An intelligent optimization method for large-scale multi-target water hammer protection based on LSMDCSO and CNN-LSTM

Abstract

This invention discloses a large-scale multi-objective intelligent optimization method for water hammer protection based on LSMDCSO and CNN-LSTM, relating to the field of hydraulic transient protection and control technology for water diversion projects. The method includes: establishing a hydraulic transient simulation model that conforms to actual operation and constructing a sample set; establishing a pump shutdown water hammer protection prediction model based on the CNN-LSTM algorithm as the objective function; using the LSMDCSO algorithm to establish a large-scale multi-objective optimization mathematical model, with the objectives of minimizing the maximum water hammer pressure, maximizing the minimum negative pressure, minimizing the maximum reverse speed of the pump, and minimizing the protection cost, to obtain the optimal design parameter solution set; applying the entropy weight-fuzzy TOPSIS method to obtain the optimal parameters under different biases; substituting the theoretically optimal parameters into the simulation model to determine the optimization effect, and performing secondary optimization when necessary. This invention solves the problems of slow convergence, insufficient diversity, limited prediction accuracy, and subjective decision-making in existing methods in high-dimensional parameter spaces, achieving efficient and accurate optimization of water hammer protection measures.

Description

Technical Field

[0001] This invention relates to the field of hydraulic transient protection and control technology for water diversion projects, and in particular to a large-scale multi-objective intelligent optimization method for water hammer protection based on LSMDCSO and CNN-LSTM. Background Technology

[0002] Water hammer is one of the most threatening transient phenomena in pressurized water transmission systems. Unexpected pump shutdowns can cause severe pressure fluctuations, leading to extreme water hammer pressure and vaporization in pipelines. In extreme cases, this can even cause serious accidents such as pipeline rupture and pump damage, thus jeopardizing the safe operation of the water transfer project. To ensure project safety, water hammer protection equipment such as bidirectional pressure regulating towers, air tanks, and post-pump valves are typically installed, but this significantly increases project investment. Therefore, finding a balance between safety and cost-effectiveness, and resolving this contradiction through optimization algorithms, is particularly important.

[0003] As water diversion projects continue to expand in scale, the optimization problem of water hammer protection exhibits complex characteristics of large scale, multiple objectives, and strong nonlinearity. The dimensionality of decision variables is rapidly increasing (such as the closing pattern parameters of multiple valves and the structural parameters of multiple pressure regulating towers). Traditional multi-objective evolutionary algorithms face significant challenges in handling such large-scale multi-objective optimization problems: slow convergence speed, degenerate solution set diversity, and limited search capabilities.

[0004] Current research mainly focuses on optimizing pump and valve closing patterns or single water hammer protection measures. This approach often overlooks the interactions between different factors, limiting its practical application. Furthermore, research on multi-objective optimization using intelligent algorithms is relatively limited, and existing methods yield inaccurate results. Considering that there is no direct linear relationship between water hammer protection equipment parameters and protection effectiveness, studying the interactions between these parameters and constructing a more accurate comprehensive multi-objective optimization mathematical model will provide strong theoretical support for the optimized design of water transfer projects.

[0005] Existing direction sampling-based methods (such as LSMOF and LSMOEA-DS) have shown good convergence, but still have the following shortcomings: (1) Directional sparsity: relying on a single reference solution to generate search directions makes it difficult to fully cover the high-dimensional decision space; (2) Insufficient diversity: lacking an effective mechanism to balance convergence and diversity, resulting in uneven distribution of the Pareto front; (3) Limited search accuracy: insufficient fine-grained search capability after direction sampling makes it difficult to approximate the true optimal solution. Summary of the Invention

[0006] The purpose of this invention is to provide a large-scale multi-directional fuzzy sampling competitive swarm optimizer (LSMDCSO) and a convolutional neural network-long short-term memory (CNN-LSTM) intelligent optimization method for water hammer protection, in order to solve the problems of slow convergence speed, insufficient diversity, limited prediction accuracy, and insufficient objectivity in the decision-making process of existing pump shutdown water hammer protection optimization methods in high-dimensional complex parameter spaces.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A large-scale multi-target water hammer protection intelligent optimization method based on LSMDCSO and CNN-LSTM includes the following steps:

[0009] S1. Establish a hydraulic transient simulation model that conforms to the actual operation of the water diversion project, and then obtain the input parameter set of the model through the Latin hypercube sampling method. Substitute it into the model for simulation and construct the desired sample set.

[0010] S2. Establish a pump stop water hammer protection prediction model based on the CNN-LSTM algorithm, and use it as the objective function of the multi-objective optimization model;

[0011] S3. Use LSMDCSO to establish a large-scale multi-objective optimization mathematical model for the optimization problem of pump shutdown water hammer protection measures, so as to obtain the optimal solution set of pump shutdown water hammer protection design parameters.

[0012] S4. Apply the entropy weight-fuzzy TOPSIS method to obtain the optimal pump stop water hammer protection design parameters under different biases;

[0013] S5. Substitute the theoretically optimal pump shutdown water hammer protection design parameters into the hydraulic transient simulation model to determine the optimization effect; judge whether the optimization effect meets the expected effect, and perform secondary optimization if necessary.

[0014] Furthermore, the specific method of step S1 includes:

[0015] Based on the topology diagram of the water diversion project, a hydraulic transient simulation model is established using the method of characteristics. The boundary conditions include: upstream boundary, downstream boundary, downstream valve boundary, air valve boundary, unidirectional pressure regulating tower boundary, and pump station boundary.

[0016] Collect and organize monitoring data on flow, pressure, water level, and operating status of pumps, downstream valves, and unidirectional pressure regulating towers along the pipeline. Select two sets of monitoring data containing 10 minutes of continuous monitoring data with a step size of 0.01 seconds under actual operating conditions. Use these data to calibrate and verify the model. Control the verification error within 10% to obtain a hydraulic transient simulation model that conforms to the actual operation of the water diversion project.

[0017] The parameters of the pump downstream valve closing behavior and the parameters of the unidirectional pressure regulating tower are selected as the target optimization variables. Then, the input parameter set of the hydraulic transient simulation model, i.e. the target optimization variable set, is obtained by Latin hypercube sampling method. The input parameter set is substituted into the model for simulation, and the maximum water hammer pressure value, minimum negative pressure value and maximum reverse speed of the pump are output to construct the desired sample set.

[0018] Furthermore, the parameters for selecting the valve closing pattern after the pump include: fast closing time, fast closing angle, slow closing time, and slow closing angle; the parameters for the unidirectional pressure regulating tower include: tower body diameter, initial effective head, and water supply pipe diameter.

[0019] Furthermore, the specific method of step S2 includes:

[0020] 80% of the expected sample set is randomly selected as the training set, and the remaining 20% ​​is used as the test set. The training set is used to build a water hammer protection prediction model based on CNN-LSTM, and the test set is used to test the trained prediction model. If the accuracy requirement is met, it is used as the objective function of the multi-objective optimization model. The input of the objective function is the target optimization variable set, and the output is the predicted maximum water hammer pressure value, the predicted minimum negative pressure value, and the predicted maximum reverse rotation speed of the water pump.

[0021] The CNN-LSTM model consists of the following layers: input layer, one-dimensional convolutional layer, max pooling layer, LSTM layer, and fully connected layer. The one-dimensional convolutional layer has a kernel size of 3 and 128 filters. The LSTM layer has 256 hidden units.

[0022] Furthermore, in step S3, a water hammer protection prediction model for pump shutdown is introduced as the objective function in the multi-objective optimization algorithm. The optimization objectives of the established large-scale multi-objective optimization mathematical model include four aspects: minimizing the maximum water hammer pressure value, maximizing the minimum negative pressure value, minimizing the maximum reverse rotation speed of the pump, and minimizing the water hammer protection cost; among them, the water hammer protection cost is characterized by minimizing the volume of the unidirectional pressure regulating tower.

[0023] Furthermore, under the premise of ensuring the safe operation of the water diversion project, the large-scale multi-objective optimization mathematical model needs to meet the following constraints:

[0024] (1) Valve closing constraint after pump:

[0025] ;

[0026] In the formula, To expedite the closing time; For quick closing angle; This refers to the slow shutdown time; For slow closing angle;

[0027] (2) Extreme value constraints of the pipeline system:

[0028]

[0029] In the formula, This represents the maximum water hammer pressure value. This represents the maximum steady-state pressure along the route. This is the minimum negative pressure value; This is the maximum reverse rotation speed of the water pump; This refers to the rated speed of the water pump;

[0030] (3) Input variable value range constraints:

[0031]

[0032] In the formula, The diameter of the tower body; This is the initial effective head; This refers to the diameter of the water supply pipe.

[0033] Furthermore, in the specific steps of iteratively searching for the optimal solution set of pump shutdown water hammer protection parameters using LSMDCSO as the main algorithm, the initialization and representative solution selection stages include:

[0034] Input initialization parameters to generate an initial population, and set the number of iterations to 1. The initialization parameters include population size N, maximum function evaluation number MaxFEs, scaling factor, number of representative solutions, number of samples, dimension d of optimization variables, and input variable constraints.

[0035] Selecting high-quality representative solutions from the current population based on angle penalty distance to form a representative solution set RS; specifically, this includes: uniformly generating reference vectors in the target space, assigning each individual to the reference vector with the smallest angle, calculating the angle penalty distance (APD) of each individual, and selecting the individual with the smallest APD in each sub-region as the representative solution.

[0036] Furthermore, in the specific steps of the iterative search, the determination of the optimization stage and the multi-directional fuzzy sampling strategy includes:

[0037] Calculate the ratio of the current function evaluation count FEs to the maximum evaluation count MaxFEs. If the ratio is less than 0.5, execute the multi-directional fuzzy sampling strategy; otherwise, directly execute the multi-source learning competitive group optimization strategy.

[0038] The multi-directional fuzzy sampling strategy includes: generating a random number rand; if rand is less than 0.5, performing convergence direction sampling; otherwise, performing diversity direction sampling.

[0039] The convergence direction sampling is as follows: For the representative solution, the approximate gradient is calculated using a simultaneous perturbation gradient estimator. Combined with the boundary sampling direction, a convergence sampling direction set is constructed. The sampling quantity of the two directions is controlled by a scaling factor, and a sampled solution is generated along the convergence direction.

[0040] Diversity directional sampling is performed as follows: the representative solution set RS is randomly divided into two equally sized subsets RS1 and RS2, and for each pair of representative solutions... Construct sampling directions and orthogonal directions to form a diverse set of sampling directions, and generate sampling solutions along the diverse directions;

[0041] The fuzzy variable operator is applied to the generated sampled solution to dynamically adjust the fuzziness of the solution, so as to achieve a smooth transition from coarse-grained search to fine-grained search and output the sampled population.

[0042] Furthermore, in the specific steps of the iterative search, the multi-source learning competitive group optimization strategy includes:

[0043] The fitness value of each individual in the population is calculated based on the displacement density estimation.

[0044] Generate a random number rand. If rand is less than 0.5, use a global sorting mechanism for pairing; otherwise, use a random pairing mechanism.

[0045] The global sorting mechanism is as follows: sort in descending order of fitness value, the first 50% is the winner set WS, the last 50% is the loser set LS, and randomly pair from WS and LS to form N / 2 pairing groups;

[0046] The random pairing mechanism is as follows: the population is randomly divided into N / 2 pairs, and the fitness of the two individuals in each pair is compared. The one with higher fitness is the winner, and the one with lower fitness is the loser.

[0047] From each pairing group, randomly select one individual to form the learning set LE, with a size of N / 2;

[0048] For each loser, the position is updated using the competitive group optimizer update formula; for each individual in the learning set, a leader is randomly selected from the representative solution set, and its position is updated using the competitive group optimizer formula.

[0049] The updated population is merged with the current population, environmental selection is performed, and offspring populations are generated.

[0050] Check if the termination condition is met: if the maximum number of function evaluations MaxFEs is reached, output the optimal solution set; otherwise, jump to the step of selecting a high-quality representative solution from the current population based on angle penalty distance, and increment the iteration count by 1.

[0051] Furthermore, in step S4, the specific steps for obtaining the optimal pump shutdown water hammer protection design parameters under different biases using the entropy weight-fuzzy TOPSIS method include:

[0052] Construct a decision matrix, where m is the number of schemes and n is the number of indicators, and then normalize it.

[0053] The objective weights of each evaluation index are calculated using the entropy weight method;

[0054] Construct a weighted fuzzy decision matrix and use triangular fuzzy numbers to represent index values;

[0055] Determine the fuzzy positive ideal solution and the fuzzy negative ideal solution;

[0056] Calculate the distance between each scheme and the fuzzy positive ideal solution, as well as the distance between each scheme and the fuzzy negative ideal solution, and calculate the relative proximity accordingly;

[0057] The solutions are ranked according to their relative proximity. The higher the relative proximity, the better the solution. The solution with the highest relative proximity is selected as the optimal compromise solution.

[0058] The weights are determined based on different biased optimization objectives, which are divided into three types: safety and reliability priority, cost and economy priority, and unbiased optimization.

[0059] The beneficial effects of this invention are:

[0060] (1) Strong adaptability to large-scale parameter space: The LSMDCSO algorithm is introduced, and its multi-directional fuzzy sampling mechanism integrates gradient approximation sampling and orthogonal sampling, which can effectively handle decision variables of up to 10,000 dimensions, perfectly adapting to the needs of multi-equipment and multi-parameter joint optimization in modern water diversion projects.

[0061] (2) Better balance between convergence and diversity: Representative solutions are selected by angle penalty distance adaptation, and orthogonal sampling is combined to maintain population diversity, avoiding the defect of traditional directional sampling methods that are prone to getting trapped in local optima, and the Pareto front distribution is more uniform.

[0062] (3) Significantly improved search accuracy: The multi-source learning competitive group optimizer performs a refined local search on the sampled solution, which makes up for the accuracy loss of directional sampling and can more accurately approximate the true Pareto optimal solution.

[0063] (4) High prediction accuracy: The CNN-LSTM hybrid neural network is used to replace the traditional random forest model, which can explore the complex nonlinear and temporal coupling relationship between water hammer protection parameters and effects more deeply.

[0064] (5) Scientific and objective decision-making process: The integration of entropy weight method and fuzzy TOPSIS avoids the arbitrariness of subjective weighting and deals with the fuzzy uncertainty in the decision-making process.

[0065] (6) High system integration: Construct a closed-loop adaptive optimization framework of “simulation-prediction-optimization-decision-verification” to improve the robustness and engineering practicality of the system. Attached Figure Description

[0066] Figure 1 This is a flowchart of the overall process of a large-scale multi-target water hammer protection intelligent optimization method based on LSMDCSO and CNN-LSTM according to the present invention.

[0067] Figure 2 This is a schematic diagram of the CNN-LSTM water hammer protection prediction model structure of the present invention;

[0068] Figure 3 This is a flowchart of the LSMDCSO multi-objective optimization algorithm of the present invention;

[0069] Figure 4 This is a flowchart of the entropy weight-fuzzy TOPSIS decision method of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0071] Reference Figures 1 to 4 The method shown is a large-scale multi-target water hammer protection intelligent optimization method based on LSMDCSO and CNN-LSTM, which includes the following steps:

[0072] S1. Establish a hydraulic transient simulation model that conforms to the actual operation of the water diversion project, and then obtain the input parameter set of the model through the Latin hypercube sampling method. Substitute the parameter set into the model for simulation and construct the desired sample set.

[0073] This step specifically involves: combining the topology diagram of the water diversion project, establishing a hydraulic transient simulation model based on the method of characteristics, with boundary conditions including: upstream boundary, downstream boundary, downstream valve boundary, air valve boundary, unidirectional pressure regulating tower boundary, and pump station boundary.

[0074] Collect and organize monitoring data on flow rate, pressure, water level, and the operating status of pumps, downstream valves, and unidirectional pressure regulating towers along the pipeline. Select two sets of monitoring data, each lasting 10 minutes under actual operating conditions with a step size of 0.01 seconds, for model calibration and verification. Control the verification error within 10% to obtain a hydraulic transient simulation model that conforms to the actual operation of the water diversion project.

[0075] Select parameters for the valve closing pattern after the pump (fast closing time) Quick-closing angle Slow closing time Slow closing angle ) and the parameters of the unidirectional surge tank (tower diameter D, initial effective head) The diameter of the water supply pipe (d) is the objective optimization variable. The input parameter set is obtained using the Latin hypercube sampling method, substituted into the model for simulation, and the maximum water hammer pressure value is output. Minimum negative pressure value and the maximum reverse rotation speed of the water pump Construct the desired sample set.

[0076] S2. A prediction model for pump stoppage water hammer protection is established based on the CNN-LSTM algorithm, and it is used as the objective function of the multi-objective optimization model.

[0077] This step specifically involves: randomly selecting 80% of the desired sample set as the training set, and the remaining 20% ​​as the test set. Constructing... Figure 2 The CNN-LSTM network shown contains one convolutional layer (kernel size 3, number of filters 128), one max pooling layer, one LSTM layer (hidden units 256) and one fully connected layer.

[0078] The network is trained using the training set, with mean squared error as the loss function. The Adam optimizer is used for parameter updates, and an early stopping mechanism is employed to prevent overfitting. The trained prediction model is tested using the test set; if the error is large, the amount of data is increased or the network hyperparameters are adjusted, and the model is retrained; if the accuracy requirements (such as the coefficient of determination) are met, the model is considered successfully trained. This is used as the objective function of the multi-objective optimization model; the algorithm expression is as follows:

[0079] .

[0080] S3. Use LSMDCSO to establish a large-scale multi-objective optimization mathematical model for the optimization problem of pump shutdown water hammer protection measures, so as to obtain the optimal solution set of pump shutdown water hammer protection design parameters.

[0081] This step specifically involves: introducing a pump shutdown water hammer protection prediction model as the objective function in a multi-objective optimization algorithm, with... minimize, maximize, Minimizes and minimizes water hammer protection costs (based on surge tank volume: Using minimization of representation as the optimization objective, and employing LSMDCSO as the main algorithm, a large-scale multi-objective optimization mathematical model is established for the optimization problem of pump shutdown water hammer protection measures, and the optimal solution set of pump shutdown water hammer protection design parameters is iteratively searched.

[0082] like Figure 3 As shown, the specific steps to obtain the optimal solution set of water hammer protection design parameters for pump shutdown are as follows:

[0083] (1) Input initialization parameters to generate an initial population, and set the iteration count to 1. The initialization parameter of LSMDCSO is the population size. Maximum number of function evaluations Scale factor , representing the number of solutions Number of samples Optimize the dimensionality of variables and input variable constraints.

[0084] (2) Select high-quality representative solutions from the current population based on angle-penalized distance to form the representative solution set RS. Specifically, this includes: uniformly generating in the target space Given a set of reference vectors, assign each individual to the reference vector with the smallest angle to it, calculate the angle penalty distance (APD) for each individual, and select the individual with the smallest APD in each sub-region as the representative solution.

[0085] (3) Optimization Stage Judgment: Calculate the ratio of the current function evaluation count FEs to the maximum evaluation count MaxFEs. If If the multi-directional fuzzy sampling strategy is not executed, then the multi-source learning competitive group optimization strategy is executed directly.

[0086] Multi-directional fuzzy sampling strategies include:

[0087] Generate random numbers rand, if If convergence direction sampling is performed, then convergence direction sampling is performed; otherwise, diversity direction sampling is performed.

[0088] Convergence direction sampling: for representative solutions The approximate gradient is calculated using a simultaneous perturbation gradient estimator. Combined with boundary sampling direction Construct a convergent sampling direction set Through the scaling factor By controlling the number of samples in both directions, a sampled solution is generated along the convergence direction.

[0089] Diversity-oriented sampling: The representative solution set RS is randomly divided into two equal-sized subsets. and For each pair of representative solutions Construct sampling direction and orthogonal direction , forming a diverse sampling direction set Sampled solutions are generated along the diversity direction;

[0090] The fuzzy variable operator is applied to the generated sampled solution to dynamically adjust the fuzziness of the solution, so as to achieve a smooth transition from coarse-grained search to fine-grained search and output the sampled population.

[0091] Multi-source learning competitive group optimization strategies include:

[0092] The fitness value of each individual in the population is calculated based on the displacement density estimation.

[0093] Generate random numbers rand, if If a pairing is not possible, a global sorting mechanism is used for pairing; otherwise, a random pairing mechanism is used.

[0094] Global sorting mechanism: Sort by fitness value in descending order, the first 50% is the winner set WS, and the last 50% is the loser set LS. Random pairing is performed from WS and LS to form N / 2 pairing groups.

[0095] Random pairing mechanism: The population is randomly divided into N / 2 pairs. The fitness of the two individuals in each pair is compared. The one with higher fitness is the winner, and the one with lower fitness is the loser.

[0096] A learning set LE is formed by randomly selecting one individual from each pairing group, with a size of N / 2.

[0097] For each loser The position is updated using the CSO update formula; for each individual in the learning set... A leader is randomly selected from the representative solution set, and its position is updated using the CSO formula.

[0098] The updated population is merged with the current population, environmental selection is performed, and offspring populations are generated.

[0099] (4) Check if the termination condition is met: the maximum number of function evaluations MaxFEs is reached. If it is met, output the optimal solution set; otherwise, jump to step 2 and increment the iteration count by 1.

[0100] The following constraints must be met during the optimization process:

[0101] Pump post-valve closing constraint:

[0102]

[0103] Piping system extreme value constraints:

[0104]

[0105] Input variable value range constraints:

[0106] .

[0107] S4. Apply the entropy weight-fuzzy TOPSIS method to obtain the optimal pump stop water hammer protection design parameters under different biases.

[0108] This step specifically involves: based on technical specifications and actual needs, and while ensuring the safe operation of the water transfer project, reducing the cost of water hammer protection, using the entropy-weighted fuzzy TOPSIS method to compare and select the optimal Pareto solution set to obtain optimal pump shutdown water hammer protection design parameters with different biases. The specific steps include:

[0109] (1) Constructing the decision matrix Where m is the number of schemes and n is the number of indicators, and normalization processing is performed;

[0110] (2) Calculate the objective weights of each evaluation index using the entropy weight method. :

[0111]

[0112] (3) Construct a weighted fuzzy decision matrix and use triangular fuzzy numbers to represent the index values;

[0113] (4) Determine the fuzzy positive ideal solution With fuzzy negative ideal solution ;

[0114] (5) Calculate the distance between each solution and the ideal solution. The distance between the sum and the negative ideal solution and relative closeness :

[0115]

[0116] (6) Based on relative proximity Sort the solutions. The larger the value, the better the solution. The largest possible solution is taken as the optimal compromise.

[0117] Weights are determined based on different biased optimization objectives. There are three types of optimization objectives: safety and reliability priority, cost and economy priority, and unbiased optimization.

[0118] S5. Substitute the theoretically optimal pump shutdown water hammer protection design parameters into the hydraulic transient simulation model to determine the optimization effect; judge whether the optimization effect meets the expected effect, and perform secondary optimization if necessary.

[0119] This step specifically involves: substituting the theoretically optimal design parameters for pump shutdown water hammer protection into the hydraulic transient simulation model to determine the optimization effect.

[0120] Furthermore, the specific method of step S1 includes:

[0121] Based on the topology diagram of the water diversion project, a hydraulic transient simulation model is established using the method of characteristics. The boundary conditions include: upstream boundary, downstream boundary, downstream valve boundary, air valve boundary, unidirectional pressure regulating tower boundary, and pump station boundary.

[0122] Collect and organize monitoring data on flow, pressure, water level, and operating status of pumps, downstream valves, and unidirectional pressure regulating towers along the pipeline. Select two sets of monitoring data containing 10 minutes of continuous monitoring data with a step size of 0.01 seconds under actual operating conditions. Use these data to calibrate and verify the model. Control the verification error within 10% to obtain a hydraulic transient simulation model that conforms to the actual operation of the water diversion project.

[0123] The parameters of the pump downstream valve closing behavior and the parameters of the unidirectional pressure regulating tower are selected as the target optimization variables. Then, the input parameter set of the hydraulic transient simulation model, i.e. the target optimization variable set, is obtained by Latin hypercube sampling method. The input parameter set is substituted into the model for simulation, and the maximum water hammer pressure value, minimum negative pressure value and maximum reverse speed of the pump are output to construct the desired sample set.

[0124] Furthermore, parameters for selecting the closing pattern of the valve after the pump include: fast closing time. Quick-closing angle Slow closing time Slow closing angle The parameters of a unidirectional pressure regulating tower include: tower body diameter. Initial effective head Water supply pipe diameter .

[0125] Furthermore, the specific method of step S2 includes:

[0126] 80% of the expected sample set is randomly selected as the training set, and the remaining 20% ​​is used as the test set. The training set is used to build a water hammer protection prediction model based on CNN-LSTM, and the test set is used to test the trained prediction model. If the accuracy requirements are met, it is used as the objective function of the multi-objective optimization model. The input of the objective function is the target optimization variable set, and the output is the predicted maximum water hammer pressure value, the predicted minimum negative pressure value, and the predicted maximum reverse rotation speed of the water pump.

[0127] The CNN-LSTM model consists of the following layers: input layer, one-dimensional convolutional layer, max pooling layer, LSTM layer, and fully connected layer. The one-dimensional convolutional layer has a kernel size of 3 and 128 filters. The LSTM layer has 256 hidden units.

[0128] Furthermore, in step S3, a water hammer protection prediction model for pump shutdown is introduced as the objective function in the multi-objective optimization algorithm. The optimization objectives of the established large-scale multi-objective optimization mathematical model include four aspects: minimizing the maximum water hammer pressure value, maximizing the minimum negative pressure value, minimizing the maximum reverse rotation speed of the pump, and minimizing the water hammer protection cost; among them, the water hammer protection cost is characterized by minimizing the volume of the unidirectional pressure regulating tower.

[0129] Furthermore, under the premise of ensuring the safe operation of the water diversion project, the large-scale multi-objective optimization mathematical model needs to meet the following constraints:

[0130] (1) Valve closing constraint after pump:

[0131] ;

[0132] In the formula, To expedite the closing time; For quick closing angle; This refers to the slow shutdown time; For slow closing angle;

[0133] (2) Extreme value constraints of the pipeline system:

[0134]

[0135] In the formula, This represents the maximum water hammer pressure value. This represents the maximum steady-state pressure along the route. This is the minimum negative pressure value; This is the maximum reverse rotation speed of the water pump; This refers to the rated speed of the water pump;

[0136] (3) Input variable value range constraints:

[0137]

[0138] In the formula, The diameter of the tower body; This is the initial effective head; This refers to the diameter of the water supply pipe.

[0139] Furthermore, in the specific steps of iteratively searching for the optimal solution set of pump shutdown water hammer protection parameters using LSMDCSO as the main algorithm, the initialization and representative solution selection stages include:

[0140] Input initialization parameters to generate an initial population, and set the number of iterations to 1. The initialization parameters include population size N, maximum function evaluation number MaxFEs, scaling factor, number of representative solutions, number of samples, dimension d of optimization variables, and input variable constraints.

[0141] Selecting high-quality representative solutions from the current population based on angle penalty distance to form a representative solution set RS; specifically, this includes: uniformly generating reference vectors in the target space, assigning each individual to the reference vector with the smallest angle, calculating the angle penalty distance (APD) of each individual, and selecting the individual with the smallest APD in each sub-region as the representative solution.

[0142] Furthermore, in the specific steps of the iterative search, the determination of the optimization stage and the multi-directional fuzzy sampling strategy includes:

[0143] Calculate the ratio of the current function evaluation count FEs to the maximum evaluation count MaxFEs. If the ratio is less than 0.5, execute the multi-directional fuzzy sampling strategy; otherwise, directly execute the multi-source learning competitive group optimization strategy.

[0144] The multi-directional fuzzy sampling strategy includes: generating a random number rand; if rand is less than 0.5, performing convergence direction sampling; otherwise, performing diversity direction sampling.

[0145] The convergence direction sampling is as follows: For the representative solution, the approximate gradient is calculated using a simultaneous perturbation gradient estimator. Combined with the boundary sampling direction, a convergence sampling direction set is constructed. The sampling quantity of the two directions is controlled by a scaling factor, and a sampled solution is generated along the convergence direction.

[0146] Diversity directional sampling is as follows: the representative solution set RS is randomly divided into two equally sized subsets RS1 and RS2, and for each pair of representative solutions ( , Construct sampling directions and orthogonal directions to form a diverse set of sampling directions, and generate sampling solutions along these diverse directions;

[0147] The fuzzy variable operator is applied to the generated sampled solution to dynamically adjust the fuzziness of the solution, so as to achieve a smooth transition from coarse-grained search to fine-grained search and output the sampled population.

[0148] Furthermore, in the specific steps of the iterative search, the multi-source learning competitive group optimization strategy includes:

[0149] The fitness value of each individual in the population is calculated based on the displacement density estimation.

[0150] Generate a random number rand. If rand is less than 0.5, use a global sorting mechanism for pairing; otherwise, use a random pairing mechanism.

[0151] The global sorting mechanism is as follows: sort in descending order of fitness value, the first 50% is the winner set WS, the last 50% is the loser set LS, and randomly pair from WS and LS to form N / 2 pairing groups;

[0152] The random pairing mechanism is as follows: the population is randomly divided into N / 2 pairs, and the fitness of the two individuals in each pair is compared. The one with higher fitness is the winner, and the one with lower fitness is the loser.

[0153] From each pairing group, randomly select one individual to form the learning set LE, with a size of N / 2;

[0154] For each loser, the position is updated using the competitive group optimizer update formula; for each individual in the learning set, a leader is randomly selected from the representative solution set, and its position is updated using the competitive group optimizer formula.

[0155] The updated population is merged with the current population, environmental selection is performed, and offspring populations are generated.

[0156] Check if the termination condition is met: if the maximum number of function evaluations MaxFEs is reached, output the optimal solution set; otherwise, jump to the step of selecting a high-quality representative solution from the current population based on angle penalty distance, and increment the iteration count by 1.

[0157] Furthermore, in step S4, the specific steps for obtaining the optimal pump shutdown water hammer protection design parameters under different biases using the entropy weight-fuzzy TOPSIS method include:

[0158] Construct a decision matrix, where m is the number of schemes and n is the number of indicators, and then normalize it.

[0159] The objective weights of each evaluation index are calculated using the entropy weight method;

[0160] Construct a weighted fuzzy decision matrix and use triangular fuzzy numbers to represent index values;

[0161] Determine the fuzzy positive ideal solution and the fuzzy negative ideal solution;

[0162] Calculate the distance between each scheme and the fuzzy positive ideal solution, as well as the distance between each scheme and the fuzzy negative ideal solution, and calculate the relative proximity accordingly;

[0163] The solutions are ranked according to their relative proximity. The higher the relative proximity, the better the solution. The solution with the highest relative proximity is selected as the optimal compromise solution.

[0164] The weights are determined based on different biased optimization objectives, which are divided into three types: safety and reliability priority, cost and economy priority, and unbiased optimization.

[0165] (1) Strong adaptability to large-scale parameter space: The present invention introduces the LSMDCSO algorithm, whose multi-directional fuzzy sampling mechanism integrates gradient approximation sampling and orthogonal sampling, which can effectively handle decision variables of up to 10,000 dimensions, perfectly adapting to the needs of multi-equipment and multi-parameter joint optimization in modern water diversion projects.

[0166] (2) Better balance between convergence and diversity: This invention selects representative solutions by angle penalty distance adaptively and maintains population diversity by combining orthogonal sampling, avoiding the defect of traditional directional sampling methods that are prone to getting trapped in local optima, and making the obtained Pareto front distribution more uniform.

[0167] (3) Significantly improved search accuracy: This invention utilizes a multi-source learning competitive group optimizer to perform a refined local search on the sampled solution, which makes up for the accuracy loss of directional sampling and can more accurately approximate the true Pareto optimal solution.

[0168] (4) High prediction accuracy: This invention uses a CNN-LSTM hybrid neural network to replace the traditional random forest model, which can more deeply explore the complex nonlinear and temporal coupling relationship between water hammer protection parameters and effects.

[0169] (5) Scientific and objective decision-making process: This invention integrates entropy weight method and fuzzy TOPSIS, which avoids the arbitrariness of subjective weighting and deals with the fuzzy uncertainty in the decision-making process.

[0170] (6) High system integration: The present invention constructs a closed-loop adaptive optimization framework of “simulation-prediction-optimization-decision-verification”, which improves the robustness and engineering practicality of the system.

[0171] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A large-scale multi-target water hammer protection intelligent optimization method based on LSMDCSO and CNN-LSTM, characterized in that, Includes the following steps: S1. Establish a hydraulic transient simulation model that conforms to the actual operation of the water diversion project, and then obtain the input parameter set of the model through the Latin hypercube sampling method. Substitute it into the model for simulation and construct the desired sample set. S2. Establish a pump stoppage water hammer protection prediction model based on the CNN-LSTM algorithm, and use it as the objective function of the multi-objective optimization model; S3. Use LSMDCSO to establish a large-scale multi-objective optimization mathematical model for the optimization problem of pump shutdown water hammer protection measures, so as to obtain the optimal solution set of pump shutdown water hammer protection design parameters. S4. Apply the entropy weight-fuzzy TOPSIS method to obtain the optimal pump stop water hammer protection design parameters under different biases; S5. Substitute the theoretically optimal pump shutdown water hammer protection design parameters into the hydraulic transient simulation model to determine the optimization effect; judge whether the optimization effect meets the expected effect, and perform secondary optimization if necessary.

2. The method according to claim 1, characterized in that, The specific method of step S1 includes: Based on the topology diagram of the water diversion project, a hydraulic transient simulation model is established using the method of characteristics. The boundary conditions include: upstream boundary, downstream boundary, downstream valve boundary, air valve boundary, unidirectional pressure regulating tower boundary, and pump station boundary. Collect and organize monitoring data on flow, pressure, water level, and operating status of pumps, downstream valves, and unidirectional pressure regulating towers along the pipeline. Select two sets of monitoring data containing 10 minutes of continuous monitoring data with a step size of 0.01 seconds under actual operating conditions. Use these data to calibrate and verify the model. Control the verification error within 10% to obtain a hydraulic transient simulation model that conforms to the actual operation of the water diversion project. The parameters of the pump downstream valve closing behavior and the parameters of the unidirectional pressure regulating tower are selected as the target optimization variables. Then, the input parameter set of the hydraulic transient simulation model, i.e. the target optimization variable set, is obtained by Latin hypercube sampling method. The input parameter set is substituted into the model for simulation, and the maximum water hammer pressure value, minimum negative pressure value and maximum reverse speed of the pump are output to construct the desired sample set.

3. The method according to claim 2, characterized in that, The parameters for selecting the valve closing pattern after the pump include: fast closing time, fast closing angle, slow closing time, and slow closing angle; the parameters for the unidirectional pressure regulating tower include: tower body diameter, initial effective head, and water supply pipe diameter.

4. The method according to claim 1, characterized in that, The specific methods for step S2 include: 80% of the expected sample set is randomly selected as the training set, and the remaining 20% ​​is used as the test set. The training set is used to construct a water hammer protection prediction model based on CNN-LSTM, and the test set is used to test the trained prediction model. If the accuracy requirement is met, it is used as the objective function of the multi-objective optimization model. The input of the objective function is the target optimization variable set, and the output is the predicted maximum water hammer pressure value, the predicted minimum negative pressure value, and the predicted maximum reverse rotation speed of the water pump. The CNN-LSTM model consists of the following layers in sequence: an input layer, a one-dimensional convolutional layer, a max pooling layer, an LSTM layer, and a fully connected layer; the one-dimensional convolutional layer has a kernel size of 3 and a filter count of 128; the LSTM layer has 256 hidden units.

5. The method according to claim 1, characterized in that, In step S3, the water hammer protection prediction model for pump shutdown is introduced as the objective function in the multi-objective optimization algorithm. The optimization objectives of the established large-scale multi-objective optimization mathematical model include four aspects: minimizing the maximum water hammer pressure value, maximizing the minimum negative pressure value, minimizing the maximum reverse rotation speed of the water pump, and minimizing the water hammer protection cost. Among them, the water hammer protection cost is characterized by minimizing the volume of the unidirectional pressure regulating tower.

6. The method according to claim 5, characterized in that, Under the premise of ensuring the safe operation of the water diversion project, the large-scale multi-objective optimization mathematical model needs to meet the following constraints: (1) Valve closing constraint after pump: ； In the formula, To expedite the closing time; For quick closing angle; This refers to the slow shutdown time; For slow closing angle; (2) Extreme value constraints of the pipeline system: ； In the formula, This represents the maximum water hammer pressure value. This represents the maximum steady-state pressure along the line. This is the minimum negative pressure value; This is the maximum reverse rotation speed of the water pump; This refers to the rated speed of the water pump; (3) Input variable value range constraints: ； In the formula, The diameter of the tower body; This is the initial effective head; This refers to the diameter of the water supply pipe.

7. The method according to claim 5, characterized in that, The initialization and representative solution selection stages in the specific steps of iteratively searching for the optimal set of pump shutdown water hammer protection design parameters using LSMDCSO as the main algorithm include: Input initialization parameters to generate an initial population, and set the number of iterations to 1; the initialization parameters include population size N, maximum function evaluation number MaxFEs, scaling factor, number of representative solutions, number of samples, dimension d of optimization variables, and input variable constraints; Selecting high-quality representative solutions from the current population based on angle penalty distance to form a representative solution set RS; specifically, this includes: uniformly generating reference vectors in the target space, assigning each individual to the reference vector with the smallest angle, calculating the angle penalty distance (APD) of each individual, and selecting the individual with the smallest APD in each sub-region as the representative solution.

8. The method according to claim 7, characterized in that, In the specific steps of iterative search, determining the optimization stage and the multi-directional fuzzy sampling strategy includes: Calculate the ratio of the current function evaluation count FEs to the maximum evaluation count MaxFEs. If the ratio is less than 0.5, execute the multi-directional fuzzy sampling strategy; otherwise, directly execute the multi-source learning competitive group optimization strategy. The multi-directional fuzzy sampling strategy includes: generating a random number rand; if rand is less than 0.5, performing convergence direction sampling; otherwise, performing diversity direction sampling. The convergence direction sampling is as follows: for a representative solution, the approximate gradient is calculated using a simultaneous perturbation gradient estimator, and a convergence sampling direction set is constructed by combining the boundary sampling direction. The sampling quantity of the two directions is controlled by a scaling factor, and a sampled solution is generated along the convergence direction. The diversity direction sampling is as follows: the representative solution set RS is randomly divided into two equally sized subsets RS1 and RS2, and for each pair of representative solutions... Construct sampling directions and orthogonal directions to form a diverse set of sampling directions, and generate sampling solutions along the diverse directions; The fuzzy variable operator is applied to the generated sampled solution to dynamically adjust the fuzziness of the solution, so as to achieve a smooth transition from coarse-grained search to fine-grained search and output the sampled population.

9. The method according to claim 8, characterized in that, The specific steps of the iterative search, including the multi-source learning competitive group optimization strategy, include: The fitness value of each individual in the population is calculated based on the displacement density estimation. Generate a random number rand. If rand is less than 0.5, use a global sorting mechanism for pairing; otherwise, use a random pairing mechanism. The global sorting mechanism is as follows: sort by fitness value in descending order, the first 50% is the winner set WS, the last 50% is the loser set LS, and randomly pair from WS and LS to form N / 2 pairing groups; The random pairing mechanism is as follows: the population is randomly divided into N / 2 pairs of individuals, and the fitness of the two individuals in each pair is compared. The one with higher fitness is the winner, and the one with lower fitness is the loser. From each pairing group, randomly select one individual to form the learning set LE, with a size of N / 2; For each loser, the position is updated using the competitive group optimizer update formula; for each individual in the learning set, a leader is randomly selected from the representative solution set, and its position is updated using the competitive group optimizer formula. The updated population is merged with the current population, environmental selection is performed, and offspring populations are generated. Check if the termination condition is met: if the maximum number of function evaluations MaxFEs is reached, output the optimal solution set; otherwise, jump to the step of selecting a high-quality representative solution from the current population based on angle penalty distance, and increment the iteration count by 1.

10. The method according to claim 1, characterized in that, In step S4, the specific steps for obtaining the optimal pump shutdown water hammer protection design parameters under different biases using the entropy weight-fuzzy TOPSIS method include: Construct a decision matrix, where m is the number of schemes and n is the number of indicators, and then normalize it. The objective weights of each evaluation index are calculated using the entropy weight method; Construct a weighted fuzzy decision matrix and use triangular fuzzy numbers to represent index values; Determine the fuzzy positive ideal solution and the fuzzy negative ideal solution; Calculate the distance between each scheme and the fuzzy positive ideal solution, as well as the distance between each scheme and the fuzzy negative ideal solution, and calculate the relative proximity accordingly; The solutions are ranked according to their relative proximity. The higher the relative proximity, the better the solution. The solution with the highest relative proximity is selected as the optimal compromise solution. The weights are determined based on different biased optimization objectives, which are divided into three types: safety and reliability priority, cost and economy priority, and unbiased optimization.

Images