A method for real-time dynamic prediction of the state of a hydroelectric generating unit based on topological multi-task learning

Abstract

This invention proposes a real-time dynamic prediction method for the status of hydropower units based on topological multi-task learning. This method involves in-depth analysis of the historical operating data and status trends of the entire unit, utilizing intelligent algorithms to construct a predictive model, and achieving accurate prediction of the unit's future operating status. This method not only enhances the intelligent operation and maintenance capabilities of hydropower units, provides early warnings of potential faults, and optimizes maintenance strategies, but also effectively reduces operation and maintenance costs, extends equipment lifespan, improves power generation efficiency, and ensures long-term stable operation of the unit, thereby comprehensively improving the safety, economic benefits, and intelligent management level of hydropower units.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for industrial equipment, and in particular to a method for real-time dynamic prediction of the status of hydropower units based on topology multi-task learning. Background Technology

[0002] The safe and stable operation of hydropower generators is directly related to the reliability of the power grid, the security of energy supply, and economic benefits. During long-term operation, hydropower units may experience problems such as vibration, abnormal sway, bearing wear, rotor imbalance, and electrical faults. Traditional periodic maintenance methods may result in over-maintenance or failure to detect faults in a timely manner. Based on big data, machine learning, and other technologies, the timing of fault occurrence can be predicted in advance. This not only prevents faults, reduces costs, and extends unit lifespan, but also ensures power grid safety, improves economic efficiency, and promotes the intelligent development of the hydropower industry.

[0003] Real-time dynamic prediction of hydropower unit status faces numerous challenges in terms of data quality, modeling accuracy, real-time calculation, model adaptability, and industrial applications. Equipment operation is influenced by a variety of factors, including mechanical, electrical, hydraulic, and thermal factors, resulting in complex status changes that are difficult to model accurately. Online operating data is extremely complex, and unit status is affected by load fluctuations, seasonal variations, and external disturbances, leading to uncertainties in the prediction model. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the prior art and to propose a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning.

[0005] This invention is achieved through the following technical solution: This invention proposes a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning, the method comprising: S1. Obtain real-time status data during unit operation as the output of the model; S2. Obtain real-time operating data during unit operation and status data during the previous unit operation as input to the model; S3. Divide the reference data of the generating units according to their different basic information and establish a reference data set belonging to each generating unit. S4. Divide the unit operation process into several stages based on the speed data; S5. Establish a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning. S6. Real-time dynamic prediction of unit status is performed using a hydropower unit status real-time dynamic prediction model based on topology multi-task learning. S7. Select data from units with similar operating conditions to compare with the predicted values. If the error exceeds the limit, correct it. If it is within the range, use it as the prediction input for the next moment to improve the prediction accuracy. S8. Regularly update the historical data set.

[0006] Furthermore, the real-time status data during the operation of the S1 unit includes: horizontal vibration of the upper frame, vertical vibration of the upper frame, horizontal vibration of the lower frame, vertical vibration of the lower frame, horizontal vibration of the stator base, vertical vibration of the stator base, horizontal vibration of the stator core, vertical vibration of the stator core, X-direction runout of the upper guide bearing, Y-direction runout of the upper guide bearing, X-direction runout of the lower guide bearing, Y-direction runout of the lower guide bearing, X-direction runout of the water guide bearing, and Y-direction runout of the water guide bearing.

[0007] Furthermore, the real-time operating data during the operation of the S2 unit includes: nozzle inlet pressure, nozzle opening, deflector opening, water distribution ring pipe inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil tank temperature, guide bearing oil level, and unit speed.

[0008] Furthermore, the basic information of the unit in S3 includes: rated speed, rated output power, inertia constant, rated voltage, thrust load, rated excitation current, number of guide bearings, rated head, main shaft diameter, and impeller diameter.

[0009] Furthermore, the method for dividing the speed data during unit operation in S4 is as follows: 0% to 25% of the rated speed is the low-speed zone, which causes initial vibration; 25% to 75% of the rated speed is the medium-speed zone, where the speed increases and the unit will pass through the resonance zone; 75% to 100% of the rated speed is the high-speed zone, where the unit's damping and control system begin to play a full role and the system enters a dynamic equilibrium state.

[0010] Furthermore, S5 establishes a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning, specifically as follows: Training data for a real-time dynamic prediction model of hydropower unit status is constructed using historical datasets. The nonlinear estimation function is then transformed into a linear estimation function in a high-dimensional feature space using the constructed optimal decision function.

[0011] in N The number of samples. x k Each sample contains unit operating status information. y pre To predict the unit's status at the next moment. w T These are the regression coefficients of the real-time dynamic prediction model for unit status. b The deviation of the real-time dynamic prediction model for unit status; The radial basis function kernel uses kernel tricks to map low-dimensional data to a high-dimensional space, solving nonlinear problems; the objective function is written as:

[0012] in and As a hyperparameter, the overall task loss L The loss consists of three parts. The task loss, which is the sum of the root mean square errors of each task, can be written as:

[0013] T The number of tasks. , and The first i The true and predicted values ​​output by the task model Indicates the first i The output of each task is uncertain; L topo For topological loss, the topological differences between tasks are calculated through persistent cohomology, specifically by calculating the Wasserstein distance between tasks to ensure that the shared topological structure of different tasks is similar.

[0014] in N The number of data points in each task. , It is a task i Sorted predicted values It is a task j The sorted predicted values, this distance measures the difference in the distribution of the outputs of the two tasks; As a regularization term, the squared norm of the weight parameters is added to the total loss. This utilizes the principle of structural risk minimization to make the parameter values ​​smaller, reducing the risk of overfitting during training. The stochastic gradient descent method is then used to find the solution that minimizes the loss function. w and b The stochastic gradient descent strategy is as follows:

[0015] in a It is a constant. and The loss function is respectively in w and Gradient of direction.

[0016] Furthermore, the specific formula for real-time dynamic prediction of unit status in S6 is as follows:

[0017] in y t+1 For the output of the next time step, MTL t+1 According to t The learning curve fitted by topological multi-task learning at time +1. x t for t Input at any time y t for t Output at all times.

[0018] Furthermore, in S8, the model is incrementally updated periodically on the historical data set, irrelevant datasets are deleted, and the adaptability of real-time prediction of unit status is improved.

[0019] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the real-time dynamic prediction method for the state of a hydropower unit based on topology multi-task learning.

[0020] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the real-time dynamic prediction method for the state of a hydropower unit based on topology multi-task learning.

[0021] The beneficial effects of this invention are: This invention proposes a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning. This method provides a better solution for research in the field of real-time state prediction of hydropower units and other black-box prediction problems involving multiple objectives, multiple variables, and high prediction accuracy. Through this innovative method, the following technical effects can be achieved: 1. By utilizing topological loss to achieve multi-task prediction based on the correlation between output tasks, each output task takes the correlation between them as input and performs joint processing through shared representation. This makes it the correlation between output tasks, improving generalization ability, data utilization efficiency, and computational performance. 2. During the prediction process, this method effectively models the complex relationships between tasks and projects data into a high-dimensional feature space through nonlinear mapping, thereby capturing richer structural information. Compared to traditional linear methods, this method can handle nonlinear problems more accurately, enabling the model to learn more precise task associations. Simultaneously, kernel tricks avoid explicitly computing high-dimensional mappings, reducing computational costs and improving training and inference efficiency. 3. Real-time prediction of the status information during unit operation, real-time tracking of vibration and sway information at various locations of the unit, overcoming the defects of on-site time delay, improving grid stability, enhancing the supporting role of hydropower in renewable energy, improving the safety of unit operation, reducing the risk of equipment damage, and providing valuable reference information for operators' control. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 Flowchart of a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning; Figure 2 A schematic diagram of the control flow of a real-time dynamic prediction method for the state of a hydropower unit based on topological multi-task learning. Figure 3 This is a structural diagram of a topology-based multi-task learning prediction method. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention proposes a real-time dynamic prediction method for the status of hydropower units based on topological multi-task learning. This method deeply analyzes the historical operating data and status trends of the entire unit, utilizes intelligent algorithms to construct a predictive model, and achieves accurate prediction of the unit's future operating status. This method not only enhances the intelligent operation and maintenance capabilities of hydropower units, provides early warnings of potential faults, and optimizes maintenance strategies, but also effectively reduces operation and maintenance costs, extends equipment lifespan, improves power generation efficiency, and ensures long-term stable operation of the unit, thereby comprehensively improving the safety, economic benefits, and intelligent management level of hydropower units.

[0026] Specifically, in combination Figures 1-3 This invention proposes a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning, the method comprising: S1. Obtain real-time status data during unit operation as the output of the model; S2. Obtain real-time operating data during unit operation and status data during the previous unit operation as input to the model; S3. Divide the reference data of the generating units according to their different basic information and establish a reference data set belonging to each generating unit. S4. Divide the unit operation process into several stages based on the speed data; S5. Establish a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning. S6. Real-time dynamic prediction of unit status is performed using a hydropower unit status real-time dynamic prediction model based on topology multi-task learning. S7. Select data from units with similar operating conditions to compare with the predicted values. If the error exceeds the limit, correct it. If it is within the range, use it as the prediction input for the next moment to improve the prediction accuracy. S8. Regularly update the historical data set.

[0027] Furthermore, the real-time status data during the operation of the S1 unit includes: horizontal vibration of the upper frame, vertical vibration of the upper frame, horizontal vibration of the lower frame, vertical vibration of the lower frame, horizontal vibration of the stator base, vertical vibration of the stator base, horizontal vibration of the stator core, vertical vibration of the stator core, X-direction runout of the upper guide bearing, Y-direction runout of the upper guide bearing, X-direction runout of the lower guide bearing, Y-direction runout of the lower guide bearing, X-direction runout of the water guide bearing, and Y-direction runout of the water guide bearing.

[0028] Furthermore, the real-time operating data during the operation of the S2 unit includes: nozzle inlet pressure, nozzle opening, deflector opening, water distribution ring pipe inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil tank temperature, guide bearing oil level, and unit speed.

[0029] Furthermore, the basic information of the unit in S3 includes: rated speed, rated output power, inertia constant, rated voltage, thrust load, rated excitation current, number of guide bearings, rated head, main shaft diameter, and impeller diameter.

[0030] Furthermore, the method for dividing the speed data during unit operation in S4 is as follows: 0% to 25% of the rated speed is the low-speed zone, which causes initial vibration; 25% to 75% of the rated speed is the medium-speed zone, where the speed increases and the unit will pass through the resonance zone; 75% to 100% of the rated speed is the high-speed zone, where the unit's damping and control system begin to play a full role and the system enters a dynamic equilibrium state.

[0031] Furthermore, S5 establishes a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning, specifically as follows: Training data for a real-time dynamic prediction model of hydropower unit status is constructed using historical datasets. The nonlinear estimation function is then transformed into a linear estimation function in a high-dimensional feature space using the constructed optimal decision function.

[0032] in N The number of samples. x k Each sample contains unit operating status information. y pre To predict the unit's status at the next moment. w T These are the regression coefficients of the real-time dynamic prediction model for unit status. b The deviation of the real-time dynamic prediction model for unit status; The radial basis function kernel uses kernel tricks to map low-dimensional data to a high-dimensional space, solving nonlinear problems; the objective function is written as:

[0033] in and These are hyperparameters used to measure the strength of each algorithm during task training, and the overall task loss. L The loss consists of three parts. The task loss, which is the sum of the root mean square errors of each task, can be written as:

[0034] T The number of tasks. , and The first i The true and predicted values ​​output by the task model Indicates the first i The output uncertainty of each task is used to balance the dynamic loss weights of uncertainty for each task in multi-task learning. L topo For topological loss, the topological differences between tasks are calculated through persistent cohomology, specifically by calculating the Wasserstein distance between tasks to ensure that the shared topological structure of different tasks is similar.

[0035] in N The number of data points in each task. , It is a task i Sorted predicted values It is a task jThe sorted predicted values, this distance measures the difference in the distribution of the outputs of the two tasks. The smaller the distance, the more similar the two tasks are. As a regularization term, the squared norm of the weight parameters is added to the total loss. This utilizes the principle of structural risk minimization to make the parameter values ​​smaller, reducing the risk of overfitting during training. The stochastic gradient descent method is then used to find the solution that minimizes the loss function. w and b The stochastic gradient descent strategy is as follows:

[0036] in a It is a constant. and The loss function is respectively in w and Gradient of direction.

[0037] Furthermore, the specific formula for real-time dynamic prediction of unit status in S6 is as follows:

[0038] in y t+1 The output for the next moment includes: data on the horizontal vibration of the upper frame, the vertical vibration of the upper frame, the horizontal vibration of the lower frame, the vertical vibration of the lower frame, the horizontal vibration of the stator base, the vertical vibration of the stator base, the horizontal vibration of the stator core, the vertical vibration of the stator core, the X-direction runout of the upper guide bearing, the Y-direction runout of the upper guide bearing, the X-direction runout of the lower guide bearing, the Y-direction runout of the lower guide bearing, the X-direction runout of the water guide bearing, and the Y-direction runout of the water guide bearing. MTL t+1 According to t The learning curve fitted by topological multi-task learning at time +1. x t for t The inputs at any given time include: nozzle inlet pressure, nozzle opening, deflector opening, water distribution ring inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil sump temperature, guide bearing oil level, and unit speed. y t for t The real-time output includes data for: horizontal vibration of the upper frame, vertical vibration of the upper frame, horizontal vibration of the lower frame, vertical vibration of the lower frame, horizontal vibration of the stator base, vertical vibration of the stator base, horizontal vibration of the stator core, vertical vibration of the stator core, X-direction runout of the upper guide bearing, Y-direction runout of the upper guide bearing, X-direction runout of the lower guide bearing, Y-direction runout of the lower guide bearing, X-direction runout of the water guide bearing, and Y-direction runout of the water guide bearing.

[0039] Furthermore, in S8, the model is incrementally updated periodically on the historical data set, irrelevant datasets are deleted, and the adaptability of real-time prediction of unit status is improved.

[0040] This invention proposes a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning. Using online monitoring data of industrial equipment, this invention employs topological multi-task learning to detect the operating status of hydropower units and to quickly and accurately predict and analyze their real-time healthy operation and abnormal faults. By incorporating topological loss and topological alignment constraints into the multi-task learning process, this invention enables the model to better utilize common features among tasks, thereby improving prediction performance. Furthermore, a multi-stage modeling method is proposed to classify the unit's state according to different processes, achieving more accurate predictions. This provides a better solution for the field of multi-objective, multi-stage real-time prediction problems.

[0041] Example This invention proposes a real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning, the process of which is as follows: Figure 1 As shown, the control flow diagram is as follows: Figure 2 As shown, the specific implementation method includes the following steps: S1. The real-time status data acquired during the unit's operation mainly includes: horizontal vibration of the upper frame, vertical vibration of the upper frame, horizontal vibration of the lower frame, vertical vibration of the lower frame, horizontal vibration of the stator base, vertical vibration of the stator base, horizontal vibration of the stator core, vertical vibration of the stator core, X-direction runout of the upper guide bearing, Y-direction runout of the upper guide bearing, X-direction runout of the lower guide bearing, Y-direction runout of the lower guide bearing, X-direction runout of the water guide bearing, and Y-direction runout of the water guide bearing. This data is used as the output of the model. S2. The real-time operating data obtained during the unit operation process mainly includes: pressure before the nozzle, nozzle opening, deflector opening, water distribution ring pipe inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil tank oil temperature, guide bearing oil level, unit speed, and the state data of the unit during the previous moment as input to the model. S3. Based on the different basic information of the unit, including: rated speed, rated output power, inertia constant, rated voltage, thrust load, rated excitation current, number of guide bearings, rated head, main shaft diameter, and impeller diameter, the reference data of the unit is divided and a reference data set belonging to each unit is established. S4. From 0% to 25% of the rated speed is the low-speed zone. During this period, the starting shock and transient phenomena may cause initial vibration. From 25% to 75% of the rated speed is the medium-speed zone. As the speed increases, hydrodynamic and electromagnetic excitation gradually increase, and the unit may pass through the resonance zone, where vibration and runout reach a peak. From 75% to 100% of the rated speed is the high-speed zone. At this point, the unit's damping and control systems begin to function effectively, and vibration and runout gradually stabilize and decrease. When the unit reaches the rated speed, it enters the stable operating zone. The unit operates within the design-allowed operating conditions, and vibration and runout are controlled within a safe range, with the system entering a dynamic equilibrium state. Thus, the unit's state data is divided into three stages. S5. Establish a topology-based multi-task learning model. The structure diagram of the topology-based multi-task learning prediction method is shown below. Figure 3 As shown, training data for a real-time dynamic prediction model of hydropower unit status is constructed using historical data sets. The nonlinear estimation function is transformed into a linear estimation function in a high-dimensional feature space using the constructed optimal decision function.

[0042] in N The number of samples. x k Each sample contains information on the unit's operating status. y pre To predict the unit's status at the next moment. w T These are the regression coefficients of the real-time dynamic prediction model for unit status. b This refers to the deviation of the real-time dynamic prediction model for unit status. The radial basis function kernel uses kernel tricks to map low-dimensional data to a high-dimensional space, solving nonlinear problems. The objective function is written as:

[0043] in and These are hyperparameters used to measure the strength of each computational during task training, and the overall task loss. L The loss consists of three parts. The task loss, which is the sum of the root mean square errors of each task, can be written as:

[0044] T The number of tasks. , and The first i The true and predicted values ​​output by the task model Indicates the first iThe output uncertainty of each task is used to balance the dynamic loss weights that account for the uncertainty of each task in multi-task learning. L topo For topological loss, the topological differences between tasks are calculated through persistent cohomology, specifically by calculating the Wasserstein distance between tasks, ensuring that the shared topological structure of different tasks is similar.

[0045] in N The number of data points in each task. , It is a task i Sorted predicted values It is a task j The sorted predicted values, this distance measures the difference in the distribution of the outputs of the two tasks. The smaller the distance, the more similar the two tasks are. As a regularization term, the squared norm of the weight parameters is added to the total loss. This utilizes the principle of structural risk minimization to make the parameter values ​​smaller, reducing the risk of overfitting during training. Regarding the loss function, the root mean square error itself is a convex function. The second reciprocal of is non-negative everywhere except zero. A linear combination of convex functions remains convex. When the regularization term's second derivative is always positive, it is strictly convex, therefore the loss function... L Since the function is convex, its minimum value can be found using optimization principles. The stochastic gradient descent method can be used to find the value that minimizes the loss function. w and b The stochastic gradient descent strategy is as follows:

[0046] in a It is a constant. and The loss function is respectively in w and Gradient of direction; S6. Real-time dynamic prediction of unit status is performed using a topology-based multi-task learning-based hydropower unit status prediction model. The formula is as follows:

[0047] in y t+1The output for the next moment includes: data on the horizontal vibration of the upper frame, the vertical vibration of the upper frame, the horizontal vibration of the lower frame, the vertical vibration of the lower frame, the horizontal vibration of the stator base, the vertical vibration of the stator base, the horizontal vibration of the stator core, the vertical vibration of the stator core, the X-direction runout of the upper guide bearing, the Y-direction runout of the upper guide bearing, the X-direction runout of the lower guide bearing, the Y-direction runout of the lower guide bearing, the X-direction runout of the water guide bearing, and the Y-direction runout of the water guide bearing. MTL t+1 According to t The learning curve fitted by topological multi-task learning at time +1. x t+1 for t The inputs at +1 include: nozzle inlet pressure, nozzle opening, deflector opening, water distribution ring inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil sump temperature, guide bearing oil level, and unit speed. y t The output at time t includes: upper frame horizontal vibration, upper frame vertical vibration, lower frame horizontal vibration, lower frame vertical vibration, stator base horizontal vibration, stator base vertical vibration, stator core horizontal vibration, stator core vertical vibration, upper guide bearing X-direction runout, upper guide bearing Y-direction runout, lower guide bearing X-direction runout, lower guide bearing Y-direction runout, water guide bearing X-direction runout, and water guide bearing Y-direction runout data; S7. Select unit data that is similar to the current unit operating status from the current unit reference data set, compare each predicted value with the formation reference data set, if the error exceeds the allowable value, then perform error correction on the real-time dynamic prediction result of the unit status, if the error is within the allowable range, then use it as the prediction input for the next moment. S8. Incrementally update the historical data set and delete datasets that are not relevant to the updated real-time dynamic prediction model for aircraft loading to enhance the model's adaptability.

[0048] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the real-time dynamic prediction method for the state of a hydropower unit based on topology multi-task learning.

[0049] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the real-time dynamic prediction method for the state of a hydropower unit based on topology multi-task learning.

[0050] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0051] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0052] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0053] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0054] The above provides a detailed description of the real-time dynamic prediction method for the state of hydropower units based on topological multi-task learning proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for real-time dynamic prediction of the state of a hydroelectric generating unit based on topological multi-task learning, characterized by, The method includes: S1. Obtain real-time status data during unit operation as the output of the model; S2. Obtain real-time operating data during unit operation and status data during the previous unit operation as input to the model; S3. Divide the reference data of the generating units according to their different basic information and establish a reference data set belonging to each generating unit. S4. Divide the unit operation process into several stages based on the speed data; S5. Establish a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning. S6. Real-time dynamic prediction of unit status is performed using a hydropower unit status real-time dynamic prediction model based on topology multi-task learning. S7. Select data from units with similar operating conditions to compare with the predicted values. If the error exceeds the limit, correct it. If it is within the range, use it as the prediction input for the next moment to improve the prediction accuracy. S8. Regularly update the historical data set.

2. The method of claim 1, wherein, The real-time status data during the operation of the S1 unit includes: horizontal vibration of the upper frame, vertical vibration of the upper frame, horizontal vibration of the lower frame, vertical vibration of the lower frame, horizontal vibration of the stator base, vertical vibration of the stator base, horizontal vibration of the stator core, vertical vibration of the stator core, X-direction runout of the upper guide bearing, Y-direction runout of the upper guide bearing, X-direction runout of the lower guide bearing, Y-direction runout of the lower guide bearing, X-direction runout of the water guide bearing, and Y-direction runout of the water guide bearing.

3. The method of claim 1, wherein, The real-time operating data of the S2 unit during operation includes: nozzle pressure, nozzle opening, deflector opening, water distribution ring inlet pressure, casing vacuum pressure, technical water supply outlet pressure, technical water supply inlet flow rate, guide bearing bearing temperature, guide bearing oil tank temperature, guide bearing oil level, and unit speed.

4. The method according to claim 1, characterized in that, The basic information of the unit in S3 includes: rated speed, rated output power, inertia constant, rated voltage, thrust load, rated excitation current, number of guide bearings, rated head, main shaft diameter, and impeller diameter.

5. The method according to claim 1, characterized in that, The method for dividing the speed data during unit operation in S4 is as follows: 0% to 25% of the rated speed is the low-speed zone, which causes initial vibration; 25% to 75% of the rated speed is the medium-speed zone, where the speed increases and the unit will pass through the resonance zone; 75% to 100% of the rated speed is the high-speed zone, where the unit's damping and control system begin to play a full role and the system enters a dynamic equilibrium state.

6. The method according to claim 1, characterized in that, The S5 establishes a real-time dynamic prediction model for the state of hydropower units based on topological multi-task learning, specifically as follows: Training data for a real-time dynamic prediction model of hydropower unit status is constructed using historical datasets. The nonlinear estimation function is then transformed into a linear estimation function in a high-dimensional feature space using the constructed optimal decision function. in N The number of samples. x k Each sample contains unit operating status information. y pre To predict the unit's status at the next moment. w T These are the regression coefficients of the real-time dynamic prediction model for unit status. b The deviation of the real-time dynamic prediction model for unit status; For radial basis function kernels, kernel tricks are used to map low-dimensional data to high-dimensional space to solve nonlinear problems; The objective function is written as: in and As a hyperparameter, the overall task loss L The loss consists of three parts. The task loss, which is the sum of the root mean square errors of each task, can be written as: T The number of tasks. , and The first i The true and predicted values ​​output by the task model Indicates the first i The output of each task is uncertain; L topo For topological loss, the topological differences between tasks are calculated through persistent cohomology, specifically by calculating the Wasserstein distance between tasks to ensure that the shared topological structure of different tasks is similar. in N The number of data points in each task. , It is a task i Sorted predicted values It is a task j The sorted predicted values, this distance measures the difference in the distribution of the outputs of the two tasks; As a regularization term, the squared norm of the weight parameters is added to the total loss. This utilizes the principle of structural risk minimization to make the parameter values ​​smaller, reducing the risk of overfitting during training. The stochastic gradient descent method is then used to find the solution that minimizes the loss function. w and b The stochastic gradient descent strategy is as follows: in a It is a constant. and The loss function is respectively in w and Gradient of direction.

7. The method according to claim 1, characterized in that, The specific formula for real-time dynamic prediction of unit status in S6 is as follows: in y t+1 For the output of the next time step, MTL t+1 According to t The learning curve fitted by topological multi-task learning at time +1. x t for t Input at any time y t for t Output at all times.

8. The method according to claim 1, characterized in that, The S8 method periodically performs incremental updates on the historical data set, deletes irrelevant datasets, and improves the adaptability of real-time prediction of unit status.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-8.

Images