A method and apparatus for predicting the performance of a water pump and turbine

Abstract

This invention relates to the field of pump-turbine performance prediction technology, and discloses a method and apparatus for predicting pump-turbine performance. The method includes acquiring historical geometric parameters, historical operating parameters, and historical performance parameters corresponding to a target pump-turbine; calculating dimensionless parameters of the pump-turbine based on the historical geometric parameters and historical operating parameters; constructing a pump-turbine performance prediction model; training the pump-turbine performance prediction model based on the historical geometric parameters, historical operating parameters, dimensionless parameters, and historical performance parameters; obtaining a trained pump-turbine performance prediction model; and using the trained model to predict the performance of the target pump-turbine to obtain the predicted performance result. This invention shortens the performance prediction cycle and provides a data foundation for the fast-paced, high-frequency optimization design of pump-turbines.

Description

Technical Field

[0001] This invention relates to the field of water pump and turbine performance prediction technology, specifically to a method and apparatus for predicting the performance of water pumps and turbines. Background Technology

[0002] As the core equipment of pumped storage power stations, pump-turbines are key hubs for realizing bidirectional energy conversion and efficient storage in power systems. Their efficient energy conversion capabilities provide reliable support for large-scale energy storage and are an important foundation for improving the flexibility, security and economy of power systems.

[0003] However, the prediction cycle of relevant pump-turbine performance prediction methods is relatively long, which cannot adapt to the fast-paced and high-frequency optimization design rhythm of pump-turbines. Summary of the Invention

[0004] This invention provides a method and apparatus for predicting the performance of a water pump turbine, in order to solve the problem that the prediction cycle of related water pump turbine performance prediction methods is too long and cannot adapt to the fast-paced, high-frequency optimization design rhythm of water pump turbines.

[0005] In a first aspect, the present invention provides a method for predicting the performance of a water pump turbine, the method comprising:

[0006] Obtain the historical geometric parameters, historical operating parameters, and historical performance parameters corresponding to the target water pump turbine;

[0007] Dimensionless parameters of the pump-turbine are calculated based on historical geometric parameters and historical operating parameters. Among them, the dimensionless parameters of the pump-turbine include the speed ratio under power generation conditions, the speed ratio under pumping conditions, the cavitation coefficient, and the symmetry factor of the bidirectional flow channel.

[0008] A performance prediction model for a water pump turbine is constructed; the performance prediction model for a water pump turbine includes a bidirectional long short-term memory neural network and a physical information neural network;

[0009] Based on historical geometric parameters, historical operating parameters, dimensionless parameters of pump-turbine, and historical performance parameters, the pump-turbine performance prediction model is trained to obtain the trained pump-turbine performance prediction model.

[0010] The performance prediction of the target pump-turbine is predicted using the trained pump-turbine performance prediction model, and the prediction results are obtained.

[0011] The pump-turbine performance prediction method provided in this embodiment calculates the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters, and constructs a pump-turbine performance prediction model. This model includes a bidirectional long short-term memory neural network and a physical information neural network. The model is trained based on historical geometric parameters, historical operating parameters, the dimensionless parameters of the pump-turbine, and historical performance parameters to enhance its predictive ability. The trained model is then used to predict the performance of the target pump-turbine, shortening the prediction cycle and providing a data foundation for fast-paced, high-frequency pump-turbine optimization design.

[0012] In one optional implementation, the dimensionless parameters of the pump-turbine are calculated based on historical geometric parameters and historical operating parameters, including:

[0013] The power generation speed ratio is calculated based on historical operating parameters, including power generation head, power generation flow rate, and power generation speed.

[0014] The pumping speed ratio is calculated based on the pumping head, pumping flow rate and pumping speed in historical operating parameters.

[0015] The cavitation coefficient is calculated based on historical geometric parameters, including pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density, and runner outlet velocity.

[0016] The symmetry factor of the bidirectional flow channel is calculated based on the power generation head and pumping head.

[0017] The pump-turbine performance prediction method provided in this embodiment calculates the power generation operating speed ratio based on historical operating parameters such as generator head, generator flow rate, and generator speed. This allows for rapid assessment of the matching degree between the actual speed and the optimal operating speed during power generation, ensuring high efficiency in power generation. The method also calculates the pumping operating speed ratio based on historical operating parameters such as pumping head, pumping flow rate, and pumping speed, accurately evaluating the compatibility between the pumping speed and operating parameters, avoiding low pumping efficiency or equipment damage due to speed imbalance. Furthermore, the method calculates the cavitation coefficient based on historical geometric parameters such as pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density, and runner outlet velocity, providing early warning of runner cavitation risks, preventing cavitation erosion and equipment damage, and extending runner service life. Finally, the method calculates the bidirectional flow channel symmetry factor based on generator head and pumping head, quantifying the structural symmetry of the bidirectional flow channel, providing a basis for flow channel optimization, and improving the stability and efficiency consistency of the pump-turbine's bidirectional operation.

[0018] In one optional implementation, a pump-turbine performance prediction model is trained based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters to obtain a trained pump-turbine performance prediction model, including:

[0019] Historical geometric parameters, historical operating parameters, dimensionless parameters of water pumps and turbines, and historical performance parameters are input into a bidirectional long short-term memory neural network to obtain intermediate feature vectors.

[0020] The physical field parameters of the water pump and turbine are obtained, and the physical field parameters and intermediate feature vectors of the water pump and turbine are input into the physical information neural network to obtain the performance prediction parameters.

[0021] Obtain the component spatial coordinates corresponding to the target water pump turbine, and perform a physical field distribution simulation of the target water pump turbine based on the component spatial coordinates to obtain the physical field distribution value of the water pump turbine;

[0022] The adaptive weighted loss value is determined based on performance prediction parameters, component spatial coordinates, physical field distribution values ​​of water pumps and turbines, and dimensionless parameters of water pumps and turbines.

[0023] The model parameters are iteratively optimized using adaptive weighted loss values ​​to obtain the trained pump-turbine performance prediction model.

[0024] The pump-turbine performance prediction method provided in this embodiment inputs historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters into a bidirectional long short-term memory neural network to obtain intermediate feature vectors. This effectively extracts temporal correlations and multi-dimensional key features from historical data, providing a comprehensive data foundation for performance prediction. The method also inputs the pump-turbine physical field parameters and intermediate feature vectors into a physical information neural network to obtain performance prediction parameters. This method integrates data-driven features and physical constraints, ensuring the physical rationality and preliminary accuracy of performance prediction. Based on component spatial coordinates, it performs... The physical field distribution simulation yields the physical field distribution values ​​of the pump-turbine, providing detailed physical field data support for subsequent loss calculations and performance analysis. Based on performance prediction parameters, component spatial coordinates, pump-turbine physical field distribution values, and dimensionless parameters of the pump-turbine, an adaptive weighted loss value is determined. This value dynamically adjusts the loss weights according to prediction deviations and physical field characteristics, enhancing the sensitivity of the loss function to key errors. Finally, the adaptive weighted loss value is used to iteratively optimize the model parameters. Through dynamic loss constraints, the parameters are continuously optimized, significantly improving the model's prediction accuracy and generalization ability for pump-turbine performance.

[0025] In one optional implementation, historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters are input into a bidirectional long short-term memory neural network to obtain an intermediate feature vector, including:

[0026] The historical geometric parameters, historical operating parameters, dimensionless parameters of water pumps and turbines, and historical performance parameters are structured to obtain the structured time series matrix.

[0027] The structured time series matrix is ​​normalized and residual noise is reduced to obtain a normalized input vector.

[0028] Time-series weights are generated based on historical operating parameters and dimensionless parameters of pumps and turbines.

[0029] The normalized temporal feature vector is encoded using a two-layer encoding process to obtain the encoded temporal feature vector.

[0030] Based on the encoded temporal feature vector and temporal weights, attention pooling and projection processing are performed to obtain intermediate feature vectors.

[0031] The pump-turbine performance prediction method provided in this embodiment obtains a structured time-series matrix by structuring historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters. This structured matrix unifies the format and dimensions of different parameters, providing a standardized data foundation for time-series feature extraction. Normalization and residual denoising are then applied to the structured time-series matrix to obtain a normalized input vector, eliminating differences in data dimensions and filtering out noise interference, thus improving the stability and consistency of the input data. Based on historical operating parameters and dimensionless parameters of the pump-turbine, time-series weights are generated, quantifying the differences in the contribution of parameters at different times to performance prediction and enhancing the weight ratio of key time-series information. The normalized time-series feature vector is then subjected to double-layer encoding to obtain an encoded time-series feature vector. This multi-layer feature transformation allows for in-depth mining of complex correlations and hidden patterns in the time-series data. Based on the encoded time-series feature vector and time-series weights, attention pooling and projection processing are performed to focus on key features and compress dimensions, outputting more representative and concise features, thus improving the processing efficiency and accuracy of subsequent models.

[0032] In one optional implementation, an adaptive weighted loss value is determined based on performance prediction parameters, component spatial coordinates, pump-turbine physical field distribution values, and pump-turbine dimensionless parameters, including:

[0033] Based on the physical field distribution values ​​of the water pump and turbine and the spatial coordinates of the components, the residuals of the continuity equation, momentum equation and boundary conditions are calculated respectively.

[0034] Determining the loss weighting coefficient based on dimensionless parameters of water pumps and turbines;

[0035] The adaptive weighted loss value is calculated based on the loss weighting coefficient, performance prediction parameters, historical performance parameters, continuity equation residuals, momentum equation residuals, and boundary condition residuals.

[0036] The pump-turbine performance prediction method provided in this embodiment calculates the residuals of the continuity equation, momentum equation, and boundary conditions based on the physical field distribution values ​​and component spatial coordinates of the pump-turbine. This quantifies the degree to which the physical equations and boundary conditions are satisfied, providing the model with hard constraint indicators that conform to the laws of fluid mechanics. The loss weighting coefficients are determined based on the dimensionless parameters of the pump-turbine, enabling the weights of different loss terms to adapt to the physical characteristics and operating conditions of the pump-turbine, thus improving the sensitivity of the adaptive weighted loss value to key errors. Finally, based on the loss weighting coefficients, performance prediction parameters, historical performance parameters, the residuals of the continuity equation, momentum equation, and boundary conditions, an adaptive weighted loss value is calculated. This ensures that the adaptive weighted loss value accurately reflects the overall deviation of the model and effectively guides the optimization direction of model parameters.

[0037] In one alternative implementation, it further includes:

[0038] The confidence level of the predicted performance of the pump-turbine is assessed, and the blade angle of the target pump-turbine is adjusted based on the confidence level assessment results.

[0039] The pump-turbine performance prediction method provided in this embodiment evaluates the confidence level of the pump-turbine performance prediction results and adjusts the blade angle based on the confidence level evaluation results. This ensures the reliability of the blade angle adjustment basis and accurately adapts to the target operating conditions, effectively optimizing the operating efficiency and stability of the pump-turbine.

[0040] In a second aspect, the present invention provides a pump-turbine performance prediction device, the device comprising:

[0041] The acquisition module is used to acquire the historical geometric parameters, historical operating parameters, and historical performance parameters of the target water pump turbine.

[0042] The calculation module is used to calculate the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters; among which, the dimensionless parameters of the pump-turbine include the power generation speed ratio, the pumping speed ratio, the cavitation coefficient, and the bidirectional flow channel symmetry factor;

[0043] The module is used to build a performance prediction model for water pumps and turbines; the performance prediction model for water pumps and turbines includes a bidirectional long short-term memory neural network and a physical information neural network.

[0044] The training module is used to train the pump-turbine performance prediction model based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters, so as to obtain the trained pump-turbine performance prediction model.

[0045] The prediction module is used to predict the performance of the target pump-turbine using the trained pump-turbine performance prediction model, and obtain the pump-turbine performance prediction results.

[0046] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the pump-turbine performance prediction method described in the first aspect or any corresponding embodiment thereof.

[0047] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the pump-turbine performance prediction method described in the first aspect or any corresponding embodiment thereof.

[0048] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the pump-turbine performance prediction method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0049] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of the first process of a method for predicting the performance of a water pump turbine according to an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of the second process of a method for predicting the performance of a water pump turbine according to an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of the third process of a pump-turbine performance prediction method according to an embodiment of the present invention;

[0054] Figure 5This is a schematic diagram of the training process of the pump-turbine performance prediction model according to an embodiment of the present invention;

[0055] Figure 6 This is a structural block diagram of a water pump turbine performance prediction device according to an embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0058] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0059] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0060] As an optional application scenario of this invention, such as Figure 1 As shown, the water pump turbine performance prediction device may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0061] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0062] Relevant methods for predicting the performance of pump-turbines include: 1) Model testing of pump-turbines, which requires building a model test bench and measuring efficiency, cavitation critical point, and pressure pulsation through a scaled-down model. However, the construction cost of the test bench exceeds ten million yuan, and the single test cycle can last for 2-4 months; 2) Predicting the performance of pump-turbines using CFD (Computational Fluid Dynamics) simulation, i.e., constructing a computational fluid dynamics simulation model to predict performance curves. However, the calculation error under biased operating conditions is too large, and the results cannot be used as a criterion for evaluating hydraulic design schemes. At the same time, the calculation cycle for pressure pulsation is too long, which cannot adapt to the current fast-paced and high-frequency hydraulic optimization design rhythm; 3) Realizing condition monitoring, fault diagnosis, and operation strategy formulation through data-driven approaches. There are currently no relevant performance prediction methods or system proposals, which cannot provide direct assistance for the hydraulic research and development of units.

[0063] According to an embodiment of the present invention, a method for predicting the performance of a water pump turbine is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0064] This embodiment provides a method for predicting the performance of a water pump turbine, which can be used in the aforementioned electronic equipment. Figure 2 This is a flowchart of a pump-turbine performance prediction method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0065] Step S201: Obtain the historical geometric parameters, historical operating parameters, and historical performance parameters corresponding to the target water pump turbine.

[0066] Specifically, historical geometric parameters, historical operating parameters, and historical performance parameters are obtained from CFD simulation data and real experimental data; among them, historical geometric parameters are static geometric quantities (time-invariant) including the number of impeller blades (e.g., 9 blades) and the angle of the movable guide vanes. α Rotor blade angle β and the diameter of the wheel D 1 / D In a 2:1 ratio, historical geometric parameters serve as static priors and conditional quantities in the pump-turbine performance prediction model, while historical operating parameters are dynamic time-series quantities, including power generation mode parameters (power generation head). Power generation flow and power generation speed Pumping mode parameters (pumping head) Pumping flow rate and pumping speed The input to the pump-turbine performance prediction model consists of a combination of a 0-1 code (0 for power generation and 1 for pumping) and a working condition identifier (0-1 encoding, 0 for power generation and 1 for pumping). After wavelet transform denoising and removal of transitional working conditions, the input is used to define the time series input. Historical performance parameters serve as the monitoring targets for the model, including efficiency. η (Power generation efficiency) η_gen With pumping efficiency η_pump ), cavitation coefficient and pressure pulsation amplitude Ap It is used as the supervisory output of the pump-turbine performance prediction model and is not directly used as input.

[0067] Step S202: Calculate the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters; wherein, the dimensionless parameters of the pump-turbine include the power generation speed ratio, the pumping speed ratio, the cavitation coefficient, and the bidirectional flow channel symmetry factor.

[0068] Specifically, historical geometric parameters and historical operating parameters are converted into dimensionless parameters of the pump-turbine, such as the flow coefficient. φ Head coefficient ψ Speed ​​ratio ns (Speed ​​ratio for power generation and speed ratio for pumping), cavitation coefficient, and bidirectional flow channel symmetry factor.

[0069] Step S203: Construct a performance prediction model for the water pump and turbine; wherein, the performance prediction model for the water pump and turbine includes a bidirectional long short-term memory neural network and a physical information neural network.

[0070] Specifically, the pump-turbine performance prediction model adopts a hybrid structure of bidirectional long short-term memory (Bi-LSTM) neural network and physical information neural network (PINN). Bi-LSTM is used to learn time-series parameters (such as pressure fluctuations over time), and simplified NS equations (continuity equation and momentum equation) are embedded in PINN to constrain the performance. By integrating historical experimental data and CFD data into a physical constraint machine learning model (i.e., the pump-turbine performance prediction model), the performance prediction problem in the pump-turbine development stage is solved.

[0071] Step S204: Based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters, train the pump-turbine performance prediction model to obtain the trained pump-turbine performance prediction model.

[0072] Specifically, during the pre-training process of the pump-turbine performance prediction model, 100,000 sets of CFD virtual data (covering the S-shaped curve region) are used to learn the basic flow laws, and 95% of real experimental data are superimposed. Transfer learning is used to optimize the sensitive parameters of the model in order to achieve fine-tuning of the parameters of the pump-turbine performance prediction model.

[0073] Furthermore, the power generation and pumping modes use the same pump-turbine performance prediction model, but during the training phase, two sets of model parameter groups are constructed to ensure the prediction accuracy of the model under two typical flow directions. The training set includes 0 / 1 mode identifiers, and some shared parameters are used to enhance bidirectional generalization ability. The non-shared part is used to refine and optimize parameters for different flow characteristics. The bidirectional flow channel symmetry factor is defined by min / max (minimum / maximum) normalization, and samples with bidirectional flow channel symmetry factors not equal to 1 are given higher weights during training to enhance robustness to bidirectional asymmetric models.

[0074] Step S205: Use the trained pump-turbine performance prediction model to predict the performance of the target pump-turbine and obtain the pump-turbine performance prediction result.

[0075] Specifically, the performance prediction results for the pump-turbine include: η_gen or η_pump Surface plot, critical boundary line of cavitation coefficient (determined based on cavitation coefficient), pressure pulsation dominant frequency and dangerous resonance warning.

[0076] The pump-turbine performance prediction method provided in this embodiment calculates the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters, and constructs a pump-turbine performance prediction model. This model includes a bidirectional long short-term memory neural network and a physical information neural network. The model is trained based on historical geometric parameters, historical operating parameters, the dimensionless parameters of the pump-turbine, and historical performance parameters to enhance its predictive ability for pump-turbine performance. The trained model is then used to predict the performance of a target pump-turbine, resulting in a shorter prediction cycle and providing a data foundation for fast-paced, high-frequency pump-turbine optimization design.

[0077] This embodiment provides a method for predicting the performance of a water pump turbine, which can be used in the aforementioned electronic equipment. Figure 3 This is a flowchart of a pump-turbine performance prediction method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0078] Step S301: Obtain the historical geometric parameters, historical operating parameters, and historical performance parameters corresponding to the target pump-turbine. For details, please refer to [link to relevant documentation]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0079] Step S302: Calculate the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters; wherein, the dimensionless parameters of the pump-turbine include the power generation speed ratio, the pumping speed ratio, the cavitation coefficient, and the bidirectional flow channel symmetry factor.

[0080] Specifically, step S302 includes:

[0081] Step S3021: Calculate the power generation operating speed ratio based on the power generation head, power generation flow rate, and power generation speed in the historical operating parameters.

[0082] Specifically, the speed ratio under power generation conditions The calculation formula is:

[0083] (1)

[0084] in, For the generator rotation speed, For power generation flow, For generating water head.

[0085] Step S3022: Calculate the pumping speed ratio based on the pumping head, pumping flow rate and pumping speed in the historical operating parameters.

[0086] Specifically, the pumping speed ratio The calculation formula is:

[0087] (2)

[0088] in, For pumping head, The pumping speed is [missing information]. This represents the pumping flow rate.

[0089] Step S3023: Calculate the cavitation coefficient based on the pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density, and runner outlet velocity from the historical geometric parameters.

[0090] Specifically, cavitation coefficient The calculation formula is:

[0091] (3)

[0092] in, The outlet pressure of the water pump turbine. The inlet pressure of the water pump and turbine. For the density of the liquid, This is the turbine exit speed.

[0093] Step S3024: Calculate the bidirectional flow channel symmetry factor based on the power generation head and pumping head.

[0094] Specifically, the symmetry factor of the bidirectional flow channel The calculation formula is:

[0095] (4)

[0096] in, and These are the rated flow rates of the turbine and the pump, respectively. The closer it is to 1, the more balanced the bidirectional flow capacity.

[0097] Furthermore, the dimensionless parameters of the pump turbine are used both as auxiliary factors in the Bi-LSTM for time-series encoding and as conditional parameters in PINN for physical residual scaling and loss weighting. Among them, the bidirectional flow channel symmetry factor is defined by maximum / minimum normalization, and the samples of the bidirectional flow channel symmetry factor are given higher weights during training to enhance the robustness to bidirectional asymmetric models.

[0098] Step S303: Construct a performance prediction model for the water pump and turbine; the performance prediction model for the water pump and turbine includes a bidirectional long short-term memory neural network and a physical information neural network. For details, please refer to... Figure 2 Step S203 of the illustrated embodiment will not be described again here.

[0099] Step S304: Based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters, the pump-turbine performance prediction model is trained to obtain the trained pump-turbine performance prediction model. For details, please refer to [link to details]. Figure 2 Step S204 of the illustrated embodiment will not be described again here.

[0100] Step S305: The performance of the target pump-turbine is predicted using the trained pump-turbine performance prediction model to obtain the pump-turbine performance prediction results. For details, please refer to [link to relevant documentation]. Figure 2 Step S205 of the illustrated embodiment will not be described again here.

[0101] The pump-turbine performance prediction method provided in this embodiment calculates the power generation operating speed ratio based on historical operating parameters such as generator head, generator flow rate, and generator speed. This allows for rapid assessment of the matching degree between the actual speed and the optimal operating speed during power generation, ensuring high efficiency in power generation. The method also calculates the pumping operating speed ratio based on historical operating parameters such as pumping head, pumping flow rate, and pumping speed, accurately evaluating the compatibility between the pumping speed and operating parameters, avoiding low pumping efficiency or equipment damage due to speed imbalance. Furthermore, the method calculates the cavitation coefficient based on historical geometric parameters such as pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density, and runner outlet velocity, providing early warning of runner cavitation risks, preventing cavitation erosion and equipment damage, and extending runner service life. Finally, the method calculates the bidirectional flow channel symmetry factor based on generator head and pumping head, quantifying the structural symmetry of the bidirectional flow channel, providing a basis for flow channel optimization, and improving the stability and efficiency consistency of the pump-turbine's bidirectional operation.

[0102] This embodiment provides a method for predicting the performance of a water pump turbine, which can be used in the aforementioned electronic equipment. Figure 4 This is a flowchart of a pump-turbine performance prediction method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps:

[0103] Step S401: Obtain the historical geometric parameters, historical operating parameters, and historical performance parameters corresponding to the target pump-turbine. For details, please refer to [link to relevant documentation]. Figure 3 Step S301 of the illustrated embodiment will not be described again here.

[0104] Step S402: Calculate the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters; wherein, the dimensionless parameters of the pump-turbine include the speed ratio under power generation conditions, the speed ratio under pumping conditions, the cavitation coefficient, and the bidirectional flow channel symmetry factor. For details, please refer to [link to relevant documentation]. Figure 3 Step S302 of the illustrated embodiment will not be described again here.

[0105] Step S403: Construct a performance prediction model for the water pump-turbine; this model includes a bidirectional long short-term memory neural network and a physical information neural network. For details, please refer to [link to details]. Figure 3 Step S303 of the illustrated embodiment will not be described again here.

[0106] Step S404: Based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters, train the pump-turbine performance prediction model to obtain the trained pump-turbine performance prediction model.

[0107] Specifically, step S404 includes:

[0108] Step S4041: Input the historical geometric parameters, historical operating parameters, dimensionless parameters of the pump and turbine, and historical performance parameters into the bidirectional long short-term memory neural network to obtain the intermediate feature vector.

[0109] Specifically, a dual-input path structure is adopted to input historical geometric parameters, historical operating parameters, dimensionless parameters of the pump and turbine, and historical performance parameters into a bidirectional long short-term memory neural network. One input is historical performance parameters, and the other input is a time series composed of historical geometric parameters, historical operating parameters, and dimensionless parameters of the pump and turbine (such as local Reynolds number, cavitation coefficient, and Reynolds stress ratio). The two inputs are encoded separately and then fused to form a time-physical dual-modal representation, which improves the ability to model the time series of complex flows. Furthermore, the 0 (power generation) / 1 (pumping) mode is encoded as a trainable embedding vector and participates in the Bi-LSTM input, which explicitly guides the model to distinguish the differences between the two-way operating flow modes.

[0110] Furthermore, after preprocessing the structured temporal matrix within the window using time-distributed normalization and residual denoising, cross-scale unsteady features are extracted sequentially through two layers of Bi-LSTM. Attention weights are generated using historical operating parameters and dimensionless parameters of the pump and turbine as conditions, and weighted pooling is performed on the encoded temporal feature vector to obtain an intermediate feature vector. The intermediate feature vector carries a fusion representation of temporal-geometric-dimensional features, which is then input into the PINN branch.

[0111] In some optional implementations, step S4041 above includes:

[0112] Step a1 involves structuring the historical geometric parameters, historical operating parameters, dimensionless parameters of the pump and turbine, and historical performance parameters to obtain the structured time series matrix.

[0113] Specifically, historical geometric parameters serve as static input features, historical operating parameters serve as temporal input features, and historical performance parameters serve as temporal inputs to Bi-LSTM to help learn the dynamic evolution of performance with operating conditions. They also serve as supervisory outputs of PINN, used to compare with prediction results and calculate supervisory loss. Dimensionless parameters of the pump-turbine system participate in Bi-LSTM temporal modeling as input features and are used in the PINN branch for scaling physical residuals and weighting loss, ensuring physical consistency across operating conditions. Through these mapping relationships, the structured pump-turbine data is divided into four categories: static geometric quantities, dynamic temporal quantities, performance target quantities, and dimensionless physical quantities. These categories respectively serve as feature inputs, temporal modeling, supervisory constraints, and physical consistency calibration in Bi-LSTM and PINN. The structured temporal matrix is ​​primarily composed of dynamic temporal quantities (i.e., historical operating parameters), and in the time dimension, it is spliced ​​together in a broadcast manner with historical geometric parameters, historical operating parameters, dimensionless pump-turbine parameters, and pattern embedding vectors (power generation / pumping condition identifiers) to prepare for subsequent Bi-LSTM processing.

[0114] Furthermore, the structured processing includes not only removing transitional conditions and wavelet denoising, but also: time alignment and uniform sampling interval, sliding time window slicing, normalization and dimensionless transformation by mode / model, time broadcasting of static geometry to temporal features, and hierarchical sampling to construct training / validation / reservation sets. The structured temporal matrix is ​​input into Bi-LSTM to achieve the splicing of static geometry and mode embedding in the time dimension. Finally, the intermediate feature vector is obtained after processing by Bi-LSTM. The intermediate feature vector carries a fusion representation of temporal-geometric-dimensionality, which is then input into PINN.

[0115] Step a2 involves normalizing and denoising the structured time series matrix to obtain a normalized input vector.

[0116] Specifically, normalized input vector The expression is:

[0117] (5)

[0118] in, This is a layer normalization operator used for feature scale standardization. This is the structured time series matrix. This is the residual signal, and the perturbation information retained after filtering.

[0119] Step a3: Generate time-series weights based on historical operating parameters and dimensionless parameters of the pump and turbine.

[0120] Specifically, a lightweight attention mechanism module is embedded before the Bi-LSTM structure to enhance the model's response to nodes with abrupt changes in operating conditions (such as mode switching points and low-load instability periods) in the input sequence, thereby enhancing the Bi-LSTM's ability to identify key disturbance points in the time series; whereby the attention weights are driven by the rate of change of operating condition indicators (power generation / pumping efficiency, cavitation coefficient, and the first derivative of flow rate).

[0121] Furthermore, temporal weights are used to reflect the importance of the current time step to the final extracted features; the formula for calculating temporal weights is as follows:

[0122] (6)

[0123] in, Representative moment The attention weights reflect the importance of that moment to the final feature. The score represents unnormalized attention. This represents the length of the timing window.

[0124] Step a4: Perform double-layer encoding on the normalized temporal feature vector to obtain the encoded temporal feature vector.

[0125] Specifically, the steps for performing two-layer encoding on the normalized temporal feature vector include: 1) encoding the normalized temporal feature vector to obtain the hidden state sequence output by the first layer Bi-LSTM; 2) encoding the hidden state sequence output by the first layer Bi-LSTM to obtain the encoded temporal feature vector; wherein, the expression for the hidden state sequence output by the first layer Bi-LSTM is... and encoding time-series feature vectors The expressions are as follows:

[0126] (7)

[0127] (8)

[0128] Step a5: Based on the encoded temporal feature vector and temporal weights, attention pooling and projection processing are performed to obtain intermediate feature vectors.

[0129] Specifically, based on the encoded temporal feature vector and temporal weights, attention pooling is performed to obtain a weighted pooled temporal feature vector. This weighted pooled temporal feature vector is then projected to obtain an intermediate feature vector. The weighted pooled temporal feature vector... and intermediate feature vectors The expressions are as follows:

[0130] (9)

[0131] (10)

[0132] in, For regularization operations, This is a projection operation.

[0133] Step S4042: Obtain the physical field parameters of the water pump and turbine, and input the physical field parameters of the water pump and turbine and the intermediate feature vector into the physical information neural network to obtain the performance prediction parameters.

[0134] Specifically, the physical field parameters of the water pump turbine include water flow velocity. u Gravity acceleration degree g and material parameters em, intermediate feature vectors Water flow velocity u Gravity acceleration degree g and material parameters em, The intermediate features are input into the physical information neural network, and loss constraints based on continuity and momentum conservation equations are embedded to obtain performance prediction parameters. These parameters include the pump-turbine output efficiency, cavitation coefficient, and pressure pulsation amplitude. Ap and characteristic frequencies f Performance prediction parameters The expressions for the loss constraints of the continuity and momentum conservation equations are as follows:

[0135] (11)

[0136] (12)

[0137] in, For speed, For fluid density, For the Laplace operator

[0138] Step S4043: Obtain the component spatial coordinates corresponding to the target water pump turbine. Based on the component spatial coordinates, perform a physical field distribution simulation on the target water pump turbine to obtain the physical field distribution value of the water pump turbine.

[0139] Specifically, the spatial coordinates of the components corresponding to the target water pump turbine are input into the physical agent head in the physical information neural network, and an approximate field of velocity and pressure is generated at a representative point, that is, the physical field distribution value of the water pump turbine.

[0140] Step S4044: Determine the adaptive weighted loss value based on performance prediction parameters, component spatial coordinates, water pump turbine physical field distribution values, and water pump turbine dimensionless parameters.

[0141] In some optional implementations, step S4044 above includes:

[0142] Step b1: Based on the physical field distribution values ​​of the water pump and turbine and the spatial coordinates of the components, calculate the residuals of the continuity equation, momentum equation, and boundary conditions, respectively.

[0143] Specifically, PINN includes a supervised prediction head and a physical proxy head. It calculates the continuity and momentum residuals at representative collocation points through automatic differentiation, and together with the boundary condition residuals, they constitute the physical loss. The loss weight is adaptively adjusted by historical performance parameters to take into account the physical consistency of different modes and load segments. During the inference phase, consistency correction is applied to the output to ensure that the predicted quantity meets the overall constraints of mass and momentum.

[0144] Furthermore, the quantization layer of "physical constraints" in the physical information neural network uses an automatic differentiation algorithm to calculate the residual between the "prediction result of the supervisory prediction head" and the "physical field distribution output by the physical proxy head" and the physical laws (continuity equation, momentum equation, boundary conditions, etc.), thus transforming "whether the physical laws are satisfied" into a calculable numerical index.

[0145] Furthermore, based on the physical field distribution values ​​of the pump-turbine and the spatial coordinates of the components, the steps for calculating the residuals of the continuity equation, momentum equation, and boundary conditions include: 1) calculating the derivatives of the physical field distribution values ​​of the pump-turbine with respect to the spatial coordinates of the components; 2) substituting the calculated derivatives into the calculation formulas for the residuals of the continuity equation, momentum equation, and boundary conditions to obtain the residuals of the continuity equation, momentum equation, and boundary conditions. Among these, the residual of the continuity equation... Momentum equation residuals and boundary condition residuals The expressions are as follows:

[0146] (13)

[0147] (14)

[0148] (15)

[0149] in, For fluid density, For radial velocity components, For the circumferential velocity component, For the axial velocity component, For radial coordinates in cylindrical coordinate system, This is a cylindrical coordinate system with circumferential coordinates. For rotation angle, 、 and These are the component's spatial coordinates.

[0150] Step b2: Determine the loss weighting coefficient based on the dimensionless parameters of the water pump and turbine.

[0151] Specifically, the dimensionless parameters of the pump turbine are used both as auxiliary factors in the Bi-LSTM for time-series encoding and as conditional parameters in PINN for physical residual scaling and loss weighting. The bidirectional flow channel symmetry factor is defined using maximum / minimum normalization, and samples with a bidirectional flow channel symmetry factor not equal to 1 are given higher weights during training to enhance robustness to bidirectional asymmetric models.

[0152] Furthermore, based on physical analysis (such as the decomposition of loss by the energy equation) or empirical rules, an appropriate model form (such as multivariate nonlinear regression, physical knowledge neural network, etc.) is selected, and the weighting coefficients of each loss term are used as the parameters to be determined in the loss weighting model, so that the loss weighting model can characterize the contribution weight of different losses through parameter combinations.

[0153] Furthermore, through experimental testing (measuring the efficiency of the pump and turbine under multiple operating conditions and back-calculating the loss values) or high-precision numerical simulation (such as CFD calculation of the loss distribution under different operating conditions), multiple sets of sample data are obtained. These samples are substituted into the loss weighting model, and optimization algorithms (such as least squares method and gradient descent method) are used to minimize the error between the model's predicted loss and the actual loss, thereby solving for the loss weighting coefficient corresponding to each loss item.

[0154] Step b3: Calculate the adaptive weighted loss value based on the loss weighting coefficient, performance prediction parameters, historical performance parameters, continuity equation residuals, momentum equation residuals, and boundary condition residuals.

[0155] Specifically, adaptive weighted loss value The calculation formula is:

[0156] (16)

[0157] in, For performance prediction parameters, These are historical performance parameters. , and The loss weighting coefficient is dynamically adjusted based on the dimensionless parameter and the flow channel symmetry factor.

[0158] Step S4045: Iteratively optimize the model parameters using adaptive weighted loss values ​​to obtain the trained pump-turbine performance prediction model.

[0159] Specifically, the parameters of the trained pump-turbine performance prediction model mainly consist of two parts: first, the parameters of the neural network itself, such as the connection weights and bias terms between neurons in each layer, which determine the mapping relationship from input (such as operating parameters and spatial coordinates) to output (such as performance indicators such as speed, pressure, and efficiency); and second, the parameters related to physical constraints or loss mechanisms, such as the viscosity coefficient in the momentum equation, the weighting coefficients of different loss terms, and the relative weight coefficients of the continuity equation and the momentum equation in the loss function.

[0160] Furthermore, such as Figure 5 As shown, the steps for iterative optimization of the pump-turbine performance prediction model include: 1) Inputting the validation set into the pump-turbine performance prediction model to obtain the predicted performance parameters; wherein, the validation set retains 5% of the actual test data to ensure coverage of the following key operating conditions: low head and low load area in power generation mode, and high head and low flow area in pumping mode; 2) Analyzing the predicted performance parameters, if the efficiency change is greater than 0.5% and the pressure pulsation amplitude error is greater than 15%, the iterative optimization mechanism is triggered; wherein, the formula for calculating the pressure pulsation amplitude error is:

[0161] (17)

[0162] in, This is for the pressure pulsation amplitude error. and These represent the pressure pulsation amplitude errors before and after the test.

[0163] 3) Initiate CFD simulation to supplement the training set data and supplement the simulation for abnormal operating conditions (at least 500 sets of new data); 4) Adjust the model structure of the pump-turbine performance prediction model, that is, adjust the number of hidden layer nodes of Bi-LSTM (increase from 128 to 256); 5) Train the structured model using the supplemented training set data, and analyze the structured model again using the validation set until the change is less than or equal to 0.5% and the pressure pulsation amplitude error is less than or equal to 15%, to obtain the trained pump-turbine performance prediction model. If the efficiency change is greater than 0.5% and the pressure pulsation amplitude error is greater than 15%, an anomaly report is generated, prompting a list of operating conditions that need to be manually checked. Through a closed-loop iterative system that verifies the pump-turbine performance prediction model in real time, the prediction accuracy is self-optimized, directly replacing the physical model test.

[0164] Step S405: Use the trained pump-turbine performance prediction model to predict the performance of the target pump-turbine and obtain the pump-turbine performance prediction result.

[0165] Specifically, the steps for predicting the performance of a target pump-turbine using the trained pump-turbine performance prediction model include: 1) The user inputs the design parameters (geometry + target operating condition) of the new pump-turbine, and determines whether it is a power generation or pumping mode based on the operating condition identifier; 2) The trained pump-turbine performance prediction model is called to perform feature inference; 3) The output performance results include the power generation or pumping efficiency surface plot, the critical boundary line of the cavitation coefficient, the main frequency of pressure pulsation, and the warning of dangerous resonance.

[0166] Furthermore, based on Dropout (a regularization method) sampling, a confidence assessment is performed on the pump-turbine performance prediction results, outputting the 95% confidence interval of the efficiency prediction value (e.g., ...). η =92.3%±0.2%), and combined with historical data model matching degree, automatically recommends geometric parameter optimization suggestions (such as blade angle adjustment range).

[0167] Furthermore, the system first searches for similar pump-turbine models in the historical database based on model matching degree, constructs a weighted local proxy model, calculates the sensitivity of geometric parameters to performance parameters, performs multi-objective optimization, and outputs the parameter range that satisfies the constraints and the corresponding confidence assessments of efficiency improvement, pressure pulsation reduction and cavitation margin change, so as to realize the automatic push from "matching degree" to "geometric suggestion".

[0168] The pump-turbine performance prediction method provided in this embodiment inputs historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters into a bidirectional long short-term memory neural network to obtain intermediate feature vectors. This effectively extracts temporal correlations and multi-dimensional key features from historical data, providing a comprehensive data foundation for performance prediction. The method also inputs the pump-turbine physical field parameters and intermediate feature vectors into a physical information neural network to obtain performance prediction parameters. This method integrates data-driven features and physical constraints, ensuring the physical rationality and preliminary accuracy of performance prediction. Based on component spatial coordinates, it performs... The physical field distribution simulation yields the physical field distribution values ​​of the pump-turbine, providing detailed physical field data support for subsequent loss calculations and performance analysis. Based on performance prediction parameters, component spatial coordinates, pump-turbine physical field distribution values, and dimensionless parameters of the pump-turbine, an adaptive weighted loss value is determined. This value dynamically adjusts the loss weights according to prediction deviations and physical field characteristics, enhancing the sensitivity of the loss function to key errors. Finally, the adaptive weighted loss value is used to iteratively optimize the model parameters. Through dynamic loss constraints, the parameters are continuously optimized, significantly improving the model's prediction accuracy and generalization ability for pump-turbine performance.

[0169] This embodiment also provides a pump-turbine performance prediction device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0170] This embodiment provides a water pump turbine performance prediction device, such as... Figure 6 As shown, it includes:

[0171] The acquisition module 601 is used to acquire the historical geometric parameters, historical operating parameters and historical performance parameters of the target water pump turbine.

[0172] The calculation module 602 is used to calculate the dimensionless parameters of the pump-turbine based on historical geometric parameters and historical operating parameters; among which, the dimensionless parameters of the pump-turbine include the speed ratio under power generation conditions, the speed ratio under pumping conditions, the cavitation coefficient, and the symmetry factor of the bidirectional flow channel.

[0173] Module 603 is used to construct a performance prediction model for a water pump and turbine; the performance prediction model for a water pump and turbine includes a bidirectional long short-term memory neural network and a physical information neural network.

[0174] Training module 604 is used to train the pump-turbine performance prediction model based on historical geometric parameters, historical operating parameters, dimensionless parameters of the pump-turbine, and historical performance parameters, so as to obtain the trained pump-turbine performance prediction model.

[0175] The prediction module 605 is used to predict the performance of the target pump-turbine using the trained pump-turbine performance prediction model, and obtain the pump-turbine performance prediction result.

[0176] In some alternative implementations, the computing module 602 includes:

[0177] The first calculation unit is used to calculate the power generation operating speed ratio based on the power generation head, power generation flow rate and power generation speed in the historical operating parameters.

[0178] The second calculation unit is used to calculate the pumping speed ratio based on the pumping head, pumping flow rate and pumping speed in the historical operating parameters.

[0179] The third calculation unit is used to calculate the cavitation coefficient based on the pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density, and runner outlet velocity from historical geometric parameters.

[0180] The fourth calculation unit is used to calculate the symmetry factor of the bidirectional flow channel based on the power generation head and pumping head.

[0181] In some alternative implementations, training module 604 includes:

[0182] The first input unit is used to input historical geometric parameters, historical operating parameters, dimensionless parameters of water pumps and turbines, and historical performance parameters into a bidirectional long short-term memory neural network to obtain intermediate feature vectors.

[0183] The second input unit is used to obtain the physical field parameters of the water pump turbine. The physical field parameters of the water pump turbine and the intermediate feature vector are input into the physical information neural network to obtain the performance prediction parameters.

[0184] The simulation unit is used to obtain the component spatial coordinates corresponding to the target water pump turbine, and to simulate the physical field distribution of the target water pump turbine based on the component spatial coordinates, thereby obtaining the physical field distribution value of the water pump turbine.

[0185] The determination unit is used to determine the adaptive weighted loss value based on performance prediction parameters, component spatial coordinates, pump-turbine physical field distribution values, and pump-turbine dimensionless parameters.

[0186] The optimization unit is used to iteratively optimize the model parameters using adaptive weighted loss values ​​to obtain the trained pump-turbine performance prediction model.

[0187] In some alternative implementations, the first input unit includes:

[0188] The structured sub-unit is used to perform structured processing on historical geometric parameters, historical operating parameters, dimensionless parameters of water pumps and turbines, and historical performance parameters to obtain a structured time series matrix.

[0189] The normalization sub-unit is used to normalize the structured time series matrix and perform residual noise reduction to obtain a normalized input vector.

[0190] The first determining sub-unit is used to generate time-series weights based on historical operating parameters and dimensionless parameters of the pump and turbine.

[0191] The encoding subunit is used to perform two-layer encoding on the normalized temporal feature vector to obtain the encoded temporal feature vector.

[0192] The pooling subunit is used to perform attention pooling and projection processing based on the encoded temporal feature vector and temporal weights to obtain intermediate feature vectors.

[0193] In some alternative implementations, the determining unit includes:

[0194] The first calculation subunit is used to calculate the residuals of the continuity equation, momentum equation, and boundary conditions based on the physical field distribution values ​​of the water pump and turbine and the spatial coordinates of the components.

[0195] The second determining sub-unit is used to determine the loss weighting coefficient based on the dimensionless parameters of the water pump and turbine.

[0196] The second calculation subunit is used to calculate the adaptive weighted loss value based on the loss weighting coefficient, performance prediction parameters, historical performance parameters, continuity equation residuals, momentum equation residuals, and boundary condition residuals.

[0197] In some alternative implementations, it also includes:

[0198] The evaluation module is used to evaluate the confidence level of the predicted performance of the water pump turbine and adjust the blade angle of the target water pump turbine based on the confidence level evaluation results.

[0199] The pump-turbine performance prediction device provided in this embodiment of the invention can execute the pump-turbine performance prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0200] The following is a detailed reference. Figure 7 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0201] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0202] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the pump-turbine performance prediction method of the embodiments of the present invention.

[0203] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0204] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the pump-turbine performance prediction method shown in the above embodiments is implemented.

[0205] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0206] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method of predicting the performance of a pump-turbine, characterized in that, The method comprises: acquiring historical geometric parameters, historical operating parameters and historical performance parameters corresponding to a target pump-turbine; calculating pump-turbine dimensionless parameters based on the historical geometric parameters and the historical operating parameters; wherein the pump-turbine dimensionless parameters comprise a power generation condition speed ratio, a pumping condition speed ratio, a cavitation coefficient and a bidirectional flow passage symmetry factor; constructing a pump-turbine performance prediction model; wherein the pump-turbine performance prediction model comprises a bidirectional long short-term memory neural network and a physical information neural network; training the pump-turbine performance prediction model based on the historical geometric parameters, the historical operating parameters, the pump-turbine dimensionless parameters and the historical performance parameters to obtain a trained pump-turbine performance prediction model; performing performance prediction on a target pump-turbine by using the trained pump-turbine performance prediction model to obtain a pump-turbine performance prediction result; the training of the pump-turbine performance prediction model based on the historical geometric parameters, the historical operating parameters, the pump-turbine dimensionless parameters and the historical performance parameters to obtain the trained pump-turbine performance prediction model comprises: inputting the historical geometric parameters, the historical operating parameters, the pump-turbine dimensionless parameters and the historical performance parameters into the bidirectional long short-term memory neural network to obtain an intermediate feature vector; acquiring pump-turbine physical field parameters, inputting the pump-turbine physical field parameters and the intermediate feature vector into the physical information neural network to obtain performance prediction parameters; acquiring component space coordinates corresponding to a target pump-turbine, performing physical field distribution simulation on the target pump-turbine based on the component space coordinates to obtain pump-turbine physical field distribution values; determining an adaptive weighted loss value based on the performance prediction parameters, the component space coordinates, the pump-turbine physical field distribution values and the pump-turbine dimensionless parameters; iteratively optimizing model parameters by using the adaptive weighted loss value to obtain the trained pump-turbine performance prediction model.

2. The method of claim 1, wherein, the calculation of the pump-turbine dimensionless parameters based on the historical geometric parameters and the historical operating parameters comprises: calculating the power generation condition speed ratio based on power generation head, power generation flow rate and power generation speed in the historical operating parameters; calculating the pumping condition speed ratio based on pumping head, pumping flow rate and pumping speed in the historical operating parameters; calculating the cavitation coefficient based on pump-turbine outlet pressure, pump-turbine inlet pressure, liquid density and runner outlet speed in the historical geometric parameters; calculating the bidirectional flow passage symmetry factor based on the power generation head and the pumping head.

3. The method of claim 1, wherein, the inputting of the historical geometric parameters, the historical operating parameters, the pump-turbine dimensionless parameters and the historical performance parameters into the bidirectional long short-term memory neural network to obtain an intermediate feature vector comprises: performing structural processing on the historical geometric parameters, the historical operating parameters, the pump-turbine dimensionless parameters and the historical performance parameters to obtain a structured time series matrix; The normalized input vector is obtained by normalizing and residual noise reduction processing on the structured time sequence matrix; Generate time sequence weight based on the historical operation parameter and the dimensionless parameter of the pump-turbine; Double-layer coding is performed on the normalized time sequence feature vector to obtain the coded time sequence feature vector; Based on the coded time sequence feature vector and the time sequence weight, attention pooling processing and projection processing are performed to obtain the intermediate feature vector.

4. The method of claim 1, wherein, The adaptive weighted loss value is determined based on the performance prediction parameter, the component spatial coordinates, the pump-turbine physical field distribution value and the dimensionless parameter of the pump-turbine, comprising: Based on the pump-turbine physical field distribution value and the component spatial coordinates, respectively calculate the continuity equation residual, momentum equation residual and boundary condition residual; Determine the loss weighting coefficient based on the dimensionless parameter of the pump-turbine; Based on the loss weighting coefficient, the performance prediction parameter, the historical performance parameter, the continuity equation residual, the momentum equation residual and the boundary condition residual, calculate the adaptive weighted loss value.

5. The method of claim 1, wherein, Also includes: Perform confidence evaluation on the performance prediction result of the pump-turbine, and adjust the blade angle of the target pump-turbine based on the confidence evaluation result.

6. A water pump water turbine performance prediction device characterized by comprising: The device comprises: An acquisition module is configured to acquire historical geometric parameters, historical operation parameters and historical performance parameters corresponding to a target pump-turbine; A calculation module is configured to calculate a dimensionless parameter of a pump-turbine based on the historical geometric parameters and the historical operation parameters; wherein the dimensionless parameter of the pump-turbine includes a generating condition speed ratio, a pumping condition speed ratio, a cavitation coefficient and a two-way flow passage symmetry factor; A construction module is configured to construct a pump-turbine performance prediction model; wherein the pump-turbine performance prediction model includes a bidirectional long short-term memory neural network and a physical information neural network; A training module is configured to train the pump-turbine performance prediction model based on the historical geometric parameters, the historical operation parameters, the dimensionless parameter of the pump-turbine and the historical performance parameters, to obtain a trained pump-turbine performance prediction model; A prediction module is configured to perform performance prediction on a target pump-turbine by using the trained pump-turbine performance prediction model, to obtain a pump-turbine performance prediction result; The training module comprises: A first input unit is configured to input historical geometric parameters, historical operation parameters, a dimensionless parameter of a pump-turbine and historical performance parameters into a bidirectional long short-term memory neural network, to obtain an intermediate feature vector; A second input unit is configured to acquire pump-turbine physical field parameters, and input the pump-turbine physical field parameters and the intermediate feature vector into a physical information neural network, to obtain a performance prediction parameter; A simulation unit is configured to acquire component spatial coordinates corresponding to a target pump-turbine, and perform physical field distribution simulation on the target pump-turbine based on the component spatial coordinates, to obtain a pump-turbine physical field distribution value; A determination unit is configured to determine an adaptive weighted loss value based on a performance prediction parameter, component spatial coordinates, a pump-turbine physical field distribution value and a dimensionless parameter of the pump-turbine. An optimization unit is configured to iteratively optimize the model parameters using the adaptive weighted loss value to obtain a trained pump-turbine performance prediction model.

7. An electronic device, comprising: The method comprises the following steps: A memory and a processor are in communication connection with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the pump-turbine performance prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing a computer to perform the pump-turbine performance prediction method according to any one of claims 1 to 5.

9. A computer program product, characterised in that, The computer instructions are configured to cause a computer to perform the pump-turbine performance prediction method according to any one of claims 1 to 5.

Images