Water pumping and storage motor set water diversion pipeline system modeling method based on data driving

Abstract

The invention discloses a water pumping and storage motor set water diversion pipeline system modeling method based on data driving. The method comprises the following steps that 1, a water diversion pipeline system hydraulic characteristic model structure is defined; 2, identifying model parameters based on data driving and a particle swarm algorithm; 3, establishing and applying a water diversion pipeline system model; according to the method, a mathematical model structure capable of representing a hydraulic loss characteristic curve of a pipeline system is constructed, and key hydraulic loss parameters in the model are accurately identified by utilizing actual operation data of a unit and a particle swarm optimization algorithm, so that a model capable of accurately reflecting hydraulic behaviors of the system under different working conditions is established.

Description

Technical Field

[0001] This invention relates to the field of hydropower generation and hydraulic system technology, and in particular to a data-driven modeling method for water intake pipeline systems of pumped-storage hydropower units. Background Technology

[0002] A diagram showing the relationships between the various elements in a pumped storage power station is shown below. Figure 1 As shown. The water diversion pipeline system (including water diversion tunnels, pressure steel pipes, etc.) of a pumped storage power station is a key part of water energy transmission. Its tunnel-pipeline elastic hydrodynamic model is shown in the figure. Figure 2 As shown, its hydraulic characteristics, especially head loss, directly affect the unit's operating efficiency and energy conversion performance. Accurately modeling the water diversion pipeline system and quantifying its hydraulic loss characteristics is of great significance for the power plant's optimized scheduling, condition assessment, fault diagnosis, and efficiency-enhancing retrofits.

[0003] Traditional methods for analyzing the hydraulic characteristics of water diversion pipeline systems mainly include:

[0004] (1) Calculation based on theoretical formulas (such as Darcy-Weisbach formula and Kolbrook-White formula). These methods rely on accurate estimation of parameters such as pipe roughness and water flow Reynolds number. However, in actual engineering, roughness is often difficult to obtain accurately and may change over time.

[0005] (2) Computational fluid dynamics (CFD) simulation. This method can provide detailed flow field information, but it requires high computing resources, is complex to model, and the accuracy of the results is highly dependent on the boundary condition settings and the selection of the turbulence model.

[0006] (3) Although the empirical formula fitting based on field test can reflect the actual situation, the test cost is high and the universality of the fitted empirical formula is poor.

[0007] With the development of sensor technology and data acquisition systems, a large amount of unit operation data has been recorded. How to utilize this data to efficiently and effectively build models that accurately reflect the actual hydraulic characteristics of water diversion pipeline systems, especially how to distinguish the loss characteristics of different pipe sections and consider the dynamic influence of operating parameters such as guide vane opening on the system's hydraulic characteristics, is a problem worthy of research in the field. Existing data-driven methods still have room for improvement in terms of model refinement and the robustness of parameter identification. Summary of the Invention

[0008] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a data-driven modeling method for the water intake pipeline system of pumped-storage hydropower units, which can accurately identify key hydraulic loss parameters in the model, thereby establishing a model that can accurately reflect the hydraulic behavior of the system under different operating conditions.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a data-driven modeling method for water diversion pipeline systems of pumped-storage hydropower units, which includes the following steps:

[0010] Step 1: Define the hydraulic characteristic model structure of the water diversion pipeline system;

[0011] Step 2: Model parameter identification based on data-driven approach and particle swarm optimization algorithm;

[0012] Step 3: Establishment and application of the water diversion pipeline system model.

[0013] Preferably, step 1 specifically includes the following process:

[0014] The final form of the hydraulic characteristic model of the water diversion pipeline system is used to calculate the dynamic head H. d And clarify the components of each head loss and the actual flow rate Q. d Relationship:

[0015] H d =H s +H loss_total

[0016]

[0017] K f_actual =f t_actual +f p_actual +f g_actual (G)

[0018] Wherein: H d : Dynamic head of the unit; H s : Static head; H loss_total Q: Total head loss of the water diversion pipeline system; d : Actual flow rate of water pumped by the unit;

[0019] K f_actual The total head loss factor is calculated based on the actual flow rate Q. d The coefficient of the square, which consists of three parts:

[0020] f t_actual Model parameters characterizing the frictional and local loss characteristics of a water diversion tunnel or tailrace tunnel section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified;

[0021] f p_actual The model parameters characterizing the friction loss and local loss characteristics of a pressure steel pipe section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified;

[0022] fg_actual (G): The model parameter component characterizing the change in head loss caused by the alteration of the flow pattern inside and near the unit due to the change in guide vane opening G. It is based on the actual flow rate Q. d The coefficient of the square varies with the guide vane opening G.

[0023] Preferably, the f g_actual (G) The specific calculation is as follows:

[0024]

[0025] Where G is the current guide vane opening, G max Q is the maximum opening of the guide vane. max This is the reference maximum flow rate used for normalization.

[0026] Preferably, step 2 includes the following process: employing the Multi-Swarm Optimization (MPSO) algorithm and using actual unit operating data to optimize the model parameters f. t_actual and f p_actual To identify.

[0027] Preferably, step 2 specifically includes the following sub-steps:

[0028] Step 2.1: Data preparation and preprocessing;

[0029] Step 2.2: Constructing the fitness function;

[0030] Step 2.3: Parameter identification process for multi-subpopulation particle swarm optimization algorithm;

[0031] Step 2.4: Final determination of model parameters and assignment of physical meaning.

[0032] Preferably, step 2.1 is performed as follows:

[0033] 1) Collect data on pumped storage power station units under pumping conditions, including different guide vane openings (G). samples Traffic Q samples Measured dynamic head H d_measured and static head H s_samples Multiple sets of operational data;

[0034] 2) Obtain the length L of the water diversion tunnel t and pressure steel pipe length L p ;

[0035] 3) Set the maximum guide vane opening G max ;

[0036] 4) Calculate the maximum flow rate Q max =max(Q samples The traffic is then normalized.

[0037]

[0038] 5) Handling guide vane opening G samples A special case where the value is 0;

[0039] 6) Calculate the component of the guide vane opening-related loss coefficient that acts on the normalized squared flow rate for each sample point:

[0040]

[0041] Preferably, step 2.2 is performed as follows:

[0042] The fitness function in the optimization process is constructed based on normalized flow to identify the fixed loss coefficient components acting on the square of normalized flow; the parameters to be optimized are defined as x = [x(1), x(2)], which are f t and f p Preliminary estimates under normalized flow; the fitness function Fitness(x) is designed as follows:

[0043]

[0044] Where λ is the L2 regularization coefficient; here x(1) and x(2) are related to f g_normalized_samples Coefficients of the same dimension all act on Q. 2 normalized_samples .

[0045] Preferably, step 2.3 is performed as follows:

[0046] In each iteration, perform the following update steps:

[0047] Inertia weight update:

[0048] Speed ​​update: v i =w·v i +c1·r1·(pbest i -x i )+c2·r2·(gbest group(i) -x i );

[0049] Location update: x i =x i +v i Where c1 and c2 are learning factors, and r1 and r2 ∈ [0, 1] are random numbers;

[0050] Individual and group optimal solution update: Each particle calculates a new fitness; if the current particle's fitness is better than its historical best, then update pbest; if the particle is better than the gbest of its subgroup, then update the subgroup optimal solution; compare the best of all subgroups and update the global best.

[0051] Elite exchange mechanism: Every few generations, the best particles are exchanged between different subpopulations to improve information exchange between subpopulations and prevent them from getting trapped in local optima;

[0052] Boundary handling: If a particle's position goes out of bounds, a constraint method or a bounce method is used to keep it within the search range;

[0053] Local perturbation mechanism: Apply a small perturbation to the current global best particle every few generations to improve the refinement of the local search and increase the convergence accuracy;

[0054] Termination condition: The maximum number of iterations is reached or the fitness function converges to a preset threshold, and the final optimal parameter solution is returned. t * ,f p * ].

[0055] Preferably, step 2.4 is performed as follows:

[0056] After optimization, the globally optimal parameter combination [x] is obtained. 1_opt ,x 2_opt ], corresponding to gbest(best_group,1) and gbest(best_group,2) in the code; these two values ​​are applied to Q. 2 normalized_samples Preliminary identification results of the fixed loss portion;

[0057] Parameter allocation and scaling transformation based on physical structure information: to obtain the final effect on the actual flow Q 2 d Model parameters f t_actual and f p_actual And allocate them reasonably according to the pipeline length:

[0058] ∑f fixed_normalized =x 1_opt +x 2_opt

[0059]

[0060] The loss coefficient component f related to guide vane opening g_actual (G) has been given in the model structure definition, and its calculation is based on:

[0061]

[0062] Preferably, step 3 is performed as follows:

[0063] Through the above steps, a set of key model parameters that can characterize the hydraulic properties of the water diversion pipeline system are finally obtained: f t_actual : Loss characteristic parameters of the tunnel section, acting on Q 2 d ;f p_actual : Loss characteristic parameters of steel pipe section, acting on Q 2 d Combined with the loss function related to guide vane opening:

[0064]

[0065] This constitutes a complete, data-driven characteristic model of the piping system:

[0066]

[0067] This model can be used to predict any given working condition (G, Q). d H s Dynamic head H under ) d and the coefficient of friction f of the water tunnel t The friction coefficient f of pressure pipeline p This can be used to evaluate the hydraulic performance of pipelines or as the basis for more complex optimization and control algorithms.

[0068] Beneficial effects of this invention:

[0069] 1. This invention establishes a mathematical model structure for the hydraulic characteristics of a water diversion pipeline system based on actual operating data. This model explicitly represents the total head loss as a function of the square of the actual flow rate, and further decomposes it into loss components related to tunnel sections, pressure steel pipe sections, and guide vane opening changes, possessing clear physical meaning and structural expression. This invention proposes a modeling method for guide vane opening-related head loss terms. By normalizing the guide vane opening and introducing an empirical function to model its impact on the total head loss, it effectively captures the dynamic influence of unit operating condition changes on the system's hydraulic characteristics. This invention employs a multi-subject particle swarm optimization (MPSO) algorithm to optimize the model parameters. This invention improves the global search capability and robustness of parameter identification by constructing a regularized fitness function and introducing a multi-subpopulation co-evolution mechanism using unit operating data under pumping conditions. Furthermore, it rationally allocates and scales the identification results by using the physical length ratio of tunnels and pressure steel pipes, achieving physical correspondence and engineering interpretation of model parameters, thus enhancing the model's practicality and generalization ability. Finally, this invention establishes a complete data-driven water diversion pipeline system model, which can be used for dynamic head prediction and system performance analysis. This model supports simulation and evaluation under different operating conditions, possesses strong engineering applicability, and can be used for optimized scheduling, hydraulic characteristic analysis, and equipment diagnosis.

[0070] 2. High model accuracy: The model identifies parameters through a data-driven approach, enabling it to more accurately reflect the complex hydraulic characteristics of the water diversion pipeline system under actual operating conditions, including the dynamic impact of guide vane opening. Clear physical meaning: The model structure decomposes the total head loss into parts related to the main pipe sections (tunnels, steel pipes) and parts related to operational adjustments (guide vane opening). The identified parameters have clear physical correspondences, aiding in understanding the sources and distribution of losses. Robust and efficient identification method: The improved multi-subpopulation particle swarm optimization algorithm employed has strong global search capabilities and convergence efficiency, effectively identifying model parameters and avoiding getting trapped in local optima. Data dependence and engineering applicability: The method relies primarily on readily available conventional operating data from power plants, eliminating the need for complex field tests or detailed geometric parameters, thus possessing good engineering applicability and promotional value. Support for optimization decision-making: The accurate hydraulic model established provides crucial technical support for the operation optimization of pumped storage power plants, equipment status assessment, hydraulic performance analysis of pumps and turbines, and pipeline system maintenance and renovation.

[0071] 3. The method of the present invention constructs a mathematical model structure that can characterize the hydraulic loss characteristic curve of the pipeline system, and uses actual unit operating data and particle swarm optimization algorithm to accurately identify the key hydraulic loss parameters in the model, thereby establishing a model that can accurately reflect the hydraulic behavior of the system under different operating conditions. Attached Figure Description

[0072] Figure 1 This is a diagram illustrating the overall relationship of a pumped storage power station.

[0073] Figure 2 Diagram of the elastic hydrodynamic model of a tunnel-water pipe;

[0074] Figure 3 A flowchart illustrating a data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system;

[0075] Figure 4 A comparison chart of dynamic head fitting;

[0076] Figure 5 This is a graph showing the algorithm's iteration curve. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0078] Example 1: As Figure 3 As shown, a data-driven modeling method for the water intake pipeline system of a pumped-storage hydropower unit includes the following steps:

[0079] Step 1: Define the hydraulic characteristic model structure of the water diversion pipeline system;

[0080] The final form of the hydraulic characteristic model of the water diversion pipeline system is used to calculate the dynamic head H. d And clarify the components of each head loss and the actual flow rate Q. d Relationship:

[0081] H d =H s +H loss_total

[0082]

[0083] K f_actual =f t_actual +f p_actual +f g_actual (G)

[0084] Wherein: H d Unit dynamic head (m); H s Static head (m); H loss_total Q: Total head loss of the water diversion pipeline system (m); d Actual pumping flow rate of the unit (m³) 3 / s);

[0085] K f_actual The total head loss factor is calculated based on the actual flow rate Q. d The coefficient of the square, which consists of three parts:

[0086] f t_actual Model parameters characterizing the frictional and local loss characteristics of a water diversion tunnel or tailrace tunnel section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified;

[0087] f p_actual The model parameters characterizing the friction loss and local loss characteristics of a pressure steel pipe section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified;

[0088] f g_actual (G): The model parameter component characterizing the change in head loss caused by the alteration of the flow pattern inside and near the unit due to the change in guide vane opening G. It is based on the actual flow rate Q. d The coefficient of the square varies with the guide vane opening G. The specific calculation is as follows:

[0089]

[0090] Where G is the current guide vane opening (%), G max Q represents the maximum guide vane opening (%). max The reference maximum flow rate (m) used for normalization 3 / s).

[0091] The model structure clearly states that the total head loss is proportional to the square of the actual flow rate, and decomposes the proportionality coefficient into a part related to the fixed geometry of the pipe (f). t_actual ,f p_actual ) and the variable components (f) related to the operating state (guide vane opening G). g_actual (G)), all these coefficients are unified to act on Q d 2 .

[0092] Step 2: Model parameter identification based on data-driven and particle swarm optimization algorithm; preferably, step 2 includes the following process: using the Multi-Swarm Particle Swarm Optimization (MPSO) algorithm, and using actual unit operating data to identify model parameters f. t_actual and f p_actual To identify.

[0093] Step 2.1: Data preparation and preprocessing;

[0094] 1) Collect data on pumped storage power station units under pumping conditions, including different guide vane openings (G). samples Traffic Q samples Measured dynamic head H d_measured and static head H s_samples Multiple sets of operational data;

[0095] 2) Obtain the length L of the water diversion tunnel t and pressure steel pipe length L p ;

[0096] 3) Set the maximum guide vane opening G max ;

[0097] 4) Calculate the maximum flow rate Q max =max(Q samples The traffic is then normalized.

[0098]

[0099] 5) Handling guide vane opening G samples A special case where the value is 0;

[0100] 6) Calculate the component of the guide vane opening-related loss coefficient that acts on the normalized squared flow rate for each sample point:

[0101]

[0102] Step 2.2: Fitness function construction (model evaluation criteria);

[0103] The fitness function in the optimization process is constructed based on normalized flow to identify the fixed loss coefficient components acting on the square of normalized flow; the parameters to be optimized are defined as x = [x(1), x(2)], which are f t and f p Preliminary estimates under normalized flow; the fitness function Fitness(x) is designed as follows:

[0104]

[0105] Where λ is the L2 regularization coefficient; here x(1) and x(2) are related to f g_normalized_samples Coefficients of the same dimension all act on Q. 2 normalized_samples .

[0106] Step 2.3: Parameter identification process for multi-subpopulation particle swarm optimization algorithm;

[0107] In each iteration, perform the following update steps:

[0108] Inertia weight update (enhancing particle search balance):

[0109] Speed ​​update: v i =w·v i +c1·r1·(pbest i -x i )+c2·r2·(gbest group(i)-x i );

[0110] Location update: x i =x i +v i Where c1 and c2 are learning factors, and r1 and r2 ∈ [0, 1] are random numbers;

[0111] Individual and group optimal solution update: Each particle calculates a new fitness; if the current particle's fitness is better than its historical best, then update pbest; if the particle is better than the gbest of its subgroup, then update the subgroup optimal solution; compare the best of all subgroups and update the global best.

[0112] Elite Exchange: Every few generations (e.g., every 10 generations), the best particles are exchanged between subpopulations to improve information exchange between subpopulations and prevent them from getting trapped in local optima.

[0113] Boundary handling: If a particle's position goes out of bounds, a constraint method or a bounce method is used to keep it within the search range;

[0114] Local perturbation mechanism (Optional): Apply a small perturbation to the current global best particle every few generations to improve the refinement of the local search and increase the convergence accuracy;

[0115] Termination condition: The maximum number of iterations is reached or the fitness function converges to a preset threshold, and the final optimal parameter solution is returned. t * ,f p * ].

[0116] Step 2.4: Final determination of model parameters and assignment of physical meaning.

[0117] After optimization, the globally optimal parameter combination [x] is obtained. 1_opt ,x 2_opt ], corresponding to gbest(best_group,1) and gbest(best_group,2) in the code; these two values ​​are applied to Q. 2 normalized_samples Preliminary identification results of the fixed loss portion;

[0118] Parameter allocation and scaling transformation based on physical structure information: to obtain the final effect on the actual flow Q 2 d Model parameters f t_actual and f p_actual And allocate them reasonably according to the pipeline length:

[0119] ∑f fixed_normalized =x1_opt +x 2_opt

[0120]

[0121] The loss coefficient component f related to guide vane opening g_actual (G) has been given in the model structure definition, and its calculation is based on:

[0122]

[0123] Step 3: Establishment and application of the water diversion pipeline system model.

[0124] Through the above steps, a set of key model parameters that can characterize the hydraulic properties of the water diversion pipeline system are finally obtained: f t_actual : Loss characteristic parameters of the tunnel section, acting on Q 2 d ;f p_actual : Loss characteristic parameters of steel pipe section, acting on Q 2 d Combined with the loss function related to guide vane opening:

[0125]

[0126] This constitutes a complete, data-driven characteristic model of the piping system:

[0127]

[0128] This model can be used to predict any given working condition (G, Q). d H s Dynamic head H under ) d and the coefficient of friction f of the water tunnel t The friction coefficient f of pressure pipeline p This can be used to evaluate the hydraulic performance of pipelines or as the basis for more complex optimization and control algorithms.

[0129] Example 2:

[0130] Suppose a pumped storage power station unit collects several sets of operating data under pumping conditions, including guide vane opening G. samples (unit: %), flow rate Q samples (unit: m) 3 / s), dynamic head H d_measured (Unit: m), Static head H s_samples (Unit: m). The length L of the water diversion system (tunnel section) for this unit is known. t =1800m, length L of pressure steel pipe section p =500m. Maximum guide vane opening G max =100%.

[0131] 1. Data preparation and preprocessing for modeling:

[0132] Load data;

[0133] Calculate the maximum flow rate: Q max =max(Q samples );

[0134] Flow normalization:

[0135] Handling zero guide vane opening: G samples (G samples ==0)=1e-6

[0136] Calculate the components of the guide vane opening loss coefficient acting on the normalized square of the flow rate:

[0137]

[0138] 2: Model parameter identification algorithm settings:

[0139] The settings are as follows:

[0140] Number of subgroups numGroups = 10;

[0141] Number of particles per subgroup nParticlesPerGroup=40;

[0142] Maximum number of iterations maxIter = 250;

[0143] Optimization dimension dim = 2 (corresponding to f) t and f p (Preliminary estimate);

[0144] The lower limit for parameter search is lb = [0,0], and the upper limit is ub = [5,5] (this upper limit can be adjusted based on experience);

[0145] Learning factors c1 = 3, c2 = 3;

[0146] Inertial weight range w max =0.95,w min =0.1;

[0147] Regularization coefficient λ = 1e-4

[0148] 3: Definition of the model fit evaluation function:

[0149] The fitness function is:

[0150]

[0151] Where x = [x(1), x(2)]

[0152] 4: Perform model parameter identification and optimization:

[0153] Particle positions are initialized using Latin hypercube sampling. Optimization is performed following the iterative logic of the multi-subject population particle swarm optimization process described earlier, including:

[0154] Update the inertia weight w;

[0155] Update individual optimal p best and p best_val ;

[0156] Update subgroup optimal g best and g best_val ;

[0157] Perform elite swaps (e.g., every 10 generations);

[0158] Update particle velocity and position, and perform boundary treatment;

[0159] Perform local perturbations (e.g., every 10 generations).

[0160] Record the convergence process.

[0161] 5: Final model parameters determined:

[0162] After optimization, the preliminarily optimized fixed loss coefficient component x acting on the normalized squared flow rate is obtained. 1_opt and x 2_opt Based on the length L of the water diversion tunnel t and pressure steel pipe length L p Perform weighted allocation and convert it into the corresponding actual traffic Q. 2 d Final model parameters:

[0163] ∑f fixed_normalized =x 1_opt +x 2_opt

[0164]

[0165] The corresponding parameter obtained in this example is: f t =0.003144,f p =0.001089.

[0166] 6: Model Construction and Verification of Water Diversion Pipeline System

[0167] For each sample data point, calculate its corresponding guide vane opening loss coefficient component that affects the square of the actual flow rate:

[0168]

[0169] Using the identified model parameters f t_actual ,f p_actual and the calculated f g_actual_samples A complete hydraulic characteristic model of the water diversion pipeline system was constructed, and the dynamic head predicted by the model was calculated:

[0170]

[0171] The dynamic head H predicted by the model d_model Compared with the measured dynamic head H d_measured A comparison was conducted to verify the accuracy and effectiveness of the established model, with specific results as follows: Figure 4 As shown.

[0172] Through the above embodiments, a data-driven model can be established that accurately describes the hydraulic characteristics of the pumped-storage hydropower unit's water intake pipeline system, with all parameters uniformly defined as the square of the actual flow rate. Figure 5 The iterative curves of the algorithm are also provided.

[0173] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A data-driven modeling method for water intake pipeline systems of pumped-storage hydropower units, characterized in that: It includes the following steps: Step 1: Define the hydraulic characteristic model structure of the water diversion pipeline system; Step 2: Model parameter identification based on data-driven approach and particle swarm optimization algorithm; Step 3: Establishment and application of the water diversion pipeline system model.

2. The data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 1, characterized in that: Step 1 is specifically... The process includes the following: The final form of the hydraulic characteristic model of the water diversion pipeline system is used to calculate the dynamic head H. d And clarify the components of each head loss and the actual flow rate Q. d Relationship: H d =H s +H loss_total K f_actual =f t_actual +f p_actual +f g_actual (G) Wherein: H d : Dynamic head of the unit; H s : Static head; H loss_total Q: Total head loss of the water diversion pipeline system; d : Actual pumping flow rate of the unit; K f_actual The total head loss factor is calculated based on the actual flow rate Q. d The coefficient of the square, which consists of three parts: f t_actual Model parameters characterizing the frictional and local loss characteristics of a water diversion tunnel or tailrace tunnel section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified; f p_actual The model parameters characterizing the friction loss and local loss characteristics of a pressure steel pipe section are specific to the actual flow rate Q. d The coefficient of the square, which is a fixed parameter to be identified; f g_actual (G): The model parameter component characterizing the change in head loss caused by the alteration of the flow pattern inside and near the unit due to the change in guide vane opening G. It is based on the actual flow rate Q. d The coefficient of the square varies with the guide vane opening G.

3. The data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 2, characterized in that: The f g_actual (G) The specific calculation is as follows: Where G is the current guide vane opening, G max Q represents the maximum opening of the guide vane. max This is the reference maximum flow rate used for normalization.

4. The data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 1, characterized in that: Step 2 The process includes the following: employing the Multi-Swarm Optimization (MPSO) algorithm, and using actual unit operating data to refine the model parameters f. t_actual and f p_actual To identify.

5. The data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 4, characterized in that: Step 2 specifically includes the following sub-steps: Step 2.1: Data preparation and preprocessing; Step 2.2: Constructing the fitness function; Step 2.3: Parameter identification process for multi-subpopulation particle swarm optimization algorithm; Step 2.4: Final determination of model parameters and assignment of physical meaning.

6. The data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 5, characterized in that: The specific process of step 2.1 is as follows: 1) Collect data on pumped storage power station units under pumping conditions, including different guide vane openings (G). samples Traffic Q samples Measured dynamic head H d_measured and static head H s_samples Multiple sets of operational data; 2) Obtain the length L of the water diversion tunnel t and pressure steel pipe length L p ; 3) Set the maximum guide vane opening G max ; 4) Calculate the maximum flow rate Q max =max(Q samples The traffic is then normalized. 5) Handling guide vane opening G samples A special case where the value is 0; 6) Calculate the components of the guide vane opening-related loss coefficients that act on the normalized squared flow rate for each sample point:

7. A data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 5, characterized in that: The specific process of step 2.2 is as follows: The fitness function in the optimization process is constructed based on normalized flow to identify the fixed loss coefficient components acting on the square of normalized flow; the parameters to be optimized are defined as x = [x(1), x(2)], which are f t and f p Preliminary estimates under normalized flow; the fitness function Fitness(x) is designed as follows: Where λ is the L2 regularization coefficient; here x(1) and x(2) are related to f g_normalized_samples Coefficients of the same dimension all act on Q. 2 normalized_samples .

8. A data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 5, characterized in that: The specific process of step 2.3 is as follows: In each iteration, perform the following update steps: Inertia weight update: Speed ​​update: v i =w·v i +c1·r1·(pbest i -x i )+c2·r2·(gbest group(i) -x i ); Location update: x i =x i +v i Where c1 and c2 are learning factors, and r1 and r2 ∈ [0, 1] are random numbers; Individual and group optimal solution update: Each particle calculates a new fitness; if the current particle's fitness is better than its historical best, then update pbest; if the particle is better than the gbest of its subgroup, then update the subgroup optimal solution; compare the best of all subgroups and update the global best. Elite exchange mechanism: Every few generations, the best particles are exchanged between different subpopulations to improve information exchange between subpopulations and prevent them from getting trapped in local optima; Boundary handling: If a particle's position goes out of bounds, a constraint method or a bounce method is used to keep it within the search range; Local perturbation mechanism: Apply a small perturbation to the current global best particle every few generations to improve the refinement of the local search and increase the convergence accuracy; Termination condition: The maximum number of iterations is reached or the fitness function converges to a preset threshold, and the final optimal parameter solution is returned. t * ,f p * ].

9. A data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 5, characterized in that: The specific process of step 2.4 is as follows: After optimization, the globally optimal parameter combination [x] is obtained. 1_opt ,x 2_opt ], corresponding to gbest(best_group,1) and gbest(best_group,2) in the code; these two values ​​are applied to Q. 2 normalized_samples Preliminary identification results of the fixed loss portion; Parameter allocation and scaling transformation based on physical structure information: to obtain the final effect on the actual flow Q 2 d Model parameters f t_actual and f p_actual And allocate them reasonably according to the pipeline length: ∑f fixed_normalized =x 1_opt +x 2_opt The loss coefficient component f related to guide vane opening g_actual (G) has been given in the model structure definition, and its calculation is based on:

10. A data-driven modeling method for a pumped-storage hydropower unit's water intake pipeline system according to claim 1, characterized in that: The specific process of step 3 is as follows: Through the above steps, a set of key model parameters that can characterize the hydraulic properties of the water diversion pipeline system are finally obtained: f t_actual : Loss characteristic parameters of the tunnel section, acting on Q 2 d ;f p_actual : Loss characteristic parameters of steel pipe section, acting on Q 2 d Combined with the loss function related to guide vane opening: This constitutes a complete, data-driven characteristic model of the piping system: This model can be used to predict any given working condition (G, Q). d H s Dynamic head H under ) d and the coefficient of friction f of the water tunnel t The friction coefficient f of pressure pipeline p This can be used to evaluate the hydraulic performance of pipelines or as the basis for more complex optimization and control algorithms.

Images