Steam turbine valve management method, device, equipment and storage medium

Abstract

The embodiment of the present invention discloses a method, device, equipment and storage medium for managing a steam turbine valve. The method comprises: obtaining a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set; writing the target DEH valve management function into a distributed control system; and controlling the steam turbine valve based on the target DEH valve management function. The technical solution provided by the embodiment of the present invention is based on an algorithm architecture based on knowledge distillation, and optimizes the flow characteristics of the steam turbine unit regulating valve by using the target digital electro-hydraulic control system DEH valve management function determined by a prediction model set, thereby achieving optimized management of the steam turbine valve.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of steam turbine power generation, and in particular to a steam turbine valve management method, device, equipment and storage medium. Background Art

[0002] In the control system of large steam turbines, valve management is achieved through a set of digital electric hydraulic control system (DEH) valve management functions. The DEH valve management function is a numerical representation of the valve flow characteristics. In order to ensure the primary frequency regulation function and automatic power generation control performance of the unit, the consistency of these functions and the actual flow of the valve must be ensured. After the unit is overhauled, flow-through modification, DEH modification, distributed control system (DCS) modification, and operation mode modification, the mismatch between the two is prone to occur. At this time, the turbine valve flow characteristic parameter test and optimization setting should be carried out to ensure the long-term safe and stable operation of the unit's frequency regulation and peak regulation functions.

[0003] At present, in practical applications, the identification of the flow characteristics of steam turbine valves usually involves valve flow characteristics tests, and the flow characteristic curve is calculated based on nozzle flow, characteristic flow or power value. This type of method not only requires a large workload and a high level of operator skills, but also has an adverse effect on the safe and stable operation of the unit due to repeated variable load experiments. In addition, after the test, technicians are required to analyze and process the test data, which will also be affected by the subjective factors of the technicians in this process.

[0004] In addition to experimental methods, some scholars have developed flow characteristic optimization methods based on numerical simulation analysis to reduce the amount of experiments. This method has two strategies, namely data analysis method and simulation platform modeling method, neither of which can completely break away from the valve flow characteristic experiment. The simulation model fitted by data analysis methods such as the least squares method cannot utilize large-scale historical data and has low accuracy. Building a model on a simulation platform is relatively ideal, and its experimental results cannot verify reliability. The latest machine learning-based method can track changes in unit characteristics in a timely manner. However, the currently published scheme still requires a large amount of data from the load increase and decrease stages of the unit, and requires the main steam pressure to be stable, which is almost equivalent to the need for a valve flow characteristic experiment. Moreover, the model is huge in scale, which is not conducive to model migration and is not convenient for rapid promotion to multiple units. Summary of the invention

[0005] The embodiment of the present invention provides a turbine valve management method, device, equipment and storage medium. Based on the algorithm architecture of knowledge distillation, the target digital electro-hydraulic control system DEH valve management function is determined by a prediction model set to optimize the flow characteristics of the turbine unit throttle valve, thereby achieving optimized management of the turbine valve.

[0006] In a first aspect, an embodiment of the present invention provides a method for managing a steam turbine valve, comprising:

[0007] Obtaining a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set;

[0008] Writing the target DEH valve management function into a distributed control system;

[0009] Based on the target DEH valve management function, the turbine valve is controlled.

[0010] In a second aspect, an embodiment of the present invention further provides a steam turbine valve management device, comprising:

[0011] A management function acquisition module, used to acquire a target digital electro-hydraulic control system DEH valve management function determined based on training of a prediction model set;

[0012] A management function writing module, used for writing the target DEH valve management function into the distributed control system;

[0013] The steam turbine valve control module is used to control the steam turbine valve based on the target DEH valve management function.

[0014] In a third aspect, an embodiment of the present invention further provides a computer device, including: a memory and one or more processors;

[0015] The memory is used to store one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the turbine valve management method as described in the first aspect above.

[0017] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, wherein the computer executable instructions, when executed by a computer processor, are used to execute the turbine valve management method as described in the first aspect.

[0018] The embodiment of the present invention discloses a method, device, equipment and storage medium for managing a steam turbine valve. The method comprises: obtaining a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set; writing the target DEH valve management function into a distributed control system; and controlling the steam turbine valve based on the target DEH valve management function. The technical solution provided by the embodiment of the present invention is based on an algorithm architecture based on knowledge distillation, and optimizes the flow characteristics of the steam turbine unit regulating valve by using the target digital electro-hydraulic control system DEH valve management function determined by a prediction model set, thereby achieving optimized management of the steam turbine valve. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic flow chart of a steam turbine valve management method provided in Embodiment 1 of the present invention;

[0020] Figure 2 A schematic diagram of the structure of the opening power prediction model and the opening flow prediction model provided in the first embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the composition of the loss function of the opening flow prediction model provided in the first embodiment of the present invention;

[0022] Figure 4 This is a flowchart of an offline optimization method provided in the first embodiment of the present invention;

[0023] Figure 5 This is a flowchart of an online optimization method provided in the first embodiment of the present invention;

[0024] Figure 6 A structural block diagram of a steam turbine valve management device provided in Embodiment 2 of the present invention;

[0025] Figure 7 This is a structural block diagram of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0027] Embodiment 1

[0028] Figure 1This is a flow chart of a steam turbine valve management method provided in Embodiment 1 of the present invention. This embodiment is applicable to the case of optimizing the management of steam turbine valves. The method can be executed by a steam turbine valve management device, which can be composed of hardware and / or software and can generally be integrated in a computer device. The method specifically includes the following steps:

[0029] S110, obtaining a target digital electro-hydraulic control system DEH valve management function determined based on the prediction model set.

[0030] In this embodiment, in the large steam turbine control system, valve management is implemented through a set of DEH valve management functions, which are numerical representations of valve flow characteristics. In order to ensure the primary frequency regulation function and automatic power generation control performance of the unit, the consistency of these functions and the actual flow of the valve must be ensured. After the unit is overhauled, flow-through modified, DEH modified, DCS modified, and operation mode modified, the problem of mismatch between the two is prone to occur. At this time, the turbine valve flow characteristic parameter test and optimization setting should be carried out to ensure the long-term safe and stable operation of the unit frequency regulation and peak regulation function.

[0031] Therefore, in order to achieve optimal management of the steam turbine valve, it is necessary to obtain the optimized target digital electro-hydraulic control system DEH valve management function, that is, to achieve the optimization of the valve flow characteristics. The ultimate goal of the valve flow characteristics optimization is to obtain the corresponding relationship between the load command and the opening command, that is, to accurately give the opening command according to the load command so that the actual power is consistent with the load command. The valve flow characteristics optimization generally refers to the relationship between the flow command and the opening command, that is, the optimization of the DEH valve management function.

[0032] The optimization method is divided into two types: multi-function and single function. The multi-function method divides the DEH valve management function into back pressure correction function, flow distribution coefficient, overlap function, and flow opening correction function under sequential valve control, and adjusts them separately. The single function method optimizes these functions as a whole, which can not only realize back pressure correction, flow distribution, and control valve overlap distribution, but also implies the actual flow characteristics of each control valve. The present invention adopts the strategy of the single function method.

[0033] Among them, the prediction model set may include an opening power prediction model and an opening flow prediction model, and the opening power prediction model and the opening flow prediction model may be obtained through training with real operation data. Based on the algorithm architecture of knowledge distillation, soft targets related to the teacher network (complex but superior reasoning performance) are introduced as part of the overall loss function to induce the training of the student network (simplified and low complexity) to achieve knowledge transfer. The reasoning performance of the teacher network is usually better than that of the student network, while the model capacity is not specifically limited and is usually large in scale. The student network is small in scale, but after being guided by the teacher network, it can obtain an accuracy close to that of the teacher network. The higher the reasoning accuracy of the teacher network, the more conducive it is to the training of the student network. The opening power prediction model and the opening flow prediction model are used as the teacher network and the student network respectively, and the core variables of the valve flow characteristics such as load, flow, and opening are integrated to obtain the trained opening flow prediction model. Then, the numerical method is used to perform inverse function processing on the opening flow prediction model to obtain the DEH valve management function. During the entire training process, a high-precision, small-scale, and transferable DEH valve management function model can be learned without the need for valve flow characteristic experiments.

[0034] S120, writing the target DEH valve management function into the distributed control system.

[0035] Specifically, the steam turbine digital electro-hydraulic control system DEH is divided into the electronic control part and the hydraulic regulation and safety part. The electronic control is mainly composed of the distributed control system DCS and the DEH special module. The target DEH valve management function is written into the DCS, which can complete tasks such as signal collection, comprehensive calculation, logic processing, and human-machine interface.

[0036] S130. Control the turbine valve based on the target DEH valve management function.

[0037] Specifically, the maintenance personnel accept the optimization results and restart the unit for operation. When receiving the flow command signal, based on the target DEH valve management function, that is, the flow characteristic curve of the turbine valve, the command electrical signal is converted into a hydraulic signal, and finally the valve opening is changed to realize the control of the turbine valve.

[0038] The embodiment of the present invention discloses a method for managing a steam turbine valve, the method comprising: obtaining a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set; writing the target DEH valve management function into a distributed control system; and controlling the steam turbine valve based on the target DEH valve management function. The technical solution provided by the embodiment of the present invention is based on an algorithm architecture based on knowledge distillation, and optimizes the flow characteristics of the steam turbine unit throttle valve by using the target digital electro-hydraulic control system DEH valve management function determined by the prediction model set, thereby achieving optimized management of the steam turbine valve.

[0039] As an optional embodiment of the embodiment of the present invention, based on the above embodiment, the step of determining the target DEH valve management function includes:

[0040] S111. Determine an optimization method for the turbine regulating valve flow characteristics according to the current state of the turbine.

[0041] Among them, the current state of the steam turbine may be in a shutdown state or in an operating state, so the optimization method of the steam turbine regulating valve flow characteristics may include an offline optimization method and an online optimization method. When the steam turbine is in a shutdown state, it can be determined to select the offline optimization method to optimize the steam turbine regulating valve flow characteristics; when the steam turbine is in an operating state, it can be determined to select the online optimization method to optimize the steam turbine regulating valve flow characteristics.

[0042] S112. Based on the optimization method and in combination with the given operating data, the opening power prediction model and the opening flow prediction model are trained.

[0043] The given operation data refers to the historical operation data or current operation data of the steam turbine unit. The given operation data may be the valve opening and actual power of a single valve or a sequential valve. It is clear that when the optimization method is an offline optimization method, the opening power prediction model and the opening flow prediction model can be trained in combination with the historical operation data of the steam turbine unit; when the optimization method is an online optimization method, the opening power prediction model and the opening flow prediction model can be trained in combination with the current operation data of the steam turbine unit.

[0044] Specifically, the machine learning algorithm can learn the relationship between valve opening and actual power based on massive historical data, that is, predict the actual power based on the opening instruction. The input of the opening power prediction model is the opening of a single valve or a sequential valve, and the output is the actual power of the unit. As a teacher network in the knowledge distillation algorithm architecture, it has extremely high accuracy and a large model scale. The training of the opening power prediction model requires more historical data, but there is no load limit, and the rich historical operation data in the DCS can be directly used.

[0045] The training of the opening flow prediction model replaces the throttle flow characteristic experiment and generates a set of opening flow functions, that is, the mapping relationship between the opening command and the flow command. This model is used as the optimization target student network. Its training only requires the knowledge transfer of the teacher network and less historical operation data to achieve higher accuracy. Based on the existing teacher network, the opening flow prediction model can be quickly iterated and optimized.

[0046] S113. Determine the target DEH valve management function according to the opening flow prediction model.

[0047] Specifically, the opening flow prediction model characterizes the mapping relationship between the opening instruction and the flow instruction. According to the opening flow prediction model obtained in step S112, the inverse function of the opening flow prediction model is obtained by adopting a numerical method, and the obtained inverse function is determined as the target DEH valve management function.

[0048] For the offline optimization method, during the unit operation phase, the algorithm system pre-trains the opening power prediction model based on the historical operation data of the steam turbine unit. The opening flow prediction model is optimized during the shutdown setting phase. Offline optimization does not require the initial accuracy of the opening flow prediction model. In order to make the model converge quickly and prevent falling into the local optimum, the opening flow prediction model is zero-initialized and the unit's current DEH valve management function is not introduced.

[0049] As an optional embodiment of the embodiment of the present invention, based on the above embodiment, combined with given operating data, training the opening power prediction model and the opening flow prediction model includes:

[0050] S1121. If the optimization method is an offline optimization method, historical operation data of the steam turbine is collected as given operation data.

[0051] Specifically, if the optimization method is an offline optimization method, historical operation data of the steam turbine is collected as given operation data, and the collected historical operation data is preprocessed, for example, the historical operation data is filtered to achieve data cleaning and obtain effective operation data. The historical operation data may be an opening instruction, a load instruction, a main steam pressure value, etc.

[0052] S1122. Train an opening power prediction model according to historical operation data to obtain a trained opening power prediction model.

[0053] Specifically, the historical operation data is used as sample data to train the opening power prediction model. In this embodiment, the load instruction is used as the true label (ground truth) to train the opening power prediction model, which serves as a teacher network to learn the relationship between valve opening and actual power, that is, to predict actual power based on the opening instruction.

[0054] S1123. Initialize the opening flow prediction model.

[0055] Specifically, when the steam turbine unit is shut down, in order to make the model converge quickly and prevent it from falling into the local optimum, the opening flow prediction model is initialized to zero, and the current DEH valve management function of the steam turbine unit is not introduced.

[0056] S1124. Train an opening flow prediction model based on historical operation data and the trained opening power prediction model.

[0057] It should be noted that the loss function of the opening flow prediction model is divided into two parts, with the real label of the flow instruction as the hard target and the power prediction value of the teacher network as the soft target. Among them, the historical operation data is used as the sample data, and the flow instruction is obtained by dividing the load instruction in the sample data by the main steam pressure value of the sample; in addition, the power prediction value of the teacher network is used as the soft target, and the flow instruction corresponding to the power prediction value is obtained by dividing the power prediction value by the main steam pressure value. In this embodiment, the flow instruction is used as the real label (groundtruth) to train the opening flow prediction model, and as a student network, it learns the relationship between the valve opening and the actual flow, that is, predicts the actual flow according to the opening instruction.

[0058] For the online optimization method, a pre-trained opening power prediction model is used, and the model is fine-tuned online according to the actual power of the operating status. Since the opening power prediction model and the opening flow prediction model are optimized at the same time, online optimization is also a knowledge distillation method for joint training of the teacher-student network, which helps the opening flow prediction model to converge quickly and reduce engineering time. In the online optimization, the algorithm system controls the DCS in a closed loop under the unit operation state and automatically adjusts the valve flow characteristics.

[0059] Optionally, based on the optimization method and in combination with the given operating data, the opening power prediction model and the opening flow prediction model are trained, including:

[0060] S1121', if the optimization method is an online optimization method, then determine the opening flow prediction model according to the current DEH valve management function corresponding to the operation of the steam turbine.

[0061] Among them, a small amount of historical operation data is retrieved to fine-tune the opening power prediction model. Online optimization has high requirements on the initial accuracy of the opening flow prediction model, which should meet the basic mechanism model constraints. Therefore, the experience of maintenance personnel is integrated into the algorithm architecture, that is, the current DEH valve management function setting of the steam turbine unit. Taking the current unit DEH valve management function as the benchmark, the inverse function is used to initialize the opening flow prediction model.

[0062] S1122', taking the current operation data of the steam turbine as the given operation data, combining with the pre-trained opening power prediction model, training the opening flow prediction model, and obtaining the trained opening flow prediction model.

[0063] It should be noted that, similarly, the loss function of the opening flow prediction model in the online optimization method is also divided into two parts, with the real label of the flow command as the hard target and the power prediction value of the teacher network as the soft target. Among them, when the unit is in operation, the current load command of the steam turbine unit is collected every t seconds as sample data, and the flow command is obtained by dividing the main steam pressure value of the sample; in addition, the power prediction value of the teacher network is used as the soft target, and the flow command corresponding to the power prediction value is obtained by dividing the power prediction value by the main steam pressure value. In this embodiment, the flow command is used as the real label (groundtruth) to train the opening flow prediction model, and as a student network, the relationship between the valve opening and the actual flow is learned, that is, the actual flow is predicted according to the opening command.

[0064] S1123', perform inverse function calculation on the trained opening flow prediction model to obtain the trained DEH valve management function.

[0065] Specifically, a numerical method is used to calculate the inverse function of the opening flow prediction model after training that reaches the accuracy index as the DEH valve management function.

[0066] S1124', training the opening power prediction model according to the trained DEH valve management function.

[0067] In this embodiment, the throttle flow characteristics are adjusted according to the DEH valve management function obtained in step 1123', the actual power of the unit is checked, and the opening power prediction model is trained and optimized; before the opening power prediction model reaches the accuracy index, it returns to step S1122' to continue iteration; after the opening power prediction model reaches the accuracy index, the online optimization ends; the maintenance personnel accept the optimization results, and the unit operates normally.

[0068] As an optional embodiment of the embodiment of the present invention, based on the above embodiment, the training step of the opening power prediction model includes:

[0069] a1) Inputting the given operating data into the opening power prediction model to obtain the first output result of the opening power prediction model.

[0070] The given operation data also includes corresponding first real label data.

[0071] It is clear that for the offline optimization method, the opening power prediction model is trained by using historical operation data as sample data; for the online optimization method, the opening power prediction model is trained by using a small amount of historical operation data and current operation data as sample data. Among them, the first output result can be specifically understood as the power prediction value, and the first real label data can be specifically understood as the load instruction.

[0072] Specifically, the single valve or sequence valve opening instruction of the steam turbine unit is input into the opening power prediction model, and the power prediction value of the unit is output as the first output result. The load instruction in the sample data is used as the first real label data and compared with the actual power prediction value.

[0073] b1) Based on the first output result and the first true label data, determine the loss function of the opening power prediction model.

[0074] Specifically, the loss function of the opening power prediction model adopts the root mean square error (RMSE) commonly used in regression prediction tasks, and performs root mean square calculation on the power prediction value (i.e., the first output result) and the load instruction (i.e., the first real data label). The formula is as follows: Where m is the number of samples, X={x1,x2......x m} is the sequence valve opening instruction, h is the power prediction value, and y is the load instruction.

[0075] c1) adjusting the parameters of the opening power prediction model according to the loss function of the opening power prediction model, returning to continue to execute inputting the given operating data into the opening power prediction model, obtaining the first output result of the opening power prediction model, until the accuracy of the output result of the opening power prediction model meets the first set condition.

[0076] Among them, the accuracy of the output result of the opening power prediction model satisfies the first set condition, which can be understood as the opening power prediction model can accurately map the relationship between the opening instruction and the load instruction, or it can be understood as the loss function of the opening power prediction model is less than a certain set threshold. Specifically, before the opening power prediction model reaches the accuracy index, it returns to step a1) to continue iteratively optimizing the opening power prediction model until the opening power prediction model reaches the accuracy index and the optimization ends.

[0077] As an optional embodiment of the embodiment of the present invention, based on the above embodiment, the training step of the opening flow prediction model includes:

[0078] a2) Calculate the given operating data to obtain the calculated given operating data.

[0079] The calculated given operating data includes the second real label data. It is clear that for the offline optimization method, the opening power prediction model is trained by using the historical operating data as sample data; for the online optimization method, the opening power prediction model is trained by using a small amount of historical operating data and current operating data as sample data. The first real label data can be specifically understood as a flow instruction. Specifically, the given operating data is used as sample data, and the load instruction in the sample data is divided by the main steam pressure value in the sample data to obtain the flow instruction.

[0080] b2) Determine the second prediction label data based on the given operating data and the opening power prediction model.

[0081] Specifically, the given operating data is used as sample data, the load instruction in the sample data is input into the opening power prediction model to obtain the output power prediction value, and the power prediction value is divided by the main steam pressure value to obtain the predicted flow instruction as the second prediction label data.

[0082] c2) inputting the given operating data into the opening flow prediction model to obtain a second output result of the opening flow prediction model.

[0083] Specifically, the single valve or sequence valve opening instruction of the steam turbine unit is input into the opening flow prediction model, and the flow prediction value of the unit is output as the second output result of the opening flow prediction model.

[0084] d2) Determine the loss function of the opening flow prediction model based on the second output result, the second predicted label and the second real label data.

[0085] In this embodiment, the training opening flow prediction model replaces the throttle flow characteristic experiment to generate a set of opening flow functions, that is, the mapping relationship between the opening instruction and the flow instruction, and the inverse function is the DEH valve management function. This model serves as the optimization target student network, and its training only requires the knowledge transfer of the teacher network and less machine history data to achieve higher accuracy. On the basis of the existing teacher network, the opening flow prediction model can be quickly iterated and optimized. The loss function of the opening flow prediction model is divided into two parts, with the real label of the flow instruction as the hard target and the power prediction value of the opening power prediction model (teacher network) as the soft target. The loss function is determined based on the second output result and the second real label data, and the loss function is determined based on the second output result and the second predicted label. The two parts of the loss function are calculated according to the weight coefficient, and finally the loss function of the opening flow prediction model is determined.

[0086] Figure 2 The schematic diagram of the opening power prediction model and the opening flow prediction model provided in the first embodiment of the present invention is as follows: Figure 2As shown, the present invention uses the load instruction as the true label (ground truth) to train the opening power prediction model (denoted as model1) as the teacher network; and uses the flow instruction as the true label to train the opening flow prediction model (denoted as model2) as the student network. The load instruction and the flow instruction are replaced with each other by multiplying and dividing the main steam pressure value. The point-by-point inversion function of model2 after knowledge distillation is obtained to obtain the DEH valve management function required for the valve flow characteristic optimization task. Specifically, the opening instruction is input into the opening power prediction model, and the output actual power is compared with the load instruction as the true label data to train the opening prediction model; the opening instruction is input into the opening flow prediction model, and the actual flow is output. The power prediction value output by the opening power prediction model and the load instruction in the sample are divided by the main steam pressure value to obtain the flow instruction, and the flow instruction is used as the true label data to compare with the actual flow to train the opening flow prediction model.

[0087] Optionally, determining a loss function of the opening flow prediction model based on the second output result, the second predicted label and the second real label data includes:

[0088] d21) Determine a first loss function based on the second output result and the second true label data.

[0089] Specifically, the first loss function uses the root mean square error (RMSE) commonly used in regression prediction tasks, and performs root mean square calculation on the traffic prediction value (i.e., the second output result) and the traffic instruction (i.e., the second real data label). The formula is as follows: Where m is the number of samples, X={x1,x2......x m} is the sequence valve opening instruction, g(xi) is the flow prediction value, and zi is the flow instruction.

[0090] d22) Determine a second loss function based on the second output result and the second predicted label data.

[0091] Specifically, the second loss function uses the root mean square error (RMSE) commonly used in regression prediction tasks, and performs root mean square calculation on the traffic prediction value (i.e., the second output result) and the predicted traffic instruction (i.e., the second predicted label data). The formula is as follows: Where m is the number of samples, X={x1,x2......x m} is the sequence valve opening instruction, where h is the power prediction value, g is the flow prediction value, and Pi is the main steam pressure value.

[0092] d23) Obtain the ratio of the first loss function to the second loss function.

[0093] Among them, the loss function of the opening flow prediction model is designed as the weighted sum of the RMSE values ​​corresponding to the soft target and the hard target. The proportion coefficient of the first loss function and the second loss function determines the contribution value of the teacher network and the student network. Exemplarily, the proportion coefficient of the first loss function can be expressed as: α, and the proportion coefficient of the second loss function can be expressed as: 1-α.

[0094] d24) Performing weighted summation of the first loss function and the second loss function according to the proportion coefficient to determine the loss function of the opening flow prediction model.

[0095] The loss function of the opening flow prediction model is designed as the weighted sum of the RMSE values ​​corresponding to the soft target and the hard target. The larger the weight coefficient of the soft target, the more the transfer induction depends on the contribution of the teacher network. This is very necessary in the early stage of training, which helps the student network to more easily identify simple samples. However, in the later stage of training, the proportion of soft targets needs to be appropriately reduced to allow real annotations to help identify difficult samples. The loss function formula of the opening flow prediction model is as follows: Where m is the number of samples, X={x1,x2......x m} is the sequence valve opening instruction, h is the power prediction value, g is the flow prediction value, Pi is the main steam pressure value, α is the proportion coefficient of the first loss function, and (1-α) is the proportion coefficient of the second loss function.

[0096] Figure 3 The schematic diagram of the loss function of the opening flow prediction model provided in the first embodiment of the present invention is as follows: Figure 3 As shown, the opening instruction is input as the input vector Xi into the teacher network (i.e., the opening power prediction model) and the student network (i.e., the opening flow prediction model), the true label of the flow instruction is used as the hard target, and the power prediction value of the teacher network is used as the soft target. The weight of the hard target is expressed as: α, and the weight of the soft target is expressed as 1-α. The RMSE values ​​corresponding to the soft target and the hard target are weighted and summed to determine the loss function of the opening flow prediction model.

[0097] e2) adjusting the parameters of the opening flow prediction model according to the loss function of the opening flow prediction model, returning to continue to execute inputting the given operating data into the opening flow prediction model, and obtaining the second output result of the opening flow prediction model, until the accuracy of the output result of the opening flow prediction model meets the second set condition.

[0098] Among them, the accuracy of the output result of the opening flow prediction model satisfies the second set condition, which can be understood as the opening flow prediction model can accurately map the relationship between the opening instruction and the flow instruction, or it can be understood as the loss function of the opening flow prediction model is less than a certain set threshold. Specifically, before the opening flow prediction model reaches the accuracy index, it returns to step a2) to continue iteratively optimizing the opening flow prediction model, and the optimization ends after the opening flow prediction model reaches the accuracy index.

[0099] Based on the teacher network model, the student network model can be quickly produced according to the actual operation of each unit and a small amount of data. A high-precision model can be obtained without calibrating a large amount of data, with a small amount of engineering and low maintenance costs. The algorithm architecture of knowledge distillation is adopted, and the obtained model is small in scale under the same accuracy conditions, fast in calculation, and easy to download on site and real-time closed-loop optimization.

[0100] In order to more clearly describe the embodiments of the present invention, Figure 4 This is an example flow chart of the offline optimization method provided in the first embodiment of the present invention, such as Figure 4 As shown, the offline optimization process can be expressed as follows:

[0101] (1) Preprocessing historical operation data;

[0102] (2) Use historical operation data to train the opening power prediction model (denoted as model1), and obtain the actual power according to the prediction of model1;

[0103] (3) The unit is shut down and the opening flow prediction model is initialized to zero (denoted as model2);

[0104] (4) For each sample load command input, divide it by the sample main steam pressure to obtain the flow command;

[0105] (5) Predicting the actual flow rate based on the current opening flow prediction model model2, and calculating the loss function of the opening flow prediction model together with the power and flow command predicted by model1;

[0106] (6) Return to (4) before model2 reaches the accuracy index and continue to iteratively optimize model2;

[0107] (7) When model2 reaches the accuracy index, the optimization ends, and the inverse function of model2 is obtained by numerical method to obtain the DEH valve management function.

[0108] Figure 5 This is a flow chart of an online optimization method provided in the first embodiment of the present invention, such as Figure 5 As shown, the process of online optimization can be expressed as follows:

[0109] (1) Retrieve a small amount of historical operating data to fine-tune the opening power prediction model (denoted as model1), and predict the actual power based on model1;

[0110] (2) Taking the current unit DEH valve management function as the benchmark, the inverse function is used to initialize the opening flow prediction model (denoted as model2);

[0111] (3) When the unit is in operation, the unit's current load command is collected every t seconds and divided by the sample's main steam pressure to obtain the flow command;

[0112] (4) Predict the actual flow rate based on the current opening flow prediction model model2, calculate the loss function of the opening flow prediction model together with the power and flow command predicted by model1, and optimize model2;

[0113] (5) Return to (3) before model2 reaches the accuracy index and continue to iteratively optimize model2;

[0114] (6) After model2 reaches the accuracy index, the updated model2 is inverted using the numerical method as the DEH valve management function;

[0115] (7) According to the DEH valve management function obtained in (6), the flow characteristics of the regulating valve are adjusted, the actual power of the unit is checked, and the optimization model 1 is trained;

[0116] (8) Return to (4) and continue iterating before model1 reaches the accuracy index;

[0117] (9) After model1 reaches the accuracy index, the online optimization ends, the maintenance personnel accept the optimization results, and the unit operates normally.

[0118] As an optional embodiment, this embodiment specifies the training process of the opening power prediction model and the opening flow prediction model, combines the knowledge distillation technology with the optimization of the valve flow characteristics, and the model is small in scale and easy to maintain, which is conducive to flexible migration between multiple units. It also specifies the steps of the online optimization method and the offline optimization method, does not rely on the valve flow characteristic experiment, and completely adjusts and optimizes the valve flow characteristics based on historical data and real-time operation data, reducing labor costs and equipment maintenance costs. Maintain the consistency between load instructions, flow instructions, and opening instructions to ensure the safety and controllability of the turbine unit. Compared with other numerical simulation methods, it has lower requirements on data quantity and quality. The offline optimization scheme can directly use historical operation data without continuous variable load data. The online optimization scheme does not even require the user to provide historical data, and it can iteratively optimize only based on real-time operation data. You can choose offline or online setting optimization and adjust the optimization scheme according to actual needs.

[0119] Embodiment 2

[0120] Figure 6 This is a structural block diagram of a steam turbine valve management device provided in the second embodiment of the present invention. This embodiment is applicable to the situation of optimizing the management of steam turbine valves. The device can be composed of hardware and / or software and can generally be integrated into a computer device. Figure 6 As shown, the device may specifically include: a management function acquisition module 21 , a management function writing module 22 and a turbine valve control module 23 .

[0121] A management function acquisition module 21 is used to acquire a target digital electro-hydraulic control system DEH valve management function determined based on training of a prediction model set;

[0122] A management function writing module 22 is used to write the target DEH valve management function into the distributed control system;

[0123] The steam turbine valve control module 23 is used to control the steam turbine valve based on the target DEH valve management function.

[0124] Optionally, the management function acquisition module 21 includes:

[0125] An optimization method determination unit is used to determine the optimization method of the flow characteristics of the steam turbine regulating valve according to the current state of the steam turbine;

[0126] A prediction model training unit, used to train an opening power prediction model and an opening flow prediction model based on an optimization method and in combination with given operating data;

[0127] The valve management function determination unit is used to determine the target DEH valve management function according to the opening flow prediction model.

[0128] Optionally, the prediction model training unit is specifically used for:

[0129] If the optimization method is an offline optimization method, the historical operation data of the steam turbine is collected as the given operation data;

[0130] According to the historical operation data, the opening power prediction model is trained to obtain the trained opening power prediction model;

[0131] Initialize the opening flow prediction model;

[0132] According to the historical operation data, combined with the trained opening power prediction model, the opening flow prediction model is trained.

[0133] Optionally, the prediction model training unit is specifically used for:

[0134] If the optimization method is an online optimization method, the opening flow prediction model is determined according to the current DEH valve management function corresponding to the operation of the steam turbine;

[0135] The current operation data of the steam turbine is used as the given operation data, and the opening flow prediction model is trained in combination with the pre-trained opening power prediction model to obtain the trained opening flow prediction model;

[0136] Perform inverse function calculation on the trained opening flow prediction model to obtain the trained DEH valve management function;

[0137] According to the trained DEH valve management function, the opening power prediction model is trained.

[0138] Optionally, the prediction model training unit includes:

[0139] A first output result obtaining subunit is used to input the given operation data into the opening power prediction model to obtain a first output result of the opening power prediction model, wherein the given operation data also includes corresponding first real label data;

[0140] A first loss function determination subunit, used to determine a loss function of an opening power prediction model based on a first output result and first true label data;

[0141] The opening power prediction model adjustment subunit is used to adjust the parameters of the opening power prediction model according to the loss function of the opening power prediction model, return to continue executing the input of the given operating data into the opening power prediction model, and obtain the first output result of the opening power prediction model until the accuracy of the output result of the opening power prediction model meets the first set condition.

[0142] Optionally, the prediction model training unit includes:

[0143] A calculation subunit, used to calculate the given operation data to obtain the calculated given operation data, wherein the calculated given operation data includes the second real label data;

[0144] A second prediction label data subunit is used to determine second prediction label data based on given operating data and an opening power prediction model;

[0145] A second output result obtaining subunit is used to input the given operation data into the opening flow prediction model to obtain a second output result of the opening flow prediction model;

[0146] A second loss function determination subunit, used to determine the loss function of the opening flow prediction model based on the second output result, the second predicted label and the second real label data;

[0147] The opening flow prediction model adjustment subunit is used to adjust the parameters of the opening flow prediction model according to the loss function of the opening flow prediction model, return to continue executing the input of the given operating data into the opening flow prediction model, and obtain the second output result of the opening flow prediction model until the accuracy of the output result of the opening flow prediction model meets the second set condition.

[0148] Optionally, the second loss function determination subunit is specifically used for:

[0149] Determine a first loss function based on the second output result and the second true label data;

[0150] Determining a second loss function based on the second output result and the second predicted label data;

[0151] Obtain the ratio of the first loss function to the second loss function;

[0152] The first loss function and the second loss function are weightedly summed according to the proportion coefficient to determine the loss function of the opening flow prediction model.

[0153] The above device can execute the turbine valve management method provided by all the above embodiments of the present invention, and has the corresponding functional modules and beneficial effects of executing the above methods. For technical details not described in detail in this embodiment, please refer to the methods provided by all the above embodiments of the present invention.

[0154] Embodiment 3

[0155] Figure 7 A structural block diagram of a computer device provided in Embodiment 3 of the present invention is shown in FIG. Figure 7 As shown, the computer device includes a processor 31, a memory 32, an input device 33 and an output device 34; the number of processors 31 in the computer device can be one or more. Figure 7 A processor 31 is taken as an example; the processor 31, memory 32, input device 33 and output device 34 in the computer device can be connected by a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0156] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the modules corresponding to the steam turbine valve management method in the embodiment of the present invention (for example, the management function acquisition module 21, the management function writing module 22 and the steam turbine valve control module 23 in the steam turbine valve management device). The processor 31 executes various functional applications and data processing of the computer device by running the software programs, instructions and modules stored in the memory 32, that is, realizes the above-mentioned steam turbine valve management method.

[0157] The memory 32 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 32 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 32 may further include a memory remotely arranged relative to the processor 31, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0158] The input device 33 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the computer device. The output device 34 may include a display device such as a display screen.

[0159] Embodiment 4

[0160] Embodiment 4 of the present invention further provides a storage medium comprising computer executable instructions, wherein the computer executable instructions are used to execute a steam turbine valve management method when executed by a computer processor, the method comprising:

[0161] Obtaining a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set;

[0162] Write the target DEH valve management function into the distributed control system;

[0163] Based on the target DEH valve management function, the turbine valve is controlled.

[0164] Of course, the computer executable instructions of a storage medium including computer executable instructions provided in an embodiment of the present invention are not limited to the operations of the method described above, but can also execute related operations in the turbine valve management method provided in any embodiment of the present invention.

[0165] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0166] It is worth noting that in the above-mentioned embodiment of the turbine valve management device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0167] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A steam turbine valve management method, characterized in that: include: Obtain a target digital electro-hydraulic control system DEH valve management function determined based on a prediction model set, wherein the prediction model set includes an opening power prediction model and an opening flow prediction model, and the step of determining the target DEH valve management function includes: based on an algorithm architecture of knowledge distillation, using the opening power prediction model and the opening flow prediction model as a teacher network and a student network respectively to obtain a trained opening flow prediction model, and using a numerical method to perform inverse function processing on the trained opening flow prediction model to obtain the target DEH valve management function; Writing the target DEH valve management function into a distributed control system; Based on the target DEH valve management function, the turbine valve is controlled.

2. The method according to claim 1, characterized in that The step of determining the target DEH valve management function comprises: Determining an optimization method for the flow characteristics of the steam turbine valve according to the current state of the steam turbine; Based on the optimization method and in combination with given operating data, an opening power prediction model and an opening flow prediction model are trained; The target DEH valve management function is determined according to the opening flow prediction model.

3. The method according to claim 2, characterized in that Based on the optimization method, combined with given operation data, training the opening power prediction model and the opening flow prediction model includes: If the optimization method is an offline optimization method, collecting historical operation data of the steam turbine as given operation data; According to the historical operation data, the opening power prediction model is trained to obtain a trained opening power prediction model; Initializing the opening flow prediction model; The opening flow prediction model is trained based on the historical operation data and in combination with the trained opening power prediction model.

4. The method according to claim 2, characterized in that: Based on the optimization method, combined with given operation data, training the opening power prediction model and the opening flow prediction model includes: If the optimization method is an online optimization method, the opening flow prediction model is determined according to the current DEH valve management function corresponding to the operation of the steam turbine; The current operating data of the steam turbine is used as given operating data, and combined with the pre-trained opening power prediction model, the opening flow prediction model is trained to obtain the trained opening flow prediction model; Performing inverse function calculation on the trained opening flow prediction model to obtain a trained DEH valve management function; The opening power prediction model is trained according to the trained DEH valve management function.

5. The method according to claim 2, characterized in that: The training step of the opening power prediction model includes: Inputting the given operating data into the opening power prediction model to obtain a first output result of the opening power prediction model, wherein the given operating data also includes corresponding first real label data; Determining a loss function of the opening power prediction model based on the first output result and the first real label data; Adjust the parameters of the opening power prediction model according to the loss function of the opening power prediction model, return to continue executing the input of the given operating data into the opening power prediction model, obtain the first output result of the opening power prediction model, and until the accuracy of the output result of the opening power prediction model meets the first set condition.

6. The method according to claim 2, characterized in that The training steps of the opening flow prediction model include: Calculating the given operating data to obtain calculated given operating data, wherein the calculated given operating data includes second real label data; Determining second prediction label data based on the given operating data and the opening power prediction model; Inputting the given operating data into the opening flow prediction model to obtain a second output result of the opening flow prediction model; Determine a loss function of the opening flow prediction model based on the second output result, the second predicted label and the second real label data; Adjust the parameters of the opening flow prediction model according to the loss function of the opening flow prediction model, return to continue executing the input of the given operating data into the opening flow prediction model, obtain the second output result of the opening flow prediction model, until the accuracy of the output result of the opening flow prediction model meets the second set condition.

7. The method according to claim 6, characterized in that The determining the loss function of the opening flow prediction model based on the second output result, the second predicted label and the second real label data includes: Determine a first loss function based on the second output result and the second true label data; Determining a second loss function based on the second output result and the second predicted label data; Obtaining the ratio of the first loss function to the second loss function; The first loss function and the second loss function are weightedly summed according to the proportion coefficient to determine the loss function of the opening flow prediction model.

8. A steam turbine valve management device, characterized in that: include: A management function acquisition module is used to acquire a target digital electro-hydraulic control system DEH valve management function determined based on training of a prediction model set, wherein the prediction model set includes an opening power prediction model and an opening flow prediction model, and the step of determining the target DEH valve management function includes: based on an algorithm architecture of knowledge distillation, using the opening power prediction model and the opening flow prediction model as a teacher network and a student network respectively to obtain a trained opening flow prediction model, and using a numerical method to perform inverse function processing on the trained opening flow prediction model to obtain the target DEH valve management function; A management function writing module, used for writing the target DEH valve management function into the distributed control system; The steam turbine valve control module is used to control the steam turbine valve based on the target DEH valve management function.

9. A computer device, characterized in that: include: memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steam turbine valve management method as described in any one of claims 1-7.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions are used to execute the steam turbine valve management method as described in any one of claims 1-7 when executed by a computer processor.

Images