Water hammer optimization method based on lightgbm-tpe model and improved multi-objective grey wolf algorithm

Abstract

The application provides a water hammer optimization method based on a LightGBM-TPE model and an improved multi-objective grey wolf algorithm. A numerical model of hydraulic transient of a pressurized water delivery system is constructed based on the method of characteristics, and variables such as slow closing valves and air valves are determined; sample data sets are generated by using Latin hypercube sampling, and the super parameters are optimized in combination with a TPE algorithm to construct a high-precision LightGBM water hammer simulation proxy model; the proxy model is analyzed by using a SHAP interpretability method to identify the core influence parameters of the control system negative pressure; a nonlinear convergence factor, a differential evolution mutation mechanism and an elite archive guide strategy are introduced to construct an improved multi-objective grey wolf algorithm, and a collaborative optimization model with the system minimum pressure maximization and the protection device size minimization as double objectives is established; the coupled proxy model is used for global multi-objective optimization, and the optimal protection scheme is output through non-dominated sorting and weighted screening. The application breaks through the traditional hydraulic transient simulation computing power bottleneck and realizes efficient and intelligent collaborative setting of the pipe network water hammer protection parameters.

Description

Technical Field

[0001] This invention relates to the fields of fluid mechanics, hydraulic engineering and intelligent optimization technology, and in particular to a water hammer optimization method based on the LightGBM-TPE model and an improved multi-objective gray wolf algorithm. Background Technology

[0002] In pressurized pipeline systems such as long-distance water transmission, pumping station water lifting, and industrial circulating cooling water systems, extreme operating conditions such as pump shutdowns and sudden valve closures can easily cause drastic changes in flow velocity within the pipeline, inducing water hammer (hydraulic transients). Severe water hammer can often lead to extreme negative pressure in the system, or even water column separation and flow interruption, seriously threatening the integrity of the pipeline structure and the safe operation of the system.

[0003] Combining air valves, slow-closing butterfly valves, and pressure regulating towers for protection is a common practice in the industry. However, traditional solutions based on empirical formulas or single-parameter trial calculations struggle to achieve optimal matching of multiple devices. Furthermore, water hammer numerical simulations based on the traditional method of characteristics (MOC) suffer from extremely long computation times when facing multi-objective intelligent optimization in high-dimensional parameter spaces due to the successive invocation of the complete hydraulic model, failing to meet the requirements for efficient optimization. In addition, existing intelligent algorithms face technical bottlenecks when solving multi-objective parameter tuning problems, including susceptibility to local optima, uneven distribution of Pareto solution sets, and a lack of interpretability in machine learning models (black box effect). Summary of the Invention

[0004] This invention proposes a water hammer optimization method based on the LightGBM-TPE model and an improved multi-objective gray wolf algorithm. It overcomes the computing power bottleneck by constructing a proxy model, uses SHAP to explain the model's decision-making mechanism, and relies on the improved multi-objective algorithm to balance the safety and economy of the water hammer protection scheme.

[0005] The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm provided by this invention adopts the following technical solution:

[0006] A water hammer optimization method based on the LightGBM-TPE model and an improved multi-objective gray wolf algorithm includes:

[0007] S1. Construct a numerical model of the hydraulic transient process of a pressurized water conveyance system under hydraulic transient conditions based on the method of characteristics, and determine the key decision variables for water hammer protection.

[0008] S2. The Latin hypercube sampling method is used to generate a sample dataset within the set range of decision variable values, and the LightGBM model is established and trained. At the same time, the TPE algorithm is used to optimize the model hyperparameters and construct a water hammer simulation proxy model.

[0009] S3. Analyze the trained LightGBM-TPE surrogate model using the SHAP interpretability method, calculate the SHAP values ​​of each key decision variable, and identify the core influencing parameters that play a dominant role in the control of water hammer negative pressure in the system.

[0010] S4. By introducing a nonlinear convergence factor, a differential evolution mutation mechanism, and an elite archive guidance strategy, an improved multi-objective gray wolf algorithm is constructed. A collaborative optimization model is established with the dual optimization objectives of maximizing the minimum system pressure and minimizing the size of the protective equipment.

[0011] S5. Using the improved multi-objective gray wolf algorithm coupled with the LightGBM-TPE surrogate model, multi-objective global optimization is performed. After non-dominated sorting and normalized weighting, the Pareto front compromise optimal solution is selected to obtain the final water hammer collaborative protection prediction value and optimal parameter configuration.

[0012] Furthermore, the pressurized water supply system includes long-distance water diversion projects, pumping station water lifting systems, and industrial circulating cooling water systems; the key decision variables include the phased closing time of the slow-closing valve and the air intake and exhaust coefficients and size parameters of the air valve.

[0013] Furthermore, the process of optimizing the model hyperparameters using the TPE algorithm in step S2 includes: constructing a probabilistic model of hyperparameters using a tree-structured Parzen estimator, selecting the next set of hyperparameter evaluation points by maximizing the expectation improvement criterion, thereby fitting the nonlinear mapping relationship between the protection parameters and the complex hydraulic response.

[0014] Furthermore, in step S4, the mathematical model for the gray wolf surrounding prey and position update in the improved multi-target gray wolf algorithm is as follows:

[0015] ;

[0016] ;

[0017] in Indicates the current iteration number; This represents the optimal solution for the objective, i.e., the position vector of the prey. This represents the current position vector of the individual gray wolf. This represents the distance vector between an individual gray wolf and its prey. and This represents the vector of synergy coefficients.

[0018] Furthermore, the synergy coefficient vector and The calculation formula is:

[0019] ;

[0020] ;

[0021] in, This represents the convergence factor, whose value decreases linearly from 2 to 0 with the number of iterations.

[0022] Furthermore, by , , The mathematical expression for the three-alpha wolf to guide the updating of the population position is:

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] in , , These represent the top three gray wolf individuals in the current population, ranked first, second, and third in fitness value, respectively. , , The position vector of the alpha wolf; , , Represents a random vector; , , Represents the convergence coefficient vector; , , These represent the updated position vectors of the individual gray wolf after being affected by the three alpha wolves; This represents the updated position vector of the individual gray wolf.

[0031] Furthermore, in step S5, the dual optimization objective function is defined as follows:

[0032] Maximize system security objectives:

[0033] ;

[0034] Minimize the system's economic objective:

[0035] ;

[0036] in, This indicates the combination of key parameters for protective equipment; This represents the minimum transient pressure occurring along the line under hydraulically transient conditions of the system; Key dimensional parameters of protective equipment that indicate associated economic costs.

[0037] The beneficial effects of the water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm provided by this invention are as follows:

[0038] 1. This invention utilizes the LightGBM-TPE algorithm to construct a proxy model for water hammer simulation, successfully replacing the traditional time-consuming characteristic line method hydraulic iterative calculation process. This model can accurately and quickly map the nonlinear relationship between multidimensional complex protection parameters and hydraulic transient response, providing a high-confidence, low-computational-cost evaluation tool for large-scale parameter optimization.

[0039] 2. This invention introduces the SHAP interpretability analysis method, which breaks through the limitations of the traditional black-box proxy model and quantitatively reveals the contribution and influence of different equipment parameters (such as valve closing time, air valve size, etc.) on suppressing extreme negative pressure, providing theoretical support for prioritizing the control of key equipment in engineering design.

[0040] 3. This invention proposes an improved multi-objective Grey Wolf algorithm (DEMOGWO) that integrates mechanisms such as differential evolution, effectively overcoming the problem of traditional optimization algorithms easily getting trapped in local optima. Under the premise of eliminating destructive water hammer (ensuring system safety), it achieves the minimization and rational configuration of protective equipment (while also considering engineering economy), providing a quantifiable and universal intelligent optimization framework for water hammer protection in fluid pipeline systems. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0042] Figure 2 A comparison chart of the predicted values ​​from the surrogate model and the actual values ​​from the feature line method;

[0043] Figure 3 Ranking of protection parameters based on SHAP values ​​and generating a bee colony diagram;

[0044] Figure 4 Pareto front solution set distribution diagram optimized for DEMOGWO algorithm;

[0045] Figure 5 A comparison chart of minimum pressure along the pipeline before and after optimization. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. 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. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0047] Example 1:

[0048] See Figures 1 to 5 As shown, this embodiment of the invention provides a water hammer optimization method based on the LightGBM-TPE model and an improved multi-objective gray wolf algorithm, including:

[0049] S1. Construct a numerical model of the hydraulic transient process of a pressurized water conveyance system under hydraulic transient conditions based on the method of characteristics, and determine the key decision variables for water hammer protection.

[0050] S2. The Latin hypercube sampling method is used to generate a sample dataset within the set range of decision variable values, and the LightGBM model is established and trained. At the same time, the TPE algorithm is used to optimize the model hyperparameters and construct a water hammer simulation proxy model.

[0051] S3. Analyze the trained LightGBM-TPE surrogate model using the SHAP interpretability method, calculate the SHAP values ​​of each key decision variable, and identify the core influencing parameters that play a dominant role in the control of water hammer negative pressure in the system.

[0052] S4. By introducing a nonlinear convergence factor, a differential evolution mutation mechanism, and an elite archive guidance strategy, an improved multi-objective gray wolf algorithm is constructed. A collaborative optimization model is established with the dual optimization objectives of maximizing the minimum system pressure and minimizing the size of the protective equipment.

[0053] S5. Using the improved multi-objective gray wolf algorithm coupled with the LightGBM-TPE surrogate model, multi-objective global optimization is performed. After non-dominated sorting and normalized weighting, the Pareto front compromise optimal solution is selected to obtain the final water hammer collaborative protection prediction value and optimal parameter configuration.

[0054] In this embodiment, the pressurized water supply system includes a long-distance water diversion project, a pumping station water lifting system, and an industrial circulating cooling water system; the key decision variables include the phased closing time of the slow-closing valve and the air intake and exhaust coefficients and size parameters of the air valve.

[0055] In this embodiment, the process of optimizing the model hyperparameters using the TPE algorithm in step S2 includes: constructing a probabilistic model of hyperparameters using a tree-structured Parzen estimator, selecting the next set of hyperparameter evaluation points by maximizing the expected improvement criterion, thereby fitting the nonlinear mapping relationship between the protection parameters and the complex hydraulic response.

[0056] In this embodiment, the mathematical model for the improved multi-target gray wolf algorithm's gray wolf surrounding prey and position update in step S4 is as follows:

[0057] ;

[0058] ;

[0059] in Indicates the current iteration number; This represents the optimal solution for the objective, i.e., the position vector of the prey. This represents the current position vector of the individual gray wolf. This represents the distance vector between an individual gray wolf and its prey. and This represents the vector of synergy coefficients.

[0060] In this embodiment, the cooperative coefficient vector and The calculation formula is:

[0061] ;

[0062] ;

[0063] in, This represents the convergence factor, whose value decreases linearly from 2 to 0 with the number of iterations.

[0064] In this embodiment, by , , The mathematical expression for the three-alpha wolf to guide the updating of the population position is:

[0065] ;

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] in , , These represent the top three gray wolf individuals in the current population, ranked first, second, and third in fitness value, respectively. , , The position vector of the alpha wolf; , , Represents a random vector; , , Represents the convergence coefficient vector; , , These represent the updated position vectors of the individual gray wolf after being affected by the three alpha wolves; This represents the updated position vector of the individual gray wolf.

[0073] In this embodiment, the dual optimization objective function is defined as follows in step S5:

[0074] Maximize system security objectives:

[0075] ;

[0076] Minimize the system's economic objective:

[0077] ;

[0078] in, This indicates the combination of key parameters for protective equipment; This represents the minimum transient pressure occurring along the line under hydraulically transient conditions of the system; Key dimensional parameters of protective equipment that indicate associated economic costs.

[0079] Example 2:

[0080] Unlike Example 1, this example uses an industrial power plant circulating cooling water system as the engineering background and optimizes water hammer protection for a full pump shutdown in case of an accident. The system uses an air valve and a two-stage slow-closing butterfly valve for coordinated protection. The specific operation steps are as follows:

[0081] Step S1: Construct a numerical model of water hammer and determine decision variables.

[0082] A transient flow numerical model of the circulating cooling water system was established using the method of characteristics. Five key control parameters for the combined protection of the two-stage slow-closing butterfly valve and air valve were selected as decision variables: air valve intake coefficient Cin, air valve exhaust coefficient Cout, air valve size D, butterfly valve first-stage closing time T1, and butterfly valve second-stage closing time T2.

[0083] Step S2: Construct the LightGBM-TPE proxy model.

[0084] Step S21: Using the Latin hypercube sampling method, 2000 sample datasets are drawn within the engineering allowable threshold of each decision variable.

[0085] Step S22: Divide the data into training and testing sets in an 8:2 ratio, using 5 variables as inputs and the minimum pressure Hmin along the system line as output.

[0086] Step S23: Establish a prediction model and use the TPE algorithm to optimize the hyperparameters of LightGBM. Test results show that the LightGBM-TPE surrogate model has a determination coefficient R2 of 0.9969 and a root mean square error RMSE of only 0.051, and its prediction accuracy is superior to neural network models such as CNN and BP.

[0087] Step S3: SHAP-based parameter sensitivity analysis.

[0088] The marginal contribution of each variable was calculated using the SHAP method. Beehive diagram analysis showed that the first-stage closing time T1 of the butterfly valve (SHAP value 0.628) and the air valve size D (SHAP value 0.326) are the core variables determining the negative pressure level of the system. Extending T1 and increasing D can effectively suppress the negative pressure peak.

[0089] Step S4: Establish a dual-objective collaborative optimization model.

[0090] A dual objective function is constructed to balance safety and economy: objective 1 is to maximize the minimum system pressure Hmin, and objective 2 is to minimize the air valve size D. The solution architecture is set up using the DEMOGWO algorithm, which integrates nonlinear convergence factor, DE mutation mechanism and elite archive.

[0091] Step S5: Model optimization and result evaluation.

[0092] Using LightGBM-TPE as the evaluator, the DEMOGWO algorithm was run (with a safety weight of 0.7 and an economy weight of 0.3). After iteration, the compromise optimal solution was extracted from the Pareto front.

[0093] The initial unoptimized scheme is: Cin=0.60, Cout=0.60, D=200mm, T1=10s, T2=60s. At this time, the system Hmin is -3.98m, which poses a risk of water hammer caused by flow interruption.

[0094] The optimized output scheme of the method of the present invention is: Cin=0.69, Cout=0.50, D=249mm, T1=30s, T2=48.6s.

[0095] Table 1 Comparison between the initial scheme and the algorithm optimization scheme

[0096] Air valve intake coefficient 0.60 0.69 0.69 Air valve exhaust coefficient 0.60 0.50 0.50 Air valve dimensions (mm) 200 249 249 Phase 1 Closure Time (s) 10 30 30 Second phase shutdown time (s) 60 48.6 48.6 System minimum pressure - safety (m) -3.98 -1.17 -1.16

[0097] As shown in Table 1, after optimization using the method of this invention, the minimum system pressure of the pipeline network was significantly increased from -3.98m to -1.16m, an improvement of 70.9%, effectively eliminating the risk of water column separation. Simultaneously, the air valve size remained within a reasonable range (approximately 250mm), achieving a perfect balance between safety protection capabilities and equipment cost. This example fully demonstrates the universality and efficiency of this invention in the field of water hammer protection for pressurized pipeline networks.

[0098] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A water hammer optimization method based on the LightGBM-TPE model and an improved multi-objective gray wolf algorithm, characterized in that, include: S1. Construct a numerical model of the hydraulic transient process of a pressurized water conveyance system under hydraulic transient conditions based on the method of characteristics, and determine the key decision variables for water hammer protection. S2. The Latin hypercube sampling method is used to generate a sample dataset within the set range of decision variable values, and the LightGBM model is established and trained. At the same time, the TPE algorithm is used to optimize the model hyperparameters and construct a water hammer simulation proxy model. S3. Analyze the trained LightGBM-TPE surrogate model using the SHAP interpretability method, calculate the SHAP values ​​of each key decision variable, and identify the core influencing parameters that play a dominant role in the control of water hammer negative pressure in the system. S4. By introducing a nonlinear convergence factor, a differential evolution mutation mechanism, and an elite archive guidance strategy, an improved multi-objective gray wolf algorithm is constructed. A collaborative optimization model is established with the dual optimization objectives of maximizing the minimum system pressure and minimizing the size of the protective equipment. S5. Using the improved multi-objective gray wolf algorithm coupled with the LightGBM-TPE surrogate model, multi-objective global optimization is performed. After non-dominated sorting and normalized weighting, the Pareto front compromise optimal solution is selected to obtain the final water hammer collaborative protection prediction value and optimal parameter configuration.

2. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 1, characterized in that, The pressurized water supply system includes long-distance water diversion projects, pumping station water lifting systems, and industrial circulating cooling water systems; the key decision variables include the phased closing time of the slow-closing valve and the air intake and exhaust coefficients and size parameters of the air valve.

3. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 2, characterized in that, The process of optimizing the model hyperparameters using the TPE algorithm in step S2 includes: constructing a probabilistic model of hyperparameters using a tree-structured Parzen estimator, selecting the next set of hyperparameter evaluation points by maximizing the expected improvement criterion, thereby fitting the nonlinear mapping relationship between the protection parameters and the complex hydraulic response.

4. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 3, characterized in that, In step S4, the mathematical model for the improved multi-target gray wolf algorithm's gray wolf surrounding prey and position update is as follows: ； ； in Indicates the current iteration number; This represents the optimal solution for the objective, i.e., the position vector of the prey. This represents the current position vector of the individual gray wolf. This represents the distance vector between an individual gray wolf and its prey. and This represents the vector of synergy coefficients.

5. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 4, characterized in that, The coordination coefficient vector and The calculation formula is: ； ； in, This represents the convergence factor, whose value decreases linearly from 2 to 0 with the number of iterations. , It is a random vector between [0,1].

6. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 5, characterized in that, Depend on , , The mathematical expression for the three-alpha wolf to guide the updating of the population position is: ； ； ； ； ； ； ； in , , These represent the top three gray wolf individuals in the current population, ranked first, second, and third in fitness value, respectively. , , The alpha wolf's position vector; , , Represents a random vector; , , Represents the convergence coefficient vector; , , These represent the updated position vectors of the individual gray wolf after being affected by the three alpha wolves; This represents the updated position vector of the individual gray wolf.

7. The water hammer optimization method based on the LightGBM-TPE model and the improved multi-objective gray wolf algorithm according to claim 6, characterized in that, In step S5, the dual optimization objective function is defined as follows: Maximize system security objectives: ； Minimize the system's economic objective: ； in, This indicates the combination of key parameters for protective equipment; This represents the minimum transient pressure occurring along the line under hydraulically transient conditions of the system; Key dimensional parameters of protective equipment that indicate associated economic costs.

Images