A method and system for hydraulic design of centrifugal pump impeller

Abstract

The present application relates to the technical field of centrifugal pump hydraulic design, in particular to a centrifugal pump impeller hydraulic design method and system, comprising collecting hydraulic data, working condition data and water area characteristic data of the centrifugal pump in different water area positions and different running time periods, and constructing a multi-dimensional original data set; a double-branch deep learning model is constructed, the input and output dimensions and the selection of the activation function are determined, the association between the water area data and the impeller geometric parameters is established, and the accurate reverse mapping of the geometric parameters is realized; the present application realizes the accurate reverse mapping of the geometric parameters and constructs a double-branch targeted deep learning model, determines the basis for the selection of the input and output dimensions and the activation function, outputs the adaptability correction coefficient through the water area adaptation branch, establishes the association mechanism of the hydraulic performance, the water area characteristics and the impeller geometric parameters, directly outputs the feasible impeller geometric shape, and the model training stability and the prediction accuracy are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of centrifugal pump hydraulic design technology, and more specifically, to a method and system for hydraulic design of centrifugal pump impellers. Background Technology

[0002] Centrifugal pumps are core equipment for fluid transportation in fields such as water conservancy, chemical industry, municipal engineering, and marine engineering. As a key component for energy conversion, the hydraulic structure design of the impeller directly determines the pump's operating efficiency, energy consumption level, and service life. Existing centrifugal pump impeller hydraulic design methods mostly adopt a general process of "data acquisition-normalization-model training-performance prediction", which has many technical defects, cannot meet the actual application needs of industrial multi-water and multi-operating conditions, and is also difficult to adapt to the inventiveness requirements of patent authorization.

[0003] The core defects of the existing technology are as follows: First, the data collection dimension is singular, only collecting hydraulic and operating condition data without considering the differences in physical, chemical and impurity characteristics of different water areas. This results in the impeller being unable to adapt to operation in multiple water areas, and is prone to corrosion, wear and a sharp drop in efficiency. Second, the model structure is vague, and the correlation between hydraulic and operating condition data and impeller geometric parameters has not been established, making it impossible to achieve reverse mapping of geometric parameters and difficult to output feasible geometric shapes. To address the aforementioned issues, a centrifugal pump impeller hydraulic design method and system are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a hydraulic design method for centrifugal pump impellers to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, a centrifugal pump impeller hydraulic design method is provided, including the following steps: S1. Collect hydraulic data, operating condition data and water characteristic data of centrifugal pumps in different water locations and different operating periods to construct a multi-dimensional raw dataset; S2. Construct a dual-branch deep learning model, clarify the input and output dimensions and activation function selection, establish the correlation between water area data and impeller geometric parameters, and achieve accurate reverse mapping of geometric parameters; S3: Multi-condition prediction and geometric parameter dual calibration. Adjust the centrifugal pump operating condition data, input the corresponding water area characteristic data, and predict the impeller geometric parameters through the optimized deep learning model to obtain multiple initial schemes. S4: Perform CFD simulations on the optimized impeller geometry under multiple water areas and operating conditions, compare the simulation results with the model prediction results, verify the hydraulic performance error and water area adaptability error, iterate and correct until the error meets the standard, output the final impeller geometry, and complete the design.

[0006] As a further improvement to this technical solution, the hydraulic data in S1 explicitly includes six core indicators: flow rate, head, power, speed, efficiency, and net positive suction head (NPSH); the operating condition data includes three key parameters: load variation range, running time, and start-stop frequency; and the water body characteristic data covers four categories and eight parameters: water body physical characteristics, chemical characteristics, impurity characteristics, and environmental auxiliary characteristics, adapting to multiple water body operation scenarios.

[0007] As a further improvement to this technical solution, the construction of the multi-dimensional element dataset in S1 includes the following steps: S1.1. Use the 3σ criterion to detect and remove abnormal data such as flow rate jumps and power anomalies, and use linear interpolation to supplement the missing key data; S1.2. The min-max normalization method is used to map all data to the [0,1] interval to eliminate the influence of units. The dataset is divided into training set, validation set and test set according to water type and working condition intensity.

[0008] As a further improvement to this technical solution, the normalization formula in S1.2 is: :( These are the measured values ​​of the parameters. , (This is the maximum or minimum value of the parameter sample).

[0009] As a further improvement to this technical solution, the construction of a dual-branch deep learning model in S2 includes the following steps: S2.1 Construct a deep learning-based dual-branch collaborative architecture. The model includes four parts: performance correlation branch, water area adaptation branch, shared hidden layer, and output layer. Each part works together to realize the bidirectional correlation between water area data, hydraulic condition data, and impeller geometric parameters. S2.2. Clearly define the input dimensions, output dimensions, and parameter definitions of each branch of the model to ensure that the input data corresponds one-to-one with the output geometric parameters and achieve accurate reverse mapping; S2.3. Based on the model architecture, input and output characteristics and reverse mapping requirements, select activation functions accordingly and clarify the selection criteria and parameter settings for activation functions at each layer. S2.4. By branch collaboration, feature fusion and parameter linkage, a direct correlation between water area data and impeller geometric parameters is established to ensure the accuracy and relevance of the reverse mapping.

[0010] As a further improvement to this technical solution, the adjustment of the centrifugal pump operating data in S3 includes the following steps: S3.1 Adjust the centrifugal pump operating data, input the adjusted operating data and the corresponding water area characteristic data into the trained and optimized model, and predict multiple initial geometric parameter schemes. S3.2 Based on the water area characteristic data, the weights of each water area dimension are allocated through the analytic hierarchy process, and the weighted sum is used to obtain the comprehensive adaptation coefficient K (value 0-1). The K value range is then adjusted accordingly. S3.3 Verify the geometric parameters and processing technology, eliminate schemes that are too thin for blades or have unreasonable flow channel dimensions, and obtain the optimized impeller geometry.

[0011] As a further improvement to this technical solution, the emergency monitoring mode is activated in step S4, and the simulation results are compared with the model prediction results in step S4, including the following steps: S4.1 Based on computational fluid dynamics (CFD) technology, perform three-dimensional flow field simulation of the optimized impeller geometry under multiple water areas and working conditions, and set the water parameters (density, viscosity) and working condition boundary conditions for the corresponding water areas. S4.2 Compare the simulation results with the model prediction results, calculate the hydraulic performance error and the water area adaptability error, until both errors of the simulation results and the prediction results meet the standards.

[0012] As a further improvement to this technical solution, the formula for the S4.2 hydraulic performance error algorithm is as follows: ; Symbol definition: The value represents the hydraulic performance error (%), and n represents the number of core hydraulic parameters involved in the calculation (in this invention, n=4, corresponding to flow rate, head, efficiency, and net positive suction head). Let i be the CFD simulation value of the i-th hydraulic parameter. Let be the model predicted value of the i-th hydraulic parameter, and |.| be the absolute value symbol.

[0013] The second objective of this invention is to provide a centrifugal pump impeller hydraulic design system, including any one of the centrifugal pump impeller hydraulic design methods described above, comprising a data processing unit, a model building unit, a parameter calibration unit, and an iterative output unit; The data processing unit collects hydraulic data, operating condition data, and water characteristic data of centrifugal pumps at different water locations and during different operating periods to construct a multi-dimensional raw dataset. The model building unit constructs a dual-branch deep learning model, clarifies the input and output dimensions and activation function selection, establishes the correlation between water area data and impeller geometric parameters, and achieves accurate reverse mapping of geometric parameters; The parameter calibration unit performs multi-condition prediction and dual calibration of geometric parameters, adjusts the centrifugal pump operating data, inputs corresponding water area characteristic data, and predicts impeller geometric parameters through an optimized deep learning model to obtain multiple initial schemes. The iterative output unit performs CFD simulations on the optimized impeller geometry under multiple water areas and operating conditions, compares the simulation results with the model prediction results, verifies the hydraulic performance error and the water area adaptability error, iteratively corrects the error until the error meets the standard, and outputs the final impeller geometry.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this centrifugal pump impeller hydraulic design method, hydraulic data, operating condition data and water characteristic data of centrifugal pumps are collected in different water locations and different operating periods. The constructed multi-dimensional uniform dataset provides high-quality and high-reliability data support for the subsequent training of the dual-branch deep learning model, significantly improving the model's generalization ability and enabling adaptation to multiple water scenarios such as freshwater and seawater.

[0015] 2. In this centrifugal pump impeller hydraulic design method, a dual-branch targeted deep learning model is constructed to clarify the input and output dimensions and the selection basis of the activation function. By outputting the adaptability correction coefficient through the water area adaptation branch, a correlation mechanism between hydraulic performance, water area characteristics and impeller geometric parameters is established. This overcomes the core defect of existing models that cannot output geometric parameters in reverse, and directly outputs feasible impeller geometry. At the same time, the application of the ELU activation function effectively solves the gradient vanishing problem, and the model training stability and prediction accuracy are significantly improved. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the centrifugal pump impeller hydraulic design method of the present invention.

[0017] The meanings of the labels in the diagram are as follows: 100. Data processing unit; 200. Model building unit; 300, Parameter calibration unit; 400, Iteration output unit. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 As shown, a method for hydraulic design of centrifugal pump impellers is provided, including the following steps: S1. Collect hydraulic data, operating condition data and water characteristic data of centrifugal pumps in different water locations and different operating periods to construct a multi-dimensional raw dataset; To acquire new water area characteristic data, clarify data indicators, improve preprocessing procedures, resolve industrial data noise interference issues, and provide reliable data support for subsequent model training, therefore... The S1 hydraulic data explicitly includes six core indicators: flow rate, head, power, speed, efficiency, and net positive suction head (NPSH). The operating condition data includes three key parameters: load variation range, operating time, and start-stop frequency. The water body characteristic data covers four categories and eight parameters: physical properties (density, kinematic viscosity), chemical properties (pH value, salinity), impurity properties (impurity particle size, volume fraction), and environmental auxiliary properties (water temperature, operating altitude). This adapts to multiple water body operating scenarios. The addition of four categories and eight parameters for water body characteristics, combined with hydraulic and operating condition data, lays a reliable data foundation for multi-water body impeller design, avoids insufficient adaptability caused by single data, and breaks through the bottleneck of single data dimensions in existing technologies.

[0020] Constructing a multi-dimensional element dataset in S1 involves the following steps: S1.1. The 3σ criterion is used to detect and remove abnormal data such as flow rate jumps and power anomalies, and the missing key data is supplemented by linear interpolation to eliminate the influence of noise and dimensions in industrial data. S1.2. The min-max normalization method is used to map all data to the [0,1] interval to eliminate the influence of units. The dataset is divided into training set, validation set and test set according to water type and working condition intensity.

[0021] The normalization formula in S1.2 is: ; ( These are the measured values ​​of the parameters. , (This is the maximum or minimum value of the parameter sample).

[0022] The dataset is divided into layers based on water type and working condition intensity, with the training set, validation set, and test set in a ratio of 7:2:1. This ensures a uniform distribution of the dataset, improves the model's generalization ability, and solves the prediction bias problem caused by the uneven distribution of the reference file dataset.

[0023] S2. Construct a dual-branch deep learning model, clarify the input and output dimensions and activation function selection, establish the correlation between water area data and impeller geometric parameters, and achieve accurate reverse mapping of geometric parameters; Building a dual-branch deep learning model in S2 includes the following steps: S2.1 Construct a deep learning-based dual-branch collaborative architecture. The model consists of four parts: a performance correlation branch, a water area adaptation branch, a shared hidden layer, and an output layer. Each part works together to achieve bidirectional correlation between water area data, hydraulic condition data, and impeller geometric parameters. The specific architecture design is as follows: Branching: A dual-branch parallel input, shared hidden layer, and unified output structure is adopted. The performance correlation branch focuses on the correlation learning between hydraulic condition data and impeller performance, while the water area adaptation branch focuses on the correlation learning between water area data and impeller adaptability. The two branches achieve feature fusion through a shared hidden layer to ensure the comprehensiveness of data correlation. Inter-layer connection: The input layers of the two branches receive corresponding data independently. After the initial feature extraction is completed by their respective initial fully connected layers, they are connected to the shared hidden layer. After the shared hidden layer completes the deep feature fusion and mapping, it is connected to the output layer to realize the reverse mapping output from the input data to the impeller geometry parameters. Model size setting: Combining the input dimension and prediction accuracy requirements, the model is set to have 4 hidden layers, and the number of neurons in each layer is 2.5 times the sum of the two branch input dimensions. This avoids insufficient mapping accuracy caused by an overly simple model, and also prevents overfitting and low training efficiency caused by an overly complex model. S2.2. Clearly define the input dimensions, output dimensions, and parameter definitions of each branch of the model to ensure that the input data corresponds one-to-one with the output geometric parameters and achieve accurate reverse mapping; The specific definitions are as follows: Water area adaptation branch input dimension (8 dimensions): It is dedicated to accessing water area data, covering 4 major categories of core parameters of water area characteristics, ensuring the direct correlation between water area data and impeller geometry parameters. Specifically, it includes: water physical properties (density, kinematic viscosity), chemical properties (pH value, salinity), impurity properties (impurity particle size, volume fraction), and environmental auxiliary properties (water temperature, operating altitude). Each parameter is valid data after S1 step standardization preprocessing (mapped to the [0,1] interval), eliminating the influence of dimensions on model training; The performance-related branch input dimensions (12 dimensions) are: the hydraulic data and operating condition data after preprocessing in step S1 are connected to help enhance the accuracy of the reverse mapping of geometric parameters. Specifically, it includes: 6 core hydraulic data (flow rate, head, power, speed, efficiency, net positive suction head) and 6 key operating condition data (load variation range, running time, start-stop frequency). All data have been preprocessed by removing outliers, filling missing values, and min-max normalization. Output layer output dimensions (8 dimensions): Directly output impeller geometric parameters to achieve accurate reverse mapping from water area data and hydraulic condition data to geometric parameters. Specifically, these include: number of blades, blade thickness, blade tilt angle, impeller inlet diameter, impeller outlet diameter, meridional profile radius of curvature, flow channel cross-sectional area, and blade offset angle. All output parameters are continuous and feasible values, meeting the basic requirements of impeller manufacturing process.

[0024] S2.3. Based on the model architecture, input / output characteristics, and back-mapping requirements, select activation functions accordingly, clarify the selection criteria and parameter settings for activation functions at each layer, and solve problems such as gradient vanishing and parameter distortion in traditional activation functions, as detailed below: Based on the model architecture, input and output characteristics, and inverse mapping requirements, select activation functions accordingly and clarify the selection criteria and parameter settings for activation functions at each layer; Input layer to hidden layer activation function: ELU activation function is adopted. The selection criteria are: ELU activation function can effectively alleviate the gradient vanishing problem in deep model training. Compared with sigmoid and tanh activation functions, it has non-zero output in the negative interval, which can retain more feature information, improve the model's ability to identify subtle differences in water area data and hydraulic conditions, and thus improve the accuracy of geometric parameter back mapping. Hidden layer to output layer activation function: Purelin linear transfer function is adopted. The selection criteria are: the output layer needs to output continuous and feasible impeller geometry parameters. The linear transfer function can ensure that the output parameters are not distorted and directly correspond to the actual numerical range of the geometry parameters, avoiding parameter offset caused by nonlinear activation function, and ensuring that the reverse-mapped geometry parameters can be directly used for subsequent S4 steps of calibration and manufacturing. Inter-layer connection of activation functions: After the input layer data is processed by the initial fully connected layer of the corresponding branch, it is connected to the ELU activation function to complete the feature activation. The activated features are passed to the shared hidden layer, and the feature fusion is gradually deepened by the ELU activation function of 4 hidden layers. Finally, it is passed to the purelin linear transfer function to output 8-dimensional impeller geometric parameters, realizing the complete reverse mapping process of "input data-feature activation-feature fusion-geometric parameter output". S2.4. Through branch collaboration, feature fusion, and parameter linkage, a direct correlation is established between water area data and impeller geometric parameters to ensure the accuracy and relevance of the reverse mapping. The specific correlation method is as follows: Branch Collaborative Association: The water area adaptation branch specifically learns the correlation between water area data and impeller adaptability. Through fully connected layers and activation functions, it transforms water area data (such as salinity and impurity particle size) into adaptability feature vectors and outputs adaptability correction coefficients. Feature fusion and association: The shared hidden layer uses four fully connected layers to fuse the feature vectors from the two branches layer by layer. Each hidden layer strengthens the features through the ELU activation function, and associates and maps the key information in the water data (such as salinity corresponding to blade corrosion protection requirements and impurity particle size corresponding to flow channel size requirements) with the performance requirements in the hydraulic condition data (such as flow rate corresponding to flow channel cross-sectional area requirements), and transforms them into fused features directly related to the impeller geometry parameters. Parameter linkage and correlation: During the model initialization phase, the model is guided to establish the correspondence between "water area data features and geometric parameters" by using labeled training samples (water area data, hydraulic condition data corresponding to known optimal impeller geometric parameters). For example, high salinity water area data corresponds to the geometric parameter correction of increased blade thickness, and large impurity particle size water area data corresponds to the parameter correction of increased flow channel cross-sectional area.

[0025] By overcoming the core defects of existing models, such as fuzzy structure and inability to output geometric parameters in reverse, a precise correlation between hydraulic data and impeller geometric parameters is established through a dual-branch (performance correlation + water area adaptation) collaborative architecture, enabling direct reverse mapping output of geometric parameters.

[0026] S3: Multi-condition prediction and geometric parameter dual calibration. Adjust the centrifugal pump operating condition data, input the corresponding water area characteristic data, and predict the impeller geometric parameters through the optimized deep learning model to obtain multiple initial schemes. To ensure that the predicted geometric parameters are adapted to complex real-world operating scenarios and to avoid insufficient practicality due to single-condition design, S3 includes the following steps: S3.1 Adjust the centrifugal pump operating data (simulate different loads, start-stop frequencies and water area switching scenarios), input the adjusted operating data and corresponding water area characteristic data into the trained and optimized model, and predict multiple initial geometric parameter schemes. S3.2 Based on the water area characteristic data, the weights of each water area dimension are allocated through the analytic hierarchy process, and the weighted sum is used to obtain the comprehensive adaptation coefficient K (value 0-1). The K value range is then adjusted accordingly. S3.3 Verify the geometric parameters and processing technology, eliminate schemes that are too thin for blades or have unreasonable flow channel dimensions, and obtain the optimized impeller geometry.

[0027] The first stage of water area adaptation calibration: Based on water area characteristic data, the weights of each water area dimension are allocated using the analytic hierarchy process (AHP), and a weighted sum is obtained to obtain the comprehensive adaptation coefficient K (value 0-1). Adjustments are made according to the K value range: when K < 0.3, the flow channel cross-sectional area is increased by 10%-20%, and the blade thickness is increased by 5%-8% to resist impurity deposition and corrosion; when 0.3 ≤ K ≤ 0.7, the blade tilt angle is fine-tuned by ±1°-2° to adapt to changes in fluid density; when K > 0.7, the meridional profile curvature is optimized to improve hydraulic efficiency. The second stage of manufacturing feasibility calibration: The geometric parameters are verified to meet the processing requirements, eliminating unmanufacturable solutions such as excessively thin blades or unreasonable flow channel dimensions. The final optimized impeller geometry is obtained, resolving the issue of the reference document's output geometry potentially being unmanufacturable.

[0028] S4: Perform CFD simulations on the optimized impeller geometry under multiple water areas and operating conditions, compare the simulation results with the model prediction results, verify the hydraulic performance error and water area adaptability error, iterate and correct until the error meets the standard, output the final impeller geometry, and complete the design.

[0029] To ensure that the design results meet performance requirements and address the shortcomings of the original solution, such as lack of verification and insufficient reliability, the simulation results are compared with the model prediction results in S4, including the following steps: S4.1 Based on computational fluid dynamics (CFD) technology, perform three-dimensional flow field simulation of the optimized impeller geometry under multiple water areas and working conditions, and set the water parameters (density, viscosity) and working condition boundary conditions for the corresponding water areas. S4.2 Compare the simulation results with the model prediction results, calculate the hydraulic performance error and the water area adaptability error, until both errors of the simulation results and the prediction results meet the standards.

[0030] S4.2 Hydraulic Performance Error (Corely used to verify the deviation between the simulated values ​​and the predicted values ​​of the impeller's hydraulic performance, covering four core hydraulic parameters: flow rate, head, efficiency, and net positive suction head (NPSH). The arithmetic mean of the errors of each parameter is taken as the final hydraulic performance error). The algorithm formula is as follows: ; Symbol definition: The value represents the hydraulic performance error (%), and n represents the number of core hydraulic parameters involved in the calculation (in this invention, n=4, corresponding to flow rate, head, efficiency, and net positive suction head). Let i be the CFD simulation value of the i-th hydraulic parameter. Let be the model predicted value of the i-th hydraulic parameter, and |.| be the absolute value symbol.

[0031] The water area adaptability error calculation (corely used to verify the deviation between the impeller adaptability simulation score and the model prediction score, based on the adaptability requirements of water area characteristics, combined with the adaptability of four key water area parameters: density, viscosity, salinity, and impurity particle size, to calculate the comprehensive deviation) is as follows: ; Symbol definition: The percentage represents the water area adaptability error. The comprehensive score of CFD simulation for water adaptability is given (the score range is 0-1, which is calculated by weighting the risk of impurity deposition, corrosion risk and fluid flow smoothness in the flow field simulation). The comprehensive score for the water adaptability model is predicted (consistent with the calculation logic of the simulation score to ensure a unified comparison benchmark). Error determination and iteration logic: If ≤3 and ≤2 If both errors meet the standards, the final impeller geometry is determined and the design is completed. If either error fails to meet the standards, the model hyperparameters (number of hidden layers, regularization coefficient) are adjusted or corresponding water area training samples are added. Training, prediction, and simulation verification are carried out again, and the iteration is repeated until the design requirements are met. This ensures that the final impeller geometry meets the performance requirements of multiple water areas and multiple operating conditions, completely solving the defects of existing technologies that lack verification and have unreliable design results. At the same time, it improves the impeller operating efficiency and reduces energy consumption.

[0032] The second objective of this invention is to provide a centrifugal pump impeller hydraulic design system, including any one of the centrifugal pump impeller hydraulic design methods described above, comprising a data processing unit 100, a model building unit 200, a parameter calibration unit 300, and an iterative output unit 400. The data processing unit 100 collects hydraulic data, operating condition data and water characteristic data of centrifugal pumps at different water locations and different operating periods to construct a multi-dimensional raw dataset. The model building unit 200 constructs a dual-branch deep learning model, clarifies the input and output dimensions and activation function selection, establishes the correlation between water area data and impeller geometric parameters, and achieves accurate reverse mapping of geometric parameters; The parameter calibration unit performs dual calibration of more than 300 operating conditions prediction and geometric parameters, adjusts the centrifugal pump operating condition data, inputs corresponding water area characteristic data, and predicts impeller geometric parameters through the optimized deep learning model to obtain multiple initial schemes. The iterative output unit 400 performs multi-water area and multi-condition CFD simulations on the optimized impeller geometry, compares the simulation results with the model prediction results, verifies the hydraulic performance error and the water area adaptability error, iteratively corrects the error until the error meets the standard, and outputs the final impeller geometry.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for hydraulic design of a centrifugal pump impeller, characterized in that: Includes the following steps: S1. Collect hydraulic data, operating condition data and water characteristic data of centrifugal pumps in different water locations and different operating periods to construct a multi-dimensional raw dataset; S2. Construct a dual-branch deep learning model, clarify the input and output dimensions and activation function selection, establish the correlation between water area data and impeller geometric parameters, and achieve accurate reverse mapping of geometric parameters; S3: Multi-condition prediction and geometric parameter dual calibration. Adjust the centrifugal pump operating condition data, input the corresponding water area characteristic data, and predict the impeller geometric parameters through the optimized deep learning model to obtain multiple initial schemes. S4: Perform CFD simulations on the optimized impeller geometry under multiple water areas and operating conditions, compare the simulation results with the model prediction results, verify the hydraulic performance error and water area adaptability error, iterate and correct until the error meets the standard, and output the final impeller geometry.

2. The centrifugal pump impeller hydraulic design method according to claim 1, characterized in that: The hydraulic data in S1 explicitly includes six core indicators: flow rate, head, power, speed, efficiency, and net positive suction head (NPSH); the operating condition data includes three key parameters: load variation range, operating time, and start-stop frequency; and the water body characteristic data covers four categories and eight parameters: physical characteristics, chemical characteristics, impurity characteristics, and environmental auxiliary characteristics, adapting to multiple water body operating scenarios.

3. The centrifugal pump impeller hydraulic design method according to claim 2, characterized in that: The construction of the multi-degree element dataset in S1 includes the following steps: S1.

1. Use the 3σ criterion to detect and remove abnormal data such as flow rate jumps and power anomalies, and use linear interpolation to supplement the missing key data; S1.

2. The min-max normalization method is used to map all data to the [0,1] interval to eliminate the influence of units. The dataset is divided into training set, validation set and test set according to water type and working condition intensity.

4. The centrifugal pump impeller hydraulic design method according to claim 3, characterized in that: The normalization formula in S1.2 is as follows: :( These are the measured values ​​of the parameters. , (This is the maximum or minimum value of the parameter sample).

5. The centrifugal pump impeller hydraulic design method according to claim 1, characterized in that: The construction of the dual-branch deep learning model in S2 includes the following steps: S2.1 Construct a deep learning-based dual-branch collaborative architecture. The model includes four parts: performance correlation branch, water area adaptation branch, shared hidden layer, and output layer. Each part works together to realize the bidirectional correlation between water area data, hydraulic condition data, and impeller geometric parameters. S2.

2. Clearly define the input dimensions, output dimensions, and parameter definitions of each branch of the model to ensure that the input data corresponds one-to-one with the output geometric parameters and achieve accurate reverse mapping; S2.

3. Based on the model architecture, input and output characteristics and reverse mapping requirements, select activation functions accordingly and clarify the selection criteria and parameter settings for activation functions at each layer. S2.

4. By branch collaboration, feature fusion and parameter linkage, a direct correlation between water area data and impeller geometric parameters is established to ensure the accuracy and relevance of the reverse mapping.

6. The centrifugal pump impeller hydraulic design method according to claim 1, characterized in that: The adjustment of centrifugal pump operating data in S3 includes the following steps: S3.1 Adjust the centrifugal pump operating data, input the adjusted operating data and the corresponding water area characteristic data into the trained and optimized model, and predict multiple initial geometric parameter schemes. S3.2 Based on the water area characteristic data, the weights of each water area dimension are allocated through the analytic hierarchy process, and the weighted sum is used to obtain the comprehensive adaptation coefficient K (value 0-1). The K value range is then adjusted accordingly. S3.3 Verify the geometric parameters and processing technology, eliminate schemes that are too thin for blades or have unreasonable flow channel dimensions, and obtain the optimized impeller geometry.

7. The centrifugal pump impeller hydraulic design method according to claim 1, characterized in that: The comparison of simulation results with model prediction results in S4 includes the following steps: S4.1 Based on computational fluid dynamics technology, perform three-dimensional flow field simulation of the optimized impeller geometry under multiple water areas and working conditions, and set the water parameters and working condition boundary conditions for the corresponding water areas. S4.2 Compare the simulation results with the model prediction results, calculate the hydraulic performance error and the water area adaptability error, until both errors of the simulation results and the prediction results meet the standards.

8. The centrifugal pump impeller hydraulic design method according to claim 7, characterized in that: The formula for the S4.2 hydraulic performance error algorithm is as follows: ； Symbol definition: The value represents the hydraulic performance error (%), and n represents the number of core hydraulic parameters involved in the calculation (in this invention, n=4, corresponding to flow rate, head, efficiency, and net positive suction head). Let i be the CFD simulation value of the i-th hydraulic parameter. Let be the model predicted value of the i-th hydraulic parameter, and |.| be the absolute value symbol.

9. A centrifugal pump impeller hydraulic design system for executing a centrifugal pump impeller hydraulic design method as described in any one of claims 1 to 8, comprising a data processing unit (100), a model building unit (200), a parameter calibration unit (300), and an iterative output unit (400). The data processing unit (100) collects hydraulic data, operating condition data and water characteristic data of centrifugal pumps in different water locations and different operating periods to construct a multi-dimensional raw dataset. The model building unit (200) constructs a dual-branch deep learning model, clarifies the input and output dimensions and activation function selection, establishes the correlation between water area data and impeller geometric parameters, and achieves accurate reverse mapping of geometric parameters; The parameter calibration unit (300) performs multi-condition prediction and dual calibration of geometric parameters, adjusts the centrifugal pump operating condition data, inputs the corresponding water area characteristic data, and predicts the impeller geometric parameters through the optimized deep learning model to obtain multiple initial schemes; The iterative output unit (400) performs multi-water area and multi-condition CFD simulations on the optimized impeller geometry, compares the simulation results with the model prediction results, verifies the hydraulic performance error and water area adaptability error, iteratively corrects until the error meets the standard, and outputs the final impeller geometry.

Images