Urban rainwater overflow modeling method and device based on sensitivity analysis and server

Abstract

The application provides a city rainwater overflow modeling method and device based on sensitivity analysis and a server, relates to the technical field of city water management, and comprises the following steps: simulating input data by using a preset simulation model, and constructing an integrated model, wherein the integrated model comprises a drainage pipe network system model, a sewage treatment plant model in a drainage system and a river model; performing sensitive parameter screening processing on multi-dimensional model parameters in the integrated model by using a regional sensitivity analysis model to obtain key sensitive parameters; and performing parameter optimization processing on the key sensitive parameters based on a two-stage adaptive genetic algorithm to obtain target sensitive parameters and a target integrated model corresponding to the target sensitive parameters. The application can significantly reduce equipment operation cost while guaranteeing control requirements.

Description

Technical Field

[0001] This invention relates to the technical field of urban water management, and in particular to a method, apparatus and server for modeling urban stormwater overflow based on sensitivity analysis. Background Technology

[0002] Currently, related technologies suggest that existing urban stormwater overflow modeling methods typically focus on a single subsystem, such as optimizing only the overflow of drainage pipe networks, rather than an integrated model. If an integrated model of multiple systems is chosen, the lack of coordination between models and poor adaptability to complex operating conditions often leads to longer decision-making time, lower response efficiency, and thus significantly increases the operating cost of the model and the energy consumption of model decision-making calculations. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method, device and server for modeling urban stormwater overflow based on sensitivity analysis, which can significantly reduce equipment operating costs while ensuring control requirements.

[0004] In a first aspect, embodiments of the present invention provide a method for modeling urban stormwater overflow based on sensitivity analysis. The method includes: using a preset simulation model to simulate and process input data to construct an integrated model, wherein the integrated model includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model; using a regional sensitivity analysis model to screen the multidimensional model parameters in the integrated model for sensitive parameters to obtain key sensitive parameters; and using a two-stage adaptive genetic algorithm to optimize the key sensitive parameters to obtain target sensitive parameters and a target integrated model corresponding to the target sensitive parameters.

[0005] In one embodiment, the preset simulation model includes: a first preset simulation model. The step of using the preset simulation model to simulate and process input data to construct an integrated model includes: using the first preset simulation model, generating a drainage network system model using input pipe topology information, pipe diameter information, and slope information, and constructing a sewage treatment plant model within the drainage system based on input pollutant concentration information and reagent dosing information.

[0006] In one embodiment, the preset simulation model further includes a second preset simulation model. The step of using the preset simulation model to simulate and process the input data to construct an integrated model further includes: using the second preset simulation model to simulate the input water quality information and flow information based on the preset cascade water tank method to obtain a river model.

[0007] In one implementation, the step of screening multidimensional model parameters in an integrated model for sensitive parameters using a regional sensitivity analysis model to obtain key sensitive parameters includes: extracting sampling parameters from the multidimensional model parameters using Latin hypercube sampling, and performing a three-level physical constraint test on the sampling parameters to remove model parameters that fail the test, thereby obtaining target model parameters. The three-level physical constraint test includes: single-parameter test, multi-parameter synergy test, and model stability test; and determining the parameter dimension corresponding to the target model parameters as key sensitive parameters.

[0008] In one implementation, the two-stage adaptive genetic algorithm includes an exploration phase and a convergence phase. Based on the two-stage adaptive genetic algorithm, the steps of optimizing key sensitive parameters to obtain target sensitive parameters and the target ensemble model corresponding to the target sensitive parameters include: in the exploration phase, by increasing the crossover rate and mutation rate of the genetic algorithm, performing a wide-area search to obtain an initial high-quality solution set that covers the sensitive parameter space and satisfies physical constraints; in the convergence phase, performing a population reconstruction process centered on the initial high-quality solution set to obtain a reconstructed population, and using the reconstructed population as a basis, performing a fine search process on the initial high-quality solution set to obtain the target sensitive parameters and the target ensemble model.

[0009] In one implementation, the steps of obtaining an initial high-quality solution set that covers the sensitive parameter space and satisfies physical constraints by increasing the crossover rate and mutation rate of the genetic algorithm and performing wide-area search processing include: initializing the population by evenly distributing the key sensitive parameters in the sensitive parameter space to obtain initial individuals, and calculating the fitness of the initial individuals based on the weighted relative error corresponding to each initial individual to obtain a fitness ranking; extracting a first target individual based on the fitness ranking, and performing wide-area search processing on the first target individual based on a preset high crossover rate and high mutation rate to obtain an initial high-quality solution set.

[0010] In one implementation, the steps of performing a fine-search process on the initial high-quality solution set based on the reconstructed population to obtain target sensitivity parameters and target ensemble model include: calculating the fitness of the reconstructed population using an error gradient weighted fitness function to obtain a gradient weighted fitness ranking, and extracting a second target individual based on the gradient weighted fitness ranking; and performing a fine-search process on the second target individual based on a preset low crossover rate and low mutation rate to obtain target sensitivity parameters and target ensemble model.

[0011] Secondly, embodiments of the present invention also provide an urban stormwater overflow modeling device based on sensitivity analysis. The device includes: a data simulation module, which uses a preset simulation model to simulate and process input data to construct an integrated model, wherein the integrated model includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model; a sensitivity analysis module, which uses a regional sensitivity analysis model to screen and process the multidimensional model parameters in the integrated model to obtain key sensitive parameters; and a parameter optimization module, which uses a two-stage adaptive genetic algorithm to optimize the key sensitive parameters to obtain target sensitive parameters and the target integrated model corresponding to the target sensitive parameters.

[0012] Thirdly, embodiments of the present invention also provide a server, including a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement any of the methods provided in the first aspect.

[0013] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement any of the methods provided in the first aspect.

[0014] The embodiments of the present invention bring the following beneficial effects: This invention provides a method, device, and server for modeling urban stormwater overflow based on sensitivity analysis. The method uses a preset simulation model to process input data and construct an integrated model. Then, through a regional sensitivity analysis model, it screens the multidimensional model parameters in the integrated model for sensitive parameters to obtain key sensitive parameters. Finally, based on a two-stage adaptive genetic algorithm, it optimizes the key sensitive parameters to obtain target sensitive parameters and the corresponding target integrated model. This invention can significantly reduce equipment operating costs while ensuring control requirements are met.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a method for modeling urban stormwater overflow based on sensitivity analysis, provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the specific process of a method for modeling urban stormwater overflow based on sensitivity analysis, provided in an embodiment of the present invention; Figure 3 A comparison diagram of convergence curves for an optimization process provided in an embodiment of the present invention; Figure 4 A schematic diagram of a rainfall process line provided in an embodiment of the present invention; Figure 5 A schematic diagram of a key node water level provided in an embodiment of the present invention; Figure 6 A schematic diagram of a modeling device for urban stormwater overflow based on sensitivity analysis provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a server provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. 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.

[0020] Currently, overflow pollution from urban drainage systems does not occur randomly; its occurrence is directly related to multiple factors, including drainage system defects, extreme weather, and system operation and maintenance. The consequences of overflow spread gradually along a chain affecting water quality, ecology, people's livelihoods, and urban operations, causing multi-dimensional impacts on sustainable urban development. Urban drainage systems experience a series of problems during rainy season overflows, including but not limited to the following: 1. Water quality deterioration: In the initial stages of rainfall, raindrops absorb pollutants from the air. After landing, surface runoff carries in pollutants such as mud, oil, and household waste. These pollutants then enter the drainage network, further eroding existing pipes and significantly increasing the total amount of pollution. If these pollutants are not properly diverted, intercepted, and purified, they will directly enter urban water bodies, leading to water quality deterioration. Related studies show that in some southern cities, the pollution load on urban water bodies can increase by 30%–50% during the rainy season, and in some older urban areas with combined sewer systems, it can even exceed 60%.

[0021] 2. Structural problems in drainage systems are becoming increasingly prominent: Taking the common intercepting combined sewer system as an example, on the one hand, the design standards for combined sewer networks to cope with the rainy season flow are too low, and the actual interception multiple in many cities is less than 1.0, resulting in a large amount of mixed sewage not being effectively collected and intercepted; on the other hand, there are many problems in the construction and operation and maintenance of the sewer system, such as poor construction quality and inadequate post-construction maintenance and management, which leads to serious misconnection and mixing of rainwater and sewage pipe networks and river networks, backflow of river water, groundwater intrusion, etc. Various external water sources seriously encroach on the space of sewage pipes and weaken the pipe network's ability to transport sewage.

[0022] 3. Structural problems in the drainage system are becoming increasingly prominent: Rainwater runoff pollution and overflow pollution caused by urban drainage systems have become the root cause of urban water bodies turning black after rain, seriously affecting the sensory quality of urban water bodies and putting enormous pressure on the stable achievement of water quality standards at all levels of assessment sections.

[0023] 4. Threatened ecological environment: Urban rivers have low flow rates and slow speeds, resulting in poor water circulation and weak resistance to pollution. Sewage overflowing during the rainy season carries large amounts of pollutants, damaging the habitat of plants and animals in urban water bodies, threatening the balance of the aquatic ecosystem, and consequently affecting the entire city's ecological environment.

[0024] Based on this, the present invention provides a method, device, and server for urban stormwater overflow modeling based on sensitivity analysis. It constructs a drainage network model (NSE ≥ 0.75, MRE ≤ 15% after calibration) and a wastewater treatment plant model (effect water quality simulation NSE ≥ 0.80 after nitrogen and phosphorus removal parameter calibration) based on SWMM, and a river model (river water quality simulation NSE ≥ 0.70) based on MIKEII. Subsequently, it uses the HSY algorithm to conduct regional sensitivity analysis, generating parameter combinations through Latin hypercube sampling. Valid samples are screened through a two-layer test at the parameter level (single-parameter physical constraints and multi-parameter collaborative logic) and the model level (model operational stability). Sensitive parameters are identified by combining prior and posterior distribution differences. Finally, a phased genetic algorithm is used to finely optimize the sensitive parameters. In the exploration phase, the search range is expanded by dynamically decreasing crossover rate, mutation rate, and exponential crossover. In the convergence phase, the parameter space is narrowed near the optimal solution. Precise optimization is achieved through error gradient weighted fitness and Gaussian fine-tuning, resulting in a significant improvement in the accuracy of overflow pollution control, economic efficiency, and anti-interference capability of the drainage system. This method is applicable to sustainable urban water management scenarios.

[0025] See Figure 1 The diagram shows a flowchart of a method for modeling urban stormwater overflow based on sensitivity analysis. The method mainly includes the following steps S102 to S106: Step S102: Using a preset simulation model, the input data is simulated and processed to construct an integrated model. The integrated model includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model. The preset simulation model includes: a first preset simulation model and a second preset simulation model.

[0026] In one implementation, a drainage network system model can be generated using a first preset simulation model, utilizing input pipe topology, pipe diameter, and slope information. Based on input pollutant concentration and reagent dosage information, a wastewater treatment plant model within the drainage system can be constructed. Specifically, the drainage network system model is built based on SWMM. Input data includes: pipe topology, pipe diameter (DN300-DN2000), and slope (0.3‰~2.5‰). By calibrating sensitive parameters (including Manning coefficient n=0.013-0.017), the calibrated NSE ≥ 0.75 and MRE ≤ 15%. SWMM is simulation software used to simulate underground pipe networks and wastewater treatment plants. It simulates how rainwater and sewage flow in pipes, whether overflows occur, and when pump stations operate. Furthermore, a wastewater treatment plant model within the drainage system can be constructed based on SWMM to calibrate nitrogen and phosphorus removal related parameters (including nitrification rate 0.3%). ~0.5 The simulated effluent water quality NSE is ≥ 0.80.

[0027] In another implementation, a second preset simulation model is used to simulate the input water quality and flow information based on the preset cascade tank method to obtain a river model. That is, a river model is built within the integrated system based on MIKEII, and the cascade tank method is used to simulate water flow. The dispersion coefficient is calibrated, and the river water quality simulation NSE is ≥ 0.70. MIKEII is a simulation software used to simulate surface rivers. It is used to simulate the river water level and water quality after pipeline overflow into the river. After the above model is built, regional sensitivity analysis is performed using the HSY algorithm.

[0028] Step S104: Using the regional sensitivity analysis model, the multidimensional model parameters in the integrated model are screened for sensitivity parameters to obtain key sensitive parameters.

[0029] In one implementation, Latin hypercube sampling (LHS) can be used. The number of sampling groups (Ns) is determined according to the principle of number of parameters × 50 (e.g., 1000 groups for 20 parameters), with a minimum of 500 groups (to avoid insufficient sample size leading to distribution distortion). First, the prior distribution interval ([ai,bi]) of each parameter is divided into (Ns) sub-intervals. Then, one sample point is randomly selected in each sub-interval. Finally, the sample points of all parameters are randomly combined to form (Ns) groups of non-overlapping parameter combinations.

[0030] Step S106: Based on the two-stage adaptive genetic algorithm, the key sensitive parameters are optimized to obtain the target sensitive parameters and the target ensemble model corresponding to the target sensitive parameters. The two-stage adaptive genetic algorithm includes an exploration phase and a convergence phase.

[0031] In one implementation, the search scope can be expanded in the exploration phase, and then a computationally efficient optimization algorithm can be used in the convergence phase to optimize only a few key sensitive parameters selected in the exploration phase, so as to find the optimal parameter values ​​that minimize the model error. That is, the optimization goal of first wide-area search and then fine convergence can be achieved by dynamically adjusting the algorithm parameters (such as crossover rate and mutation rate).

[0032] The urban stormwater overflow modeling method based on sensitivity analysis provided in this invention adopts a parameter optimization technique that integrates regional sensitivity analysis and a phased genetic algorithm: First, parameter combinations are generated through Latin hypercube sampling (LHS), and effective samples are screened through a two-layer test at the parameter level (single-parameter physical constraints and multi-parameter collaborative logic) and the model level (model running stability). Then, sensitive parameters are identified based on the HSY algorithm through prior and posterior distribution differences. Subsequently, a phased genetic algorithm is used. In the exploration phase (generations 1 to 80), the search range is expanded by means of high crossover rate (0.9 to 0.7), high mutation rate (0.05 to 0.02), and exponential crossover. In the convergence phase (generations 81 to 200), the parameter space is reduced (20% of the original range) near the optimal solution. The sensitive parameters are finely optimized by error gradient weighted fitness and Gaussian fine-tuning, which solves the problem of dimensionality curse and local optima caused by traditional algorithms optimizing all parameters.

[0033] Furthermore, this invention also employs a multi-objective collaborative integrated control technology for urban drainage systems: constructing a full-chain integrated system encompassing drainage pipe networks (SWMM model), sewage treatment plants (SWMM model), and rivers (MIKEII model), clearly defining control variables such as pumping stations, gates, and dosing pumps; designing multi-objective functions for environment (total overflow and effluent water quality and river water quality), economy (energy consumption and chemical consumption), and operation (facility frequency), and forming a global reward function through weighted integration (e.g., environmental sub-objective weight 0.5, economy 0.1, operation 0.1); and setting three types of constraints for equipment capacity (range of control variable values), operational stability (5-minute control step for pipe networks / 30-minute control step for sewage treatment plants), and system state (water quantity and quality dynamic equations), thereby achieving collaborative optimization of the hydraulic, water quality, economic, and operational aspects of the drainage system, and addressing the shortcomings of existing technologies that suffer from single objectives and insufficient subsystem collaboration.

[0034] See Figure 2 The diagram illustrates a specific process for a method of modeling urban stormwater overflow based on sensitivity analysis. This invention also provides an implementation method for modeling urban stormwater overflow, as detailed in (A) to (C) below: (A) Determine the prior distribution of each parameter (usually assumed to follow a uniform distribution), draw samples from the prior distribution, and calculate the error. If the error is less than a pre-set acceptable threshold, accept the parameter group to form the posterior distribution. Then, identify sensitive parameters based on the difference between the prior and posterior distributions. Use the Nash-Satcliffe efficiency coefficient (NSE) and mean relative error (MRE) to characterize the simulation error of the model. The calculation formulas are as follows:

[0035]

[0036] Where NSE is the Nash-Satcliffe efficiency coefficient; MRE is the mean relative error; N is the number of monitoring sample points output by the module; obs i For the i-th monitoring sample value; simu i These are the simulated values ​​from the corresponding model; This represents the average value of the monitored samples.

[0037] This paper proposes a method to perform regional sensitivity analysis by combining a regional sensitivity analysis algorithm with a genetic algorithm (GA). First, the regional sensitivity analysis algorithm is used to perform a global screening to quickly identify sensitive parameters. Then, the computational resources are concentrated on the fine optimization of a few sensitive parameters. This method avoids the huge computational burden and dimensionality curse problem caused by directly using a genetic algorithm to optimize all parameters.

[0038] (B) HSY algorithm for parameter sensitivity screening and ranking: Sample parameters are extracted from multidimensional model parameters by Latin hypercube sampling, and three-level physical constraint tests are performed on the sample parameters to remove model parameters that fail the test and obtain target model parameters. Then, the parameter dimensions corresponding to the target model parameters are determined as key sensitive parameters. The three-level physical constraint tests include: single parameter test, multi-parameter synergy test and model stability test.

[0039] In one implementation, sample rationality verification is crucial and necessary after random sampling of parameters for the regional sensitivity analysis algorithm. Samples with contradictory physical meanings, model failures, or invalid parameter combinations are removed to ensure the reliability of subsequent simulation results. The verification process needs to consider both engineering physical constraints and model operation rules, and is divided into two levels: parameter-level verification and model-level verification.

[0040] The first level of testing is the physical boundary testing of the parameters. Each parameter has a clear physical meaning and must meet the physical constraints of each parameter: First, single-parameter testing: extract a single parameter value from the sample, substitute it into the corresponding test formula, and determine whether the physical constraints are met; if any parameter violates the constraint, the sample group is directly removed (no need to proceed to subsequent tests); Second, multi-parameter testing: there are cooperative constraint relationships between parameters, and the rationality of the combination needs to be judged in conjunction with the physical formula. Common combination tests and formulas are as follows: 1. Inspection of the rationality of pump operation in the pumping station: The pump's flow rate Q, head H, and power P must satisfy the pump characteristic equation. If the sample includes pump parameters (such as head and flow rate), verification is required.

[0041] Where P is the pump shaft power (kW), and P ≤ pump rated power P rated; ρ is the density of water (taken as 1000kg / m³); g is the acceleration due to gravity (taken as 9.81m / s²); Q is the pump flow rate (m³ / s) (the parameter value in the sample); H is the pump head (m) (the parameter value in the sample); η is the pump efficiency (taken as 0.7~0.85, an empirical value for engineering).

[0042] If P calculated according to the sample parameters is greater than P rated ×1.1 (allowing a 10% overload margin), this set of samples is unreasonable (the water pump cannot operate normally).

[0043] 2. Hydraulic balance verification of the drainage pipe network: The pipe diameter D - flow rate Q - slope i of the pipeline must satisfy the Manning formula, to verify whether the pipeline parameter combination in the sample meets the hydraulic balance:

[0044] Wherein, Q is the design flow rate of the pipeline (m³ / s) (the parameter value in the sample); n is the Manning coefficient (the parameter value in the sample); A is the cross-sectional flow area of the pipeline (m²), for a circular pipeline (D is the pipe diameter); R is the hydraulic radius (m), when the circular pipeline is full of flow (R=D / 4); i is the pipeline slope (‰).

[0045] Substitute n, D, and i in the sample into the formula to calculate the theoretical maximum flow rate Q max ; If the actual flow rate Q in the sample is greater than Q max ×1.2 (allowing 20% over-design flow), the pipeline cannot convey this flow rate, and the sample is unreasonable.

[0046] 3. DO balance verification for the aeration tank of the sewage treatment plant: The DO concentration, aeration rate, and oxygen transfer efficiency of the aeration tank must satisfy the oxygen balance equation, to verify the aeration parameter combination in the sample:

[0047] Wherein, OTR is the oxygen transfer rate (kgO2 / (m³ h)), OTR ≥ the microbial oxygen consumption rate OUR (e.g. 2kgO2 / (m³ h)); KLa is the overall oxygen transfer coefficient (h - ⁻¹); C S is the saturated dissolved oxygen concentration in water (mg / L, taken as 9.17mg / L at 20℃); C is the actual DO concentration in the aeration tank (mg / L) (generally 2mg / L).

[0048] If the calculated OTR < OUR × 0.8 (the oxygen transfer amount cannot meet the microbial demand), the aeration parameter combination is unreasonable, and this sample is eliminated.

[0049] 4. Finally, the model is tested to see if the sample parameters can make the model run normally and output reasonable results. Although some sample parameters meet the physical constraints, they may cause the model to diverge (such as the water level rising indefinitely) or output abnormal values ​​(such as the water concentration being negative). A simplified simulation is needed for quick testing.

[0050] Assuming the model runs a 24-hour rainfall simulation, the procedure to verify its validity is as follows: First, substitute the sample parameters into the model, input constant conditions, run for 24 hours, and observe whether the model operates normally. If the model reports an error or terminates, remove that sample group. The objects being tested include the following: Drainage network water level inspection: The water level h in the pipe must be ≤ pipe diameter D (to avoid overflow and loss of control). Inspection formula: h≤D×1.1 (10% over-pipe height allowed, corresponding to emergency overflow scenarios); Wastewater treatment plant effluent quality testing: effluent pollutant concentration is generally ≥0 and ≤ influent concentration, therefore the corresponding parameter should be 0≤C. effluent ≤C influent ×1.2; where C effluent The simulated effluent concentration (mg / L) from a wastewater treatment plant; C influent The influent concentration is (mg / L, fixed as the measured average, e.g., COD = 300 mg / L); if C effluent <0 or C effluent >C influent ×1.2, the sample is obviously unreasonable and needs to be removed.

[0051] River model flow continuity test: Inflow flow Q at the river cross section in Outflow Q out Changes in storage volume Then the water balance equation must be satisfied:

[0052] in, This represents the change in water storage volume (m³) between cross sections. For simulating step size; Q in Q out The simulated inflow / outflow rate (m³ / s); if the simulation results If the water imbalance error is greater than 10%, the sample is unreasonable.

[0053] Based on the single-parameter test, multi-parameter test, and model test mentioned in the above three schemes, a step-by-step test is conducted, and finally the remaining samples are retained.

[0054] (C) Genetic algorithm performs fine optimization of sensitive parameters: Assuming the number of iterations is 200 generations, then the exploration phase (generations 1-80) focuses on expanding the search range to avoid getting trapped in local optima; the convergence phase (generations 81-200) focuses on fine search near high-quality solutions to accelerate convergence to the optimal solution.

[0055] In one implementation, during the exploration phase, by increasing the crossover and mutation rates of the genetic algorithm, a wide-area search is performed to obtain an initial high-quality solution set that covers the sensitive parameter space and satisfies physical constraints: the population is initialized, the key sensitive parameters are evenly distributed in the sensitive parameter space to obtain initial individuals, and the fitness of the initial individuals is calculated based on the weighted relative error corresponding to each initial individual to obtain a fitness ranking. Then, the first target individual is extracted according to the fitness ranking, and a wide-area search is performed on the first target individual based on a preset high crossover and high mutation rate to obtain an initial high-quality solution set. Assuming there are 80 initial individuals, each individual is a set of values ​​for the sensitive parameters, the exploration phase specifically includes the following (1) to (5): (1) Population initialization: First, the population is initialized to generate an initial population that is evenly distributed and covers the entire sensitive parameter space, thereby improving the quality of the initial solution; the constraint interval [Li,Ui] of each sensitive parameter is divided into P sub-intervals according to the population size P; for each parameter dimension, one point is randomly sampled in the sub-interval to ensure that each sub-interval is selected only once. Then the initialization formula is: the value of the i-th parameter of the j-th individual is:

[0056] Among them, K ji ∈{1,2,...,P}, is the sub-interval index (ensuring no repetition) of the j-th individual in the i-th parameter dimension; R ji ~ U (0,1) represents the random perturbation within the subinterval; L i , U i These are the upper and lower limits of the sensitivity parameters output by the HSY algorithm.

[0057] (2) Fitness function calculation: Minimizing simulation error and minimizing parameter adjustment cost (e.g., the parameter adjustment cost of sewage treatment plant is higher than that of pipeline network) are combined into a weighted objective function: J j = w 1×MRE j + w 2×Cost j Among them, MRE j Cost is the weighted relative error of the j-th individual (as defined previously);j Adjust the cost for the parameters (dimensionless, such as setting the Manning coefficient adjustment cost to 1 and the nitrification rate to 1.5). w 1 = 0.7, w 2 = 0.3 (The weight can be adjusted according to project requirements).

[0058] The constraints include: hard constraints, NSE ≥ 0.8 (below this value the model has no practical value); and soft constraints: fitness function.

[0059] (3) Selection operation: to retain high-quality individuals while preventing a single high-quality individual from quickly dominating the population (maintaining diversity); retain the top 5% of individuals in fitness in each generation (e.g., retain 4 out of 80 individuals) and directly enter the next generation to ensure that high-quality genes are not lost; perform tournament selection on the remaining individuals (replacing roulette and reducing the monopoly of individuals with extremely high fitness): randomly select 3 individuals from the population (tournament size = 3); select the individual with the highest fitness among them to enter the offspring; repeat this process until the population size is filled (80-4=76 individuals), where the probability of the individual with fitness ranked r being selected in the tournament selection is approximately: (More evenly distributed than roulette, avoiding an excessive proportion of offspring from highly fit individuals).

[0060] (4) Crossover operation: In the exploration phase, a high crossover rate promotes gene recombination and expands the search range; Adaptive crossover rate: In the exploration phase, the crossover rate dynamically decreases with iteration (from 0.9 to 0.7):

[0061] Where g is the current generation (1≤g≤80), ensuring a high crossover rate in the early stage to explore new solutions, and gradually reducing it in the later stage to stabilize high-quality genes.

[0062] Hybrid crossover operator: Employs differentiated crossover methods for different types of parameters. Continuous parameters (such as Manning coefficient and nitrification rate): use exponential crossover (to generate offspring with a wider range of variation):

[0063]

[0064] Here, β=0.5 is the scaling factor, and G=80 is the total number of generations in the exploration phase, ensuring that crossover in the early stages generates greater mutations.

[0065] Discrete parameters (such as pump station operation mode): Single-point crossover is used, and gene fragments are exchanged by randomly selecting crossover points.

[0066] (5) Mutation operation: In the exploration phase, new genes are introduced through a high mutation rate to avoid premature population convergence; adaptive mutation rate: dynamically reduced with iteration (from 0.05 to 0.02) to ensure high mutation rate exploration of the unknown solution space in the early stage:

[0067] Non-uniform variation: For continuous parameters, the variation amplitude decreases with iteration (large variation in the early stage, small adjustment in the later stage):

[0068] Wherein, the variable asynchronous length δ(g) is:

[0069] Where r∈U(-1,1) is a random direction; b=2 is a shape parameter (controlling the step size decay rate); ensuring that the variation amplitude is large in the early stage (exploring new regions), and gradually shrinks X in the later stage.

[0070] In another implementation, during the convergence phase, population reconstruction is performed centered on the initial high-quality solution set to obtain a reconstructed population. Based on the reconstructed population, a fine search is performed on the initial high-quality solution set to obtain target sensitivity parameters and a target ensemble model. The fitness of the reconstructed population is calculated using the error gradient weighted fitness function to obtain a gradient weighted fitness ranking. Based on the gradient weighted fitness ranking, the second target individual is extracted. Then, based on a preset low crossover rate and low mutation rate, a fine search is performed on the second target individual to obtain target sensitivity parameters and a target ensemble model. For details of the convergence phase, please refer to (1) to (7) below: (1) Population reconstruction. New individuals are densely generated near the current optimal solution to focus on fine-grained search: using the optimal individuals from the exploration phase. Centered on the parameter space, resample in a reduced parameter space:

[0071] in, r ji ~ U(-1,1), the parameter space is reduced to 20% of the original range (ensuring that the search is near a high-quality solution).

[0072] (2) Fitness calculation using error gradient weighting. An error gradient term is introduced to give higher fitness to individuals that are close to the optimal solution:

[0073] Among them, WRE avg This represents the average weighted error of the current generation population. Gradient weights amplify high-quality individuals (WRE) j <WRE avgThe adaptability advantage of ).

[0074] (3) Selection operation. For the top 30% of individuals in terms of fitness, calculate the Euclidean distance between individuals (standardized distance to eliminate the influence of parameter dimensions):

[0075] If d jk If the distance is less than 0.1 (too close), individuals with lower fitness will be eliminated to ensure that the population is evenly distributed within the high-quality solution area.

[0076] (4) Employ a combination of low-probability crossover and arithmetic crossover. First, maintain a fixed low crossover rate: during the convergence phase, the crossover rate is fixed at 0.6 (lower than during the exploration phase) to reduce the disruption of high-quality solutions by gene recombination; arithmetic crossover: apply arithmetic crossover to all parameters (generating offspring with slight adjustments).

[0077]

[0078] Among them, the parent gene has a higher weight to ensure that the offspring are close to the high-quality solution.

[0079] (5) Mutation operation. Targeted fine-tuning of sensitive parameters accelerates convergence: First, Gaussian fine-tuning is performed, reducing the mutation amplitude to ±1% of the parameter value (±5% during the exploration phase):

[0080] The standard deviation is 0.01 to ensure fine-tuning.

[0081] Next, adaptive mutation of parameters is performed: for highly sensitive parameters of HSY identification (such as the Manning coefficient), the mutation probability is increased by 50% (from 0.02 to 0.03), and for insensitive parameters, it is decreased by 50% (from 0.02 to 0.01), prioritizing the optimization of key parameters.

[0082] (6) Iteration Termination and Optimal Solution Selection. The iteration terminates when the following three conditions are met simultaneously: Convergent stability: The change in fitness of the best individual over 20 consecutive generations. Fit≤1×10-4.

[0083] Accuracy achieved: The optimal individual has a WRE ≤ 5 and an NSE ≥ 0.85.

[0084] Population diversity: The average distance d between individuals within the population is ≤0.05 (the population is highly concentrated).

[0085] (7) Optimal solution verification. The final selected optimal individual undergoes three verifications: Training set validation: The WRE of the rainfall events used for optimization is ≤5%.

[0086] Independent validation set verification: WRE ≤ 6% for rainfall events not involved in optimization.

[0087] Extreme value verification: When the parameters fluctuate within ±10% of the optimal value, the WRE increase is ≤20% (solution stability).

[0088] Therefore, the aforementioned cascaded parameter high-efficiency calibration framework can overcome the computational bottleneck of multi-model parameter calibration in complex drainage systems through a cascaded strategy of HSY global sensitive parameter screening and two-stage adaptive genetic optimization. First, Latin hypercube sampling combined with three levels of physical constraint checks (single-parameter boundary, multi-parameter collaboration, and model stability) is used to quickly identify key sensitive parameters. Then, a genetic algorithm is used for phased optimization: in the exploration phase (first 80 generations), the search space is expanded with a dynamically high crossover rate (0.9 to 0.7) and a high mutation rate (0.05 to 0.02); in the convergence phase (last 120 generations), the population is reconstructed near the optimal solution, and Gaussian fine-tuning of mutation (±1% amplitude) and error gradient weighted fitness are used to accelerate optimization. Ultimately, a practical engineering solution with a 90% reduction in computational resource consumption and an NSE ≥ 0.8 is achieved.

[0089] Furthermore, the present invention conducted the following experiments: I. Experimental Setup and Comparison Standards (a) Comparison Objects 1. The method of this invention: adopts a parameter calibration framework that combines the "HSY algorithm with three-level physical constraint verification" and the "two-stage adaptive genetic algorithm (TSAGA)".

[0090] 2. Closest to existing technology (comparison method): The parameter calibration framework combining the "standard HSY algorithm (without three-level physical constraint test)" and the "standard genetic algorithm (SGA)" in the published technology is adopted. The SGA adopts a fixed crossover rate (0.8) and a fixed mutation rate (0.01).

[0091] 3. Baseline method: trial and error based on human experience, with manual parameter adjustments made based on the engineer's experience.

[0092] (II) Experimental Environment and Fairness Guarantee 1. Same hardware platform: Intel Xeon Platinum 8378A processor, 32GB RAM.

[0093] 2. Same software environment: Python 3.9, SWMM 5.2, MIKE II 2022.

[0094] 3. Same integrated model: An integrated model of the same drainage network (SWMM), sewage treatment plant (ASM2d), and river (RWQM1) was constructed for a case area in a certain urban area of ​​Hangzhou City, Zhejiang Province.

[0095] 4. Same training / validation dataset: Training set (for parameter optimization): 5 historical rainfall events in 2022 (covering single-peak, double-peak, and uniform patterns).

[0096] Validation set (used to evaluate generalization ability): 3 historical rainfall events in 2023 that were not included in the training.

[0097] 5. Same optimization objective and termination condition: The objective is to maximize the average Nash efficiency coefficient (NSE) of the ensemble model on the training set. The maximum number of iterations is set to 200 generations, or the model will terminate early when the improvement of the optimal solution is less than 0.001 for 20 consecutive generations.

[0098] II. Core Comparative Experiments and Data Presentation Experiment 1: Comparison of parameter optimization efficiency and convergence.

[0099] Objective: To demonstrate the significant advantages of the present invention, TSAGA, in terms of search efficiency and convergence speed, as shown in Table 1 below: Table 1: Performance Comparison of Parameter Optimization Process

[0100] For the corresponding figure in Table 1, please refer to... Figure 3 The graph shows a comparison of convergence curves for an optimization process. The horizontal axis represents the number of iterations (0-200 generations), and the vertical axis represents the NSE of the best individual (0.6-0.9).

[0101] Curve 1 (in this invention): It rises rapidly from 0.68 in the first iteration, reaches 0.80 in the 45th iteration and enters a plateau period, and finally converges to 0.941, with no significant fluctuations throughout.

[0102] Curve 2 (comparison method): The rise is slow, reaching 0.80 after 100 generations, and finally converges to 0.815. It shows local stabilization in the 50-70 generation (signs of getting stuck in a local optimum).

[0103] Curve 3 (manual trial and error): No obvious upward trend, fluctuating in the range of 0.75-0.78, unable to reach the target accuracy of 0.80.

[0104] The results of Experiment 1 demonstrate that the two-stage adaptive genetic algorithm (TSAGA) employed in this invention rapidly locates high-quality solution regions through high crossover / mutation rates during the exploration phase, and stably approaches the global optimum through population reconstruction and refined search during the convergence phase. Compared to standard GA, it reduces the number of optimization iterations by 51.1% and the total computation time by 53.6% while achieving higher model accuracy. This proves that the TSAGA structure brings unexpectedly high-efficiency convergence characteristics for this specific high-dimensional, nonlinear parameter optimization problem, an effect that cannot be easily obtained by those skilled in the art through simple application of standard GA.

[0105] Experiment 2: Verification of model accuracy after calibration.

[0106] Objective: To demonstrate that the model calibrated using the method of this invention has superior generalization ability (i.e., the ability to simulate unseen rainfall), as shown in Table 2 below: Table 2: Model accuracy metrics (NSE) on the calibration validation set

[0107] Data series: Model of this invention (HSY+TSAGA calibration), with specific values ​​marked (e.g., drainage network NSE=0.82, overall MRE=8.7%); Comparison method model (HSY+SGA calibration), with specific values ​​marked (e.g., drainage network NSE=0.79, overall MRE=12.3%); Uncalibrated initial model, with specific values ​​marked (e.g., drainage network NSE=0.65, overall MRE=22.5%).

[0108] Experiment 2 Effect Analysis and Conclusions: On an independent validation set, the integrated model calibrated using the method of this invention systematically and significantly outperformed the model calibrated using the comparative method in terms of simulated Nash efficiency coefficients (NSE) for all key water quality and quantity indicators. The improvement in simulation accuracy for ammonia nitrogen in wastewater treatment plant effluent and river ammonia nitrogen was particularly significant. This is directly attributed to the parameter combination selected by this invention through three levels of physical constraint checks (especially collaborative checks on aeration DO balance and pump station operational rationality), which better reflects actual physical processes and system linkages, thus endowing the model with stronger physical consistency and extrapolation prediction capabilities. This significant improvement in overall accuracy is difficult for those skilled in the art to anticipate when using the conventional HSY+SGA method.

[0109] Experiment 3: Final test of real-time control performance based on calibration model.

[0110] Objective: To transfer the "upstream" advantages of parameter calibration to "downstream" control performance, demonstrating the superiority of the whole-chain technical solution.

[0111] Testing Method: The optimal models calibrated by the method of this invention and the comparative method respectively are embedded as "environment models" into the same multi-agent reinforcement learning control framework for training. Then, the performance of the control policy is evaluated on the same test rainfall set, as shown in Table 3 below: Table 3: Comparison of System Control Performance (Average Values ​​of Test Set)

[0112] See Figure 4 The diagram shows a rainfall process curve: the horizontal axis represents time (0-6 hours), and the vertical axis represents rainfall intensity (0-150 mm / h), showing a bimodal rainstorm process (the peaks occur at 2.1 hours and 4.5 hours respectively).

[0113] See Figure 5 The diagram shows a key node water level: the two curves represent the node water level changes corresponding to the model of the present invention and the comparative method model, respectively. The water level of the model of the present invention is always lower than that of the comparative method, and there is no situation where the water level exceeds the safe level.

[0114] The pump station start-up and shutdown frequency of the model of this invention is more gradual, while the comparative method model operates frequently during the peak period of rainstorms. Furthermore, the peak COD concentration of the river corresponding to the model of this invention (200 mg / L) is lower than that of the comparative method (300 mg / L), and the peak duration is shortened by 40%.

[0115] Conclusion 3 of this experiment: The tests show that a more accurate integrated model is the foundation for high-quality real-time control. Because the model calibrated by this invention more realistically reflects system dynamics, the intelligent controller trained on it can make decisions closer to the global optimum. In 135 rainfall tests, the proposed solution comprehensively outperformed the controller based on the comparative method model in all core indicators, including overflow control, water quality protection, and operational economy. Most importantly, under extreme rainstorm scenarios, the performance advantage of the controller based on this invention further expands, with the overflow reduction rate advantage increasing from an average of 3.7 percentage points to 5.2 percentage points. This demonstrates that the model calibrated by the proposed method has stronger robustness and the ability to characterize extreme conditions. This chain-like gain effect, from better parameters to a better model and then to better control, is a highly convincing and unexpected technical effect.

[0116] In summary, this invention can significantly improve the accuracy of overflow pollution control in drainage systems and improve water quality. Existing technologies often focus on a single subsystem (such as optimizing only the overflow of drainage pipe networks), and parameter optimization suffers from the curse of dimensionality, resulting in insufficient control accuracy. This invention uses the HSY algorithm to screen sensitive parameters (such as Manning coefficient, nitrification rate, and other key parameters) and a staged genetic algorithm for fine optimization, which significantly reduces the simulation error of the integrated system (drainage pipe network, sewage treatment plant, and river).

[0117] Furthermore, this invention can reduce system operating costs and energy consumption, and improve economic and operational efficiency. Existing smart drainage systems often suffer from problems such as over-operation of pipeline pumping stations and waste of chemicals in sewage treatment plants due to insufficient coordination. This invention integrates economic sub-objectives (energy consumption and chemical consumption) and operational sub-objectives (facility frequency) through multi-objective functions to achieve optimal allocation of resources across the entire system.

[0118] This invention also enhances the system's anti-interference capability and real-time response efficiency, improving robustness and practicality. Existing technologies (such as Model Predictive Control, MPC) are sensitive to rainfall forecast errors and monitoring noise, and their decision-making is time-consuming (more than 200 seconds for a single-step optimization), making them difficult to adapt to complex operating conditions. This invention significantly improves the system's robustness and real-time performance through a staged genetic algorithm.

[0119] Regarding the urban stormwater overflow modeling method based on sensitivity analysis provided in the foregoing embodiments, this invention provides an urban stormwater overflow modeling device based on sensitivity analysis. (See attached image) Figure 6 The diagram shows a structural schematic of an urban stormwater overflow modeling device based on sensitivity analysis. The device includes the following parts: The data simulation module 602 uses a preset simulation model to simulate and process the input data to build an integrated model, which includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model. Sensitivity analysis module 604 uses a regional sensitivity analysis model to screen the multidimensional model parameters in the integrated model for sensitive parameters, thereby obtaining key sensitive parameters. The parameter optimization module 606, based on a two-stage adaptive genetic algorithm, performs parameter optimization on key sensitive parameters to obtain target sensitive parameters and the target ensemble model corresponding to the target sensitive parameters.

[0120] The urban stormwater overflow modeling device based on sensitivity analysis provided in this application constructs an integrated system model covering drainage pipe networks, sewage treatment plants, and receiving rivers. It employs the HSY algorithm for regional sensitivity analysis, generates parameter combinations through Latin hypercube sampling, and identifies sensitive parameters by screening effective samples through parameter-level and model-level dual-layer checks. Subsequently, a phased genetic algorithm is used to finely optimize the sensitive parameters. In the exploration phase, the search range is expanded by dynamically adjusting the crossover rate and mutation rate; in the convergence phase, the parameter space is narrowed near the optimal solution. Precise optimization is achieved by combining error gradient weighted fitness and Gaussian fine-tuning. This significantly improves the accuracy of overflow pollution control, reduces system operating costs and energy consumption, and enhances the system's anti-interference and real-time response capabilities, making it suitable for sustainable urban water management scenarios.

[0121] In one embodiment, the preset simulation model includes: a first preset simulation model. When performing the step of using the preset simulation model to simulate and process the input data to construct an integrated model, the data simulation module 602 is further used to: generate a drainage network system model using the first preset simulation model and the input pipeline topology information, pipe diameter information, and slope information, and construct a sewage treatment plant model within the drainage system based on the input pollutant concentration information and reagent dosing information.

[0122] In one embodiment, the preset simulation model further includes a second preset simulation model. When performing the step of using the preset simulation model to simulate and process the input data to construct an integrated model, the data simulation module 602 is also used to: simulate the input water quality information and flow information using the second preset simulation model based on the preset cascade water tank method to obtain a river model.

[0123] In one embodiment, when performing the step of screening sensitive parameters of multidimensional model parameters in the integrated model through regional sensitivity analysis model to obtain key sensitive parameters, the sensitivity analysis module 604 is further configured to: extract sampling parameters from the multidimensional model parameters through Latin hypercube sampling, and perform three-level physical constraint tests on the sampling parameters to remove model parameters that fail the tests, thereby obtaining target model parameters. The three-level physical constraint tests include: single parameter test, multi-parameter synergy test, and model stability test; and determine the parameter dimension corresponding to the target model parameters as key sensitive parameters.

[0124] In one embodiment, the two-stage adaptive genetic algorithm includes an exploration phase and a convergence phase. During the step of optimizing key sensitive parameters based on the two-stage adaptive genetic algorithm to obtain target sensitive parameters and the corresponding target ensemble model, the parameter optimization module 606 is further configured to: in the exploration phase, perform a wide-area search by increasing the crossover and mutation rates of the genetic algorithm to obtain an initial high-quality solution set covering the sensitive parameter space and satisfying physical constraints; in the convergence phase, perform population reconstruction processing centered on the initial high-quality solution set to obtain a reconstructed population, and based on the reconstructed population, perform a fine-grained search on the initial high-quality solution set to obtain the target sensitive parameters and the target ensemble model.

[0125] In one embodiment, when performing the step of increasing the crossover rate and mutation rate of the genetic algorithm to perform wide-area search processing and obtain an initial high-quality solution set that covers the sensitive parameter space and satisfies physical constraints, the parameter optimization module 606 is further configured to: perform initialization processing on the population, evenly distribute the key sensitive parameters in the sensitive parameter space to obtain initial individuals, and calculate the fitness of the initial individuals based on the weighted relative error corresponding to each initial individual to obtain a fitness ranking; extract the first target individual according to the fitness ranking, and perform wide-area search processing on the first target individual based on a preset high crossover rate and high mutation rate to obtain an initial high-quality solution set.

[0126] In one implementation, when performing the step of fine-searching the initial high-quality solution set based on the reconstructed population to obtain target sensitivity parameters and target ensemble model, the parameter optimization module 606 is further configured to: calculate the fitness of the reconstructed population using the error gradient weighted fitness function to obtain a gradient weighted fitness ranking, and extract the second target individual based on the gradient weighted fitness ranking; and perform fine-searching on the second target individual based on a preset low crossover rate and low mutation rate to obtain target sensitivity parameters and target ensemble model.

[0127] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0128] This invention provides a server, specifically, the server includes a processor and a storage device; the storage device stores a computer program, which, when run by the processor, executes the method described in any of the above embodiments.

[0129] Figure 7 This is a schematic diagram of the structure of a server provided in an embodiment of the present invention. The server 100 includes: a processor 70, a memory 71, a bus 72, and a communication interface 73. The processor 70, the communication interface 73, and the memory 71 are connected through the bus 72. The processor 70 is used to execute executable modules, such as computer programs, stored in the memory 71.

[0130] The memory 71 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 73 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0131] Bus 72 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0132] The memory 71 is used to store programs. After receiving an execution instruction, the processor 70 executes the programs. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 70 or implemented by the processor 70.

[0133] The processor 70 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 70 or by instructions in software form. The processor 70 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 71. Processor 70 reads the information in memory 71 and, in conjunction with its hardware, completes the steps of the above method.

[0134] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0135] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0136] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for modeling urban stormwater overflow based on sensitivity analysis, characterized in that, The method includes: Using a pre-set simulation model, the input data is simulated and processed to construct an integrated model, which includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model; By using a regional sensitivity analysis model, the multidimensional model parameters in the integrated model are screened for sensitivity parameters to obtain key sensitive parameters. Based on a two-stage adaptive genetic algorithm, the key sensitive parameters are optimized to obtain the target sensitive parameters and the target ensemble model corresponding to the target sensitive parameters. The step of using a regional sensitivity analysis model to screen the multidimensional model parameters in the integrated model for sensitive parameters to obtain key sensitive parameters includes: extracting sampling parameters from the multidimensional model parameters using Latin hypercube sampling, and performing a three-level physical constraint test on the sampling parameters to remove model parameters that fail the test, thereby obtaining target model parameters. The three-level physical constraint test includes: single-parameter test, multi-parameter synergy test, and model stability test; the parameter dimension corresponding to the target model parameters is determined as the key sensitive parameter. The two-stage adaptive genetic algorithm includes an exploration phase and a convergence phase. The step of optimizing the key sensitive parameters based on the two-stage adaptive genetic algorithm to obtain the target sensitive parameters and the corresponding target ensemble model includes: in the exploration phase, performing a wide-area search by increasing the crossover and mutation rates of the genetic algorithm to obtain an initial high-quality solution set covering the sensitive parameter space and satisfying physical constraints; in the convergence phase, performing a population reconstruction process centered on the initial high-quality solution set to obtain a reconstructed population, and using the reconstructed population as a basis, performing a fine-grained search process on the initial high-quality solution set to obtain the target sensitive parameters and the target ensemble model.

2. The urban stormwater overflow modeling method based on sensitivity analysis according to claim 1, characterized in that, The preset simulation model includes: a first preset simulation model; the step of using the preset simulation model to simulate and process the input data to construct an integrated model includes: Using the first preset simulation model, the drainage network system model is generated by utilizing the input pipeline topology information, pipe diameter information, and slope information. Based on the input pollutant concentration information and reagent dosing information, a sewage treatment plant model within the drainage system is constructed.

3. The urban stormwater overflow modeling method based on sensitivity analysis according to claim 1, characterized in that, The preset simulation model further includes: a second preset simulation model; the step of using the preset simulation model to simulate and process the input data to construct an integrated model further includes: The river model is obtained by simulating the input water quality and flow information using the second preset simulation model based on the preset cascade water tank method.

4. The urban stormwater overflow modeling method based on sensitivity analysis according to claim 1, characterized in that, The steps of obtaining an initial high-quality solution set that covers the sensitive parameter space and satisfies physical constraints by improving the crossover and mutation rates of the genetic algorithm and performing wide-area search processing include: The population is initialized by equally distributing the key sensitive parameters within the sensitive parameter space to obtain initial individuals. Based on the weighted relative error corresponding to each initial individual, the fitness of the initial individuals is calculated to obtain a fitness ranking. The first target individual is extracted based on the fitness ranking, and a wide-area search is performed on the first target individual based on a preset high crossover rate and high mutation rate to obtain the initial high-quality solution set.

5. The urban stormwater overflow modeling method based on sensitivity analysis according to claim 1, characterized in that, The step of performing a fine-grained search on the initial high-quality solution set based on the reconstructed population to obtain the target sensitivity parameters and the target ensemble model includes: The fitness of the reconstructed population is calculated using the error gradient weighted fitness function to obtain a gradient weighted fitness ranking, and the second target individual is extracted based on the gradient weighted fitness ranking. Based on the preset low crossover rate and low mutation rate, a fine search process is performed on the second target individual to obtain the target sensitivity parameters and the target ensemble model.

6. A modeling device for urban stormwater overflow based on sensitivity analysis, characterized in that, The device includes: The data simulation module uses a preset simulation model to simulate and process the input data to construct an integrated model, wherein the integrated model includes: a drainage pipe network system model, a sewage treatment plant model within the drainage system, and a river model; The sensitivity analysis module uses a regional sensitivity analysis model to screen the multidimensional model parameters in the integrated model for sensitive parameters, thereby obtaining key sensitive parameters. The parameter optimization module, based on a two-stage adaptive genetic algorithm, performs parameter optimization on the key sensitive parameters to obtain the target sensitive parameters and the target ensemble model corresponding to the target sensitive parameters. The step of using a regional sensitivity analysis model to screen the multidimensional model parameters in the integrated model for sensitive parameters to obtain key sensitive parameters includes: extracting sampling parameters from the multidimensional model parameters using Latin hypercube sampling, and performing a three-level physical constraint test on the sampling parameters to remove model parameters that fail the test, thereby obtaining target model parameters. The three-level physical constraint test includes: single-parameter test, multi-parameter synergy test, and model stability test; the parameter dimension corresponding to the target model parameters is determined as the key sensitive parameter. The two-stage adaptive genetic algorithm includes an exploration phase and a convergence phase. The step of optimizing the key sensitive parameters based on the two-stage adaptive genetic algorithm to obtain the target sensitive parameters and the corresponding target ensemble model includes: in the exploration phase, performing a wide-area search by increasing the crossover and mutation rates of the genetic algorithm to obtain an initial high-quality solution set covering the sensitive parameter space and satisfying physical constraints; in the convergence phase, performing a population reconstruction process centered on the initial high-quality solution set to obtain a reconstructed population, and using the reconstructed population as a basis, performing a fine-grained search process on the initial high-quality solution set to obtain the target sensitive parameters and the target ensemble model.

7. A server, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method described in any one of claims 1 to 5.

Images