applying the optimized control vector to the once-through steam generator, and using sensors to perform real-time measurements on the once-through steam generator in online operation to obtain thermal hydraulic data;

[0065] S4, repeating steps S2-S4 to achieve iterative optimization of the control vector, and finally implementing the operating condition change plan on the once-through steam generator.

[0066] The present invention adopts the following technical solutions, which specifically include: Figure 2 The process shown in the figure. This scheme is divided into two main stages: offline training and online operation. It should be noted that offline training does not include the process of providing the initial value of the control scheme by the historical database. The method for determining the initial value of the control scheme is: the control amount of the new control step is the average of the control amounts of the previous four control steps.

[0067] Specifically, in said S1, said once-through steam generator simulation calculation model includes an operation simulation model and a fast solution model of the once-through steam generator;

[0068] The construction method of the running simulation model is as follows:

[0069] According to the structural parameters of the DC steam generator (such as the size of the heat transfer tube, the layout, the shell size, etc.) and the static and dynamic operating characteristics, a one-dimensional thermal analysis model of the full-scale DC steam generator is constructed as the operation simulation model of the DC steam generator; the static characteristics: consider the characteristics of the DC steam generator under stable conditions, including temperature distribution, pressure distribution, etc.; the dynamic characteristics: consider the dynamic response of the DC steam generator under load changes, temperature fluctuations, etc. The one-dimensional thermal analysis model of the full-scale DC steam generator includes the one-dimensional thermal analysis of all key components, and can provide detailed changes in parameters such as temperature, pressure, and flow.

[0070] The method for constructing the fast solution model is as follows:

[0071] According to the actual situation of the direct current steam generator, a single heat transfer tube is taken according to the arrangement of the heat transfer tubes to establish a one-dimensional thermal analysis model of the direct current steam generator as a fast solution model. In the simulation calculation link, the steam pressure, steam temperature, steam flow, feed water flow, feed water temperature, feed water pressure and related parameters of the primary inlet and outlet working fluid are stored after the calculation is completed. In the fast solution model, the main goal to be solved is the secondary side pressure drop, so this method can save computing resources and speed up the calculation. It is particularly important to note that when the fast solution model and the running simulation model are interacting with data, the primary side inlet and outlet flow and the secondary side inlet and outlet flow need to be magnified by the corresponding multiples.

[0072] This is because the fast solution model is based on a single heat transfer tube, while the actual once-through steam generator consists of multiple heat transfer tubes. The magnification factor is usually determined based on the number of heat transfer tubes.

[0073] Before S1, the operation simulation model is trained offline to build an operation database. The specific method is as follows:

[0074] According to the operation requirements of the once-through steam generator, a large number of examples with different load change rates and different start and stop conditions are set up and input into the operation simulation model and the fast solution model. Offline training is performed using the synchronous perturbation random approximation method. Through the offline training process, simulation calculation and control vector solution are performed to obtain the operation data and corresponding control vectors under different load change rates and different start and stop conditions.

[0075] The operation data and the corresponding control vectors are divided into several groups based on the load change rate. The operation data and the corresponding control vectors are arranged and stored in each group based on the initial operating conditions and the target operating conditions to build an operation database.

[0076] The purpose of the present invention to conduct offline training is to accelerate the convergence speed by optimizing the initial value of the control vector during operation. Therefore, an operation database is established to generalize the experience of offline training to the online operation process. The specific method is: query the historical data of the same working condition through the operation database, and fit the control vector obtained by the synchronous perturbation random approximation method during offline training to serve as the initial value of the control vector during online operation.

[0077] After completing the simulation model establishment work, it is necessary to determine the control principle according to the selected operation scheme and reflect it through the loss function. The present invention selects the operation scheme with constant secondary side steam pressure during the test, so the steam pressure is directly selected as the controlled quantity and the loss function is constructed based on it.

[0078] The control step refers to the time interval at which the controller updates its output or performs a control action. It determines how often the controller adjusts the control direction. At the beginning of each control step, the controller needs to be initialized and the initial value of the control vector to be optimized is selected. The selection of the initial value of the control scheme is different from that of offline training. In offline training, the control quantity newly included in the moving time domain as time moves forward is obtained by the weighted average of the first few elements of the control vector. When running online, it is directly provided by the operation database based on the starting operating conditions and load change rate. The specific method is to first determine the range of the target control scheme based on the load change rate, and then determine the required control scheme based on the starting and ending conditions. The feedwater flow in the control scheme is fitted according to the change trend of the steam flow, and at the same time ensure that the fitting curve is equal to the integral of the feedwater flow over time. Finally, this fitting curve is used as the initial value of the control quantity during online operation.

[0079] Specifically, the S1 inputs the operating condition change plan into a simulation calculation model of a once-through steam generator based on numerical calculation, predicts the dynamic behavior of the once-through steam generator for a period of time in the future, and obtains simulation results, which specifically include:

[0080] Based on the operating condition change plan, the simulation model is run for simulation calculation to obtain the expected load change rate and start and end conditions. The so-called operating condition change plan is the design of the start and end points and paths of the operating condition change of the once-through steam generator set by the operator according to the needs and environment, such as Figure 2 It is shown that the water flow rate q is q0 before time t0 and q1 after time t1, and changes linearly between time t0 and time t1;

[0081] Based on the constructed operation database, the selection range of the control vector is first determined according to the expected load change rate, and then the required control vector is determined according to the expected start and end conditions. The feedwater flow in the control vector is fitted according to the change trend of the steam flow, and at the same time, the fitting curve is ensured to be equal to the integral of the feedwater flow over time. Finally, this fitting curve is used as the initial value of the control vector during online operation.

[0082] After obtaining the initial value of the control vector, the control vectors of the forward and reverse disturbances are simultaneously input into the fast solution model to predict the dynamic behavior of the DC evaporator in the future and obtain the simulation results.

[0083] Specifically, the S2, based on the simulation results, uses a generalized predictive control algorithm to optimize the control vector of the once-through steam generator, specifically including:

[0084] The simulation results are processed to obtain the gradient of the loss function under the current control vector, and the control vector is optimized based on this gradient. This process is repeated until the termination criterion is met. After that, the first item of the control vector is passed to the running simulation model, and the inlet flow is pushed forward based on this, and the moving domain moves forward.

[0085] The present invention uses a synchronous perturbation random approximation method to optimize the control vector. The synchronous perturbation random approximation method generally controls the optimization of the result by the number of iterations. For simple problems, a larger number of iterations can be directly set to ensure that the result converges to the optimal solution. In the present invention, the following convergence criteria are selected:

[0086] (1) The number of iterations must be greater than 20;

[0087] (2) The loss function value corresponding to the current control vector is less than the minimum value of the loss function that occurs during the iteration process;

[0088] The iteration ends when (1) and (2) are both satisfied. When (1) is satisfied but (2) is not satisfied, the control vector corresponding to the minimum value of the loss function during the iteration is used as the result to end the iteration.

[0089] Since online operation requires computing speed, continuing to use the iteration termination criteria used in offline training may lead to the problem of too slow solution. The iteration termination criteria for online operation is: continuously optimize and iterate the control vector, update the control vector with the smallest corresponding loss function after each iteration, stop optimization after a control step in real time, directly output the first item of this control vector, and move the time domain backward to start solving the next control step.

[0090] The method also includes updating the execution database as follows:

[0091] After the process of changing the working condition is over, the simulation data of the direct steam generator is compared and sorted with the measured data, and the erroneous data is corrected to form a case and store it in the operation database. With the operation of the direct steam generator, more and more cases are stored in the operation database. Therefore, in order to ensure the timeliness of the operation database, it is necessary to delete the redundant and outdated data to complete the update of the operation database. The update of the operation database is started below. At this time, two quantities are considered, namely, the length of time from the time a case is recorded to the current moment, and the data density of a certain area around a case. When establishing the operation database update mechanism, experts set different data densities in different areas based on experience. Frequently occurring working conditions or working conditions with more complex changes require a larger data density, and vice versa, a smaller data density is required. When updating the operation database, first traverse the historical data near the new working condition from old to new in chronological order. If it is detected that the data density around a group of historical data exceeds the local standard, the group of data is deleted. This method not only ensures that the database covers as many working conditions as possible, but also ensures that the database follows potential changes such as scaling of the direct steam generator.

[0092] Example:

[0093] 1. Simulation Calculation of Once-through Steam Generator

[0094] The embodiment of the present invention uses the Relap5 program to establish a fast solution model and a running simulation model of a direct current steam generator: taking the B&W company's straight tube direct current steam generator experimental device as an example. First, refer to the parameters shown in Table 1 to establish a fast solution model and a running simulation model. Figure 4 As shown, Figure 4 The primary and secondary sides of the once-through steam generator are divided into sixty nodes, which are connected to the control bodies representing the upstream and downstream devices of the primary and secondary circuits through time interfaces.

[0095] The next step is to use the running simulation model built by the Relap5 program to obtain the initial working conditions of the offline training example and prepare for offline training. In the development stage, the calculation is completed by reading the output text of the Relap5 program and automatically generating the next running file. In actual application, facing the specific direct steam generator control object, a one-dimensional movable boundary simulation model embedded in the control program is established as a fast solution model to improve the calculation efficiency.

[0096] Table 1 Main parameters of once-through steam generator

[0097]

[0098]

[0099] 2. Processing of control quantities and input of fast solution model

[0100] First, the control quantity is initialized. In the present invention, the control quantity in the mobile time domain, i.e., the water flow rate, is discretized into the following form according to the "black box" method, which is called the control vector:

[0101] [θ0,θ1,θ2,θ3,θ4]

[0102] The subscript of each component in the control vector represents the corresponding position of the component in the moving time domain. The so-called moving time domain is the time range [t p ,t p+N ], whose length is determined by the dimension of the control vector and the duration covered by a single control step.

[0103]

[0104] Among them, t p represents the current time, t p+N represents the end time of the current moving time domain, p represents the sequence number of the first control step in the current moving time domain in the entire example, N represents the number of control steps covered by the moving time domain, Δt i Indicates the duration covered by a single control step;

[0105] During offline training, the fifth component in the control vector is obtained by weighted average of the first four items, and during online operation, the feedwater flow in the control scheme is fitted according to the change trend of the steam flow, while ensuring that the fitting curve is equal to the integral of the feedwater flow over time. Finally, this fitting curve is used as the initial value of the control quantity during online operation.

[0106] The input quantities required for quickly solving the model include: the initial state of the system, the boundary conditions of the once-through steam generator in the moving time domain, and the initial value of the control vector.

[0107] The method for determining the initial state of the system is as follows: when the system is initialized for the first time, the once-through steam generator (running simulation model) is operated to be stable, and then a stable flow field solution is obtained by using a fast solution model under this condition, and the flow field solution obtained by the fast solution model can be used as the initial state of the system. The subsequent initial state of the system is obtained by correcting the solution of the fast solution model in the previous control step with the sensor data of the once-through steam generator (flow field solution of the running simulation model).

[0108] The boundary conditions for quickly solving the model in the mobile time domain in the present invention are the other primary and secondary side working fluid inlet and outlet parameters except the feed water flow provided by the direct current steam generator sensor data (the flow field solution of the running simulation model).

[0109] 3. Determine the loss function:

[0110] Design the loss function according to the operation scheme of the control target: The control scheme established by the present invention uses a fast solution model to determine the values ​​of the objective function and the loss function when the initial conditions and the input control vector θ are given in the moving time domain of N control steps. In this embodiment, the operation scheme with constant secondary side steam pressure is selected, so the square of the difference between the steam pressure and the rated value per unit time is selected as the loss function, and its expression is:

[0111]

[0112] Where t′ is the current time, t″ is the simulation time required to quickly solve the model, p is the steam pressure, p0 is the rated steam pressure or target steam pressure, and θ is the control vector. In practical applications, a regularization term that quantitatively describes the operation requirements of the once-through steam generator can be added to this function.

[0113] The control vector is judged according to the obtained loss function value. When the loss function value is within an acceptable range, it is considered that the once-through steam generator is basically stable and the control vector can be directly output. Otherwise, the control vector is optimized and iterated.

[0114] 4. Optimization and Iteration of Control Vector

[0115] The control vector is iterated and optimized as follows:

[0116] The first step is to initialize the SPSA (Synchronous Perturbed Stochastic Approximation) system. Set the counter k to 0 and determine the SPSA gain sequence a k and c k The coefficients a, c, α, β, and γ in the iteration process are as follows. kThe correction speed of a, too small a will lead to slow convergence speed, and too large a will lead to iterative divergence. It is worth noting that in the control process of the direct steam generator, the fixed a cannot meet the control requirements under the condition of large-span power change. Therefore, in this paper, a is expressed by a function with the current feedwater flow rate as the independent variable:

[0117] a=0.1(θ kp +1)

[0118] Among them, θ kp It is the water flow rate of the first control step in the kth iteration in the current time domain.

[0119] c determines the size of the gradient value used to update the water flow rate. As the iteration progresses, the smaller c k It can ensure the convergence of the results. According to the experience of this paper, the additional disturbance should make the steam pressure difference between the forward disturbance and the reverse disturbance within 0.005Mpa, and the effect is better when it is about 0.001Mpa. α and γ determine the reduction speed of the correction amount in each feedwater flow iteration:

[0120]

[0121] Among them, a k 、c k , k, and A are internal coefficients of the SPSA algorithm and have no specific meanings.

[0122] β is the coefficient that relates the correction amount to the load change rate and needs to be obtained through experimental adjustment:

[0123]

[0124] Wherein, Q is the load of the direct current steam generator at the current moment, and f represents the function with the load change rate as the independent variable.

[0125] The second step is to generate an N-dimensional random perturbation vector Δ by Monte Carlo method. k The classic and effective method is that each choice follows the ±1 of Bernoulli distribution, with a probability of 0.5. The disturbance vector is added to the vector of water flow in the positive and negative directions according to the following formula, which are: θ k +c k Δ k and θ k -c k Δ k .

[0126] The third step is to input the disturbed water flow vector into the fast solution model to obtain the two measured values ​​y(θ k +c k Δ k ) and y(θk -c k Δ k ) and find the gradient vector g k :

[0127]

[0128] Among them, the role of τ is to associate the gradient optimization process with the power level and adjust the order of magnitude of the obtained gradient vector. It is a single-valued function of the power level and needs to be obtained through experimental adjustment. The meaning of is: the elements of the random perturbation vector generated in the kth iteration of the moving time domain from the pth control step to the p+Nth control step.

[0129] The fourth step is to substitute the obtained gradient vector into the following formula to update the water flow rate θ k And return to the second step until the convergence condition is met.

[0130] θ k+1 =θ k -a k g k (θ k )

[0131] 5. Iteration Termination Criteria

[0132] For the SPSA method, a simple approach is to use the number of iterations as the iteration termination criterion, or to use the correction value of the input vector or the measured value of the loss function. In the present invention, a certain number of iterations and a loss function value less than the historical minimum value in the current control step optimization process are used as the iteration termination criterion.

[0133] k>k max ,y <y min

[0134] In the formula, k represents the number of iterations, k max is the maximum number of iterations, y represents the loss function value corresponding to the control vector, y min Represents the minimum value of the loss function that occurs during the iteration process.

[0135] The iteration termination criterion during online operation is: continuously optimize and iterate the control vector, update the control vector with the smallest corresponding loss function generated in the iteration process after each iteration, stop the optimization after a control step in real time, directly output the first item of this control vector, and move the time domain backward to start solving the next control step.

[0136] 6. Run database updates

[0137] like Figure 5As shown in the figure, when the once-through steam generator is regulated and stabilized in the new operating condition, the whole process of the operating condition change is recorded in the operation database with the disturbance start time as time 0. Taking the steam flow change condition as an example, the control case is stored in the operation database with the steam flow change rate as the z coordinate, the initial operating condition as the x coordinate, and the target operating condition as the y coordinate.

[0138] like Figure 6 As shown in the figure, after the control case storage is completed, the update of the operation database begins, based on the recording time of a case and the data density in a certain area R around the case. When establishing the operation database update mechanism, experts set different data densities in different areas based on experience. Frequently occurring working conditions or working conditions with more complex changes require a larger data density, and vice versa. When updating the operation database, first traverse the historical data in R near the new case in chronological order from old to new. If it is detected that the data density around a case exceeds the local standard, the group of data is deleted.

[0139] like Figure 7 As shown, the present invention can effectively stabilize the steam pressure within a small range under emergency power reduction conditions. Compared with the traditional PID control method, the present invention is still capable of completing the automatic control of the direct current steam generator under low power conditions. In the industry, the automatic control problem of the direct current steam generator under 30%-100% load is generally considered, while the present invention can still well complete the automatic control task under the low load condition of 15% of the rated power of the direct current steam generator.

[0140] The following examples cover common load regulation conditions of once-through steam generators and Figure 7 , Figure 8 It can be seen that the present invention can not only effectively reduce the pressure fluctuation of the direct current steam generator, but also greatly shorten the duration of the pressure fluctuation. It is not only conducive to improving the safety of the device, but also can extend the life of the device to a certain extent. For the long-life direct current steam generator, it can greatly reduce the probability of failure.

[0141] Table 2 Comparison of maximum pressure fluctuations of two controllers

[0142]

[0143]

[0144] like Fig. 9As shown, in the example of high-load secondary frequency modulation and power reduction, the main coefficients a, c, α, and γ of the control system in the present invention are disturbed by ±20%, and the simulation experimental results are compared to verify the robustness of the present invention. It can be found that the optimization of the control vector in the present invention mainly depends on the intrinsic connection between the feed water flow and pressure of the direct current steam generator. The controller coefficient only determines the process of the control vector approaching the optimal solution. Therefore, within a certain range, even if the controller coefficient is not carefully adjusted, the present invention can also achieve good control effects.

[0145] Finally, for the many requirements of the operation of the direct steam generator, the regularization term can be quantitatively established as a constraint. The realization of this function gives the direct steam generator the possibility of moving from automatic control to intelligent control. The following takes the fluctuation of the dry point and the outlet steam superheat, which are related to the life of the heat transfer tube, as an example:

[0146]

[0147] Where h is the height of the dry point, T(t,θ) is the outlet steam temperature, and T0 is the saturated steam temperature at the outlet pressure. For practical applications, researchers need to design the weights of each regularization term in combination with the operating criteria and requirements of the direct current steam generator to obtain satisfactory results. Taking the above formula as an example, the importance of steam pressure stability is far greater than the fluctuation of the dry point, while the importance of outlet steam superheat is the smallest. Therefore, a higher weight should be given to pressure fluctuations and a lower weight should be given to outlet steam superheat.

[0148] In summary, the generalized predictive control method for a nuclear power plant once-through steam generator provided by the present invention combines simulation calculation, generalized predictive control (GPC) and real-time feedback correction. The following is a detailed explanation and step-by-step breakdown of the entire process: Obtain the operating change plan from the database or dispatch system of the digital nuclear power system; Input the operating change plan into the simulation program based on numerical calculation; Use a suitable mathematical model (such as transfer function, state space model or one-dimensional thermal analysis model) to simulate the dynamic behavior of the direct current steam generator; Through simulation calculation, predict the key parameters of the direct current steam generator in the future, such as temperature, pressure, flow, etc.; Use the results of the simulation calculation as the input of generalized predictive control (GPC) to judge and optimize the control quantity; Use the sensors installed on the direct current steam generator to measure the key parameters in real time to obtain actual thermal hydraulic data; Compare the actual data measured by the sensor with the results of the simulation calculation; According to the difference between the actual data and the simulation results, adjust the parameters of the simulation model to ensure that the simulation model can more accurately reflect the dynamic characteristics of the direct current steam generator; Use the revised model to re-simulate and calculate to further improve the accuracy of the prediction; Apply the optimized control quantity to the direct current steam generator and gradually implement the operating change plan. During the entire process, the operating status of the system is continuously monitored, and real-time adjustments are made as needed to ensure the stability and safety of the system. This closed-loop control method not only improves the control accuracy, but also enhances the robustness and adaptability of the system.

[0149] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A generalized predictive control method for a nuclear power plant once-through steam generator, characterized in that: The following steps are involved: S1, inputting the operating condition change plan into the simulation calculation model of the once-through steam generator based on numerical calculation, predicting the dynamic behavior of the once-through steam generator in the future, and obtaining the simulation results; S2, based on the simulation results, the control vector of the once-through steam generator is optimized using the generalized predictive control algorithm; S3, applying the optimized control vector to the once-through steam generator, and using sensors to perform real-time measurements on the once-through steam generator in online operation to obtain thermal hydraulic data; S4, based on the thermal hydraulic data, the simulation results in S1 are corrected, and steps S2-S4 are repeated to achieve iterative optimization of the control vector, and finally the operating condition change plan is implemented on the once-through steam generator.

2. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 1, characterized in that: In S1, the once-through steam generator simulation calculation model includes an operation simulation model and a fast solution model of the once-through steam generator; The construction method of the running simulation model is as follows: According to the structural parameters and static and dynamic operating characteristics of the once-through steam generator, a one-dimensional thermal analysis model of the full-scale once-through steam generator is constructed as an operating simulation model of the once-through steam generator. The method for constructing the fast solution model is as follows: According to the actual situation of the once-through steam generator, a single heat transfer tube is taken according to the arrangement of the heat transfer tubes to establish a one-dimensional thermal analysis model of the once-through steam generator as a fast solution model, and the fast solution model is used to solve the secondary side pressure drop; The fast solution model interacts with the running simulation model in data, and the primary side inlet and outlet flow rates and the secondary side inlet and outlet flow rates are magnified by corresponding multiples.

3. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 2, characterized in that: Before S1, the operation simulation model is trained offline to build an operation database. The specific method is as follows: According to the operation requirements of the once-through steam generator, a large number of examples with different load change rates and different start and stop conditions are set up and input into the operation simulation model and the fast solution model. Offline training is performed using the synchronous perturbation random approximation method. Through the offline training process, simulation calculation and control vector solution are performed to obtain the operation data and corresponding control vectors under different load change rates and different start and stop conditions. The operation data and the corresponding control vectors are divided into several groups based on the load change rate. The operation data and the corresponding control vectors are arranged and stored in each group based on the initial operating conditions and the target operating conditions to build an operation database.

4. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 3, characterized in that: The S1, inputting the operating condition change plan into the simulation calculation model of the once-through steam generator based on numerical calculation, predicting the dynamic behavior of the once-through steam generator in the future, and obtaining the simulation results, specifically includes: Based on the operating condition change plan, run the simulation model to perform simulation calculations to obtain the expected load change rate and start and end conditions; Based on the constructed operation database, the selection range of the control vector is first determined according to the expected load change rate, and then the required control vector is determined according to the expected start and end conditions. The feedwater flow in the control vector is fitted according to the change trend of the steam flow, and at the same time, the fitting curve is ensured to be equal to the integral of the feedwater flow over time. Finally, this fitting curve is used as the initial value of the control vector during online operation. After obtaining the initial value of the control vector, the control vectors of the forward and reverse disturbances are simultaneously input into the fast solution model to predict the dynamic behavior of the DC evaporator in the future and obtain the simulation results.

5. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 4, characterized in that: S2, based on the simulation results, uses a generalized predictive control algorithm to optimize the control vector of the once-through steam generator, specifically including: The simulation results are processed to obtain the gradient of the loss function under the current control vector, and the control vector is optimized based on this gradient. This process is repeated until the termination criterion is met, and then the first item of the control vector is passed to the running simulation model and the inlet flow is advanced to the next control step.

6. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 5, characterized in that: The simulation results are processed to obtain the gradient of the loss function under the current control vector, and the control vector is optimized according to the gradient, and this process is repeated until the termination criterion is met, specifically including: Under the operation scheme with constant secondary side steam pressure, the square of the difference between the steam pressure per unit time and the rated value is selected as the loss function, and its expression is: Where t′ is the current time, t″ is the simulation time required to quickly solve the model, p is the steam pressure, p0 is the rated steam pressure or target steam pressure, and θ is the current control vector; The current control vector is judged according to the obtained loss function value. When the loss function value is within an acceptable range, the direct-flow steam generator is determined to be in a stable state and the current control vector is output. Otherwise, the current control vector is optimized and this process is repeated until the termination criterion is met.

7. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 6, characterized in that: The optimization of the current control vector is repeated until the termination criterion is met, including: Based on the current control vector, the simulation model is run to update the parameters, and the synchronous perturbation random approximation method is used to generate the perturbation vector, including the forward perturbation vector and the reverse perturbation vector. Through the gradient approximation, the control vector is optimized and iterated. When the maximum number of iterations is reached and the loss function value corresponding to the current control vector is less than the minimum value of the loss function that occurs during the iteration, the iteration ends; If the maximum number of iterations is reached, but the loss function value corresponding to the current control vector is not less than the minimum value of the loss function that occurs during the iteration, the control vector corresponding to the minimum value of the loss function during the iteration is used as the result to end the iteration.

8. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 7, characterized in that: In S4, steps S2-S4 are repeated to achieve the online running iteration termination criterion of iterative optimization of the control vector: The control vector is continuously optimized and iterated. After each iteration, the control vector with the smallest corresponding loss function generated during the iteration process is updated. When the real time passes a control step, the optimization is stopped and the first item of the control vector is directly output. At the same time, the time domain is moved backward to start solving the next control step.

9. The generalized predictive control method for a nuclear power plant once-through steam generator according to claim 7, characterized in that: The method also includes updating the execution database as follows: After the operating condition change plan is implemented on the once-through steam generator, the simulation results of the once-through steam generator are compared and sorted with the measurement data, and the erroneous data are corrected to form a control case and store it in the operation database; After the control case storage is completed, the update of the operation database begins. According to the recording time of a case and the data density in a certain area around the case, when establishing the operation database update mechanism, experts set different data densities in different areas based on experience. Frequently occurring working conditions or working conditions with more complex changes require a larger data density, and vice versa. When updating the running database, first traverse the historical data around the new case in chronological order from old to new. If it is detected that the data density around a case exceeds the local standard, the group of data will be deleted.

Images