Scheduling collaborative optimization method and system based on segmented agent and feasible region evaluation

Abstract

The application discloses a dispatching collaborative optimization method and system based on a segmented agent and a feasible domain evaluation. The method comprises the following steps: obtaining engineering structure and equipment dispatching parameter space and operation conditions, and constructing physical conservation constraints; predicting first safety constraint responses and second safety constraint responses by using a segmented agent model for joint parameter combinations meeting the constraints; dividing a feasible dispatching domain based on flow rate responses, extracting volume indicators of quantitative dispatching scheme proportions, and extracting spatial interaction indicators of quantitative equipment area influence distribution; solving optimal structure parameters by taking minimum outer layer structure engineering quantity as an objective and taking the above two indicators meeting performance thresholds as constraints; and outputting recommended equipment operation strategies by taking minimum inner layer flow rate as an objective in the feasible dispatching domain corresponding to the optimal structure parameters. The application can realize double-layer nested collaboration of physical structure design and equipment operation dispatching, and can improve the cross-condition dispatching adaptability and cross-region response field decoupling capability of an engineering system.

Description

Technical Field

[0001] This invention relates to the field of computer-aided engineering optimization design technology, and in particular to a scheduling collaborative optimization method and system based on segmented proxy and feasible region evaluation. Background Technology

[0002] In engineering systems involving the coordinated operation of physical structures and adjustable equipment, the physical parameters of the structures directly change the physical boundary conditions of the system, while the operation and scheduling scheme of the equipment determines the initial state and spatial distribution of the system's input load. There is a two-way coupled response relationship between the two.

[0003] For example, taking a water conservancy project's navigation infrastructure as a typical scenario, the floating dike is an important protective structure at the entrance of the approach channel. It can improve the distribution of the navigation flow field through physical barriers, ensuring that the navigation flow conditions meet requirements. The geometry of the floating dike directly alters the physical boundary conditions of the flow field, while the sluice gate scheduling scheme determines the initial momentum and spatial distribution of the incoming flow. The two-way hydraulic coupling response of these two factors directly affects the overall flow capacity and navigation indicators of the water conservancy project.

[0004] In current engineering structural optimization design and scheduling decisions, the setting of structural physical parameters and equipment operation are usually regarded as independent static engineering processes. Taking the design of a water conservancy project as an example, in the design phase, a pre-selected extreme flood condition and a fixed matching single equipment opening scheduling scheme are often chosen, with the sole objective of meeting the specifications for longitudinal and transverse flow velocities within the navigation channel. The optimal engineering structural dimensions are selected through comparison using fluid dynamics numerical simulations or physical model tests. In the scheduling and operation phase, based on the established fixed physical boundaries, and according to conventional scheduling procedures, a search is conducted for local equipment opening combinations that meet the safety performance threshold under the constraints of the flow balance equation.

[0005] The aforementioned independent static design and single-condition optimization mechanisms neglect the two-way coupling relationship between structural boundaries and equipment scheduling, resulting in poor cross-condition adaptability and strong interference from zone scheduling. This type of problem is prevalent in various engineering systems containing physical structures and adjustable equipment, such as water conservancy projects, port breakwaters, and industrial pipeline valve control systems. Summary of the Invention

[0006] Purpose of the invention: In view of the above-mentioned problems in the prior art, this application provides a scheduling collaborative optimization method and system based on segmented agent and feasible region evaluation to solve the above-mentioned technical problems.

[0007] Technical solution: Firstly, a scheduling collaborative optimization method based on segmented proxy and feasible region evaluation, comprising:

[0008] Obtain the engineering structure parameter space, equipment scheduling parameter space, and operating condition set including operating condition load parameters;

[0009] Physical conservation constraints are constructed based on load parameters.

[0010] For the joint parameter combination of structural parameters and scheduling parameters that satisfy physical conservation constraints, the first security constraint response and the second security constraint response are predicted by a segmented surrogate model.

[0011] Feasible scheduling domains are divided using safety performance thresholds, and scheduling flexibility evaluation indicators are extracted based on the feasible scheduling domains, including volume indicators for quantifying the proportion of feasible solutions and spatial interaction indicators for quantifying the regional impact distribution of each device.

[0012] In the engineering structural parameter space, with the goal of minimizing the structural engineering volume and the constraint that the volume index and the spatial interaction index meet the performance threshold, the optimal structural parameters are solved.

[0013] Within the feasible scheduling domain corresponding to the optimal structural parameters, a recommended equipment operation strategy is output with the goal of minimizing the response to the second security constraint.

[0014] The second aspect is a scheduling collaborative optimization system based on segmented proxy and feasible region evaluation. The system includes a processor, which is coupled with a memory. The memory stores instructions, and when the instructions are executed by the processor, they implement any possible implementation method in the first aspect.

[0015] Beneficial effects: By reducing global fitting error through segmented proxy model, and transforming structural design from minimizing single-point response to maximizing scheduling space capacity through feasible domain volume index and spatial interaction index, it can realize double-layer nested collaboration between physical structure design and equipment operation scheduling, and improve the cross-condition scheduling adaptability and cross-regional response field decoupling capability of engineering system. Attached Figure Description

[0016] Figure 1 This is a flowchart of the two-layer collaborative optimization method for structural scheduling in this application.

[0017] Figure 2 This is a flowchart illustrating the process of dividing a feasible scheduling domain and calculating volume indices for this application.

[0018] Figure 3 This is a flowchart of the first method for extracting spatial interaction indicators in this application.

[0019] Figure 4 This is a flowchart illustrating the second method for extracting spatial interaction indicators in this application.

[0020] Figure 5 This is a flowchart of the pre-built segmented proxy model for this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] In the technical solution of this invention, the engineering structure parameter space corresponds to the set of geometric parameters of the physical structure in the engineering system, the equipment scheduling parameter space corresponds to the set of control parameters of the adjustable operating equipment, the working condition load parameters correspond to the external input conditions in the system operation, and the physical conservation constraints correspond to the conservation law equation constraints that need to be satisfied in the system operation. The segmented proxy model can specifically be a performance response proxy model or a flow velocity field proxy model. The performance response proxy model corresponds to a mathematical regression model pre-trained based on numerical simulation data. The first safety constraint response and the second safety constraint response correspond to the performance response outputs that characterize different safety criteria, and the safety performance threshold corresponds to the performance upper limit benchmark required by the engineering specifications.

[0023] In a specific embodiment, the coordinated optimization of floating breakwater structural parameters and spillway gate scheduling parameters in a water conservancy project is used as an example for illustration. Specifically, the engineering structural parameters correspond to the floating breakwater geometric parameters, the equipment scheduling parameters correspond to the gate scheduling parameters, the operating load parameters correspond to the inflow rate, the physical conservation constraints correspond to the flow balance constraints, the first safety constraint response corresponds to the navigation velocity response, the second safety constraint response corresponds to the outflow velocity response, and the safety performance threshold corresponds to the flow velocity safety threshold. However, the method framework of this invention is also applicable to other engineering optimization scenarios involving physical structures and adjustable equipment, such as port breakwater parameter optimization, joint scheduling of river dam and gate groups, and industrial pipeline valve regulation.

[0024] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Through research and analysis, the applicant discovered that conventional methods only optimize the floating embankment parameters for a single preset scheduling condition, without considering combinations of different load parameters and power station operating modes throughout the year. When the hub switches to a non-designed operating condition, the original optimal structural parameters can easily lead to a sharp contraction in the feasible space for equipment scheduling, potentially creating unsolvable dead zones.

[0026] Correspondingly, the existing system treats the flow velocity in the gate area of ​​the navigation channel as the total output response of the entire gate, without quantifying the cross-regional lateral jet interference mechanism behind the abrupt changes in the macroscopic hydrodynamic field structure. The degree of influence and cross-interference of equipment areas at different locations on the navigation flow field varies greatly, that is, there is a lack of measurement and control criteria for the physical coupling effect between regions, which makes multi-region joint commissioning decision-making complex and difficult to achieve true physical isolation.

[0027] To solve these problems, combined with Figures 1 to 5 The present invention will be specifically described through the following embodiments.

[0028] According to one aspect of this application, this embodiment proposes a two-layer collaborative optimization method for structure scheduling based on segmented proxy and feasible region evaluation, specifically including:

[0029] Step 101: Obtain the engineering structure parameter space, the equipment scheduling parameter space, and the set of operating conditions including operating condition load parameters.

[0030] In this embodiment, the geometric parameter space of the floating breakwater consists of continuous geometric variables characterizing the three-dimensional dimensions and position of the breakwater. Specifically, the geometric parameter vector of the breakwater is defined as x, and its components include the breakwater length L, draft D, width W, and distance S from the centerline of the navigation channel. The upper and lower boundaries of each parameter can be predetermined according to the feasibility requirements of the hydraulic engineering project.

[0031] Accordingly, the gate scheduling parameter space consists of variables characterizing the opening status of each section of the spillway. The spillway is divided into multiple control zones, and the corresponding gate scheduling vector *u* includes the number of open gates and the uniform opening ratio for each zone. The number of open gates is an integer variable, and the uniform opening ratio is a continuous variable. The operating condition set reflects the various combinations of natural inflows and power station scheduling during the actual operation of the water conservancy project. Each operating condition corresponds to a defined total inflow, i.e., the operating condition load parameter, combined with the power station's power generation flow.

[0032] Furthermore, the hydraulic performance requirements of floating dikes vary under different inflow rates. Extending the fixed single operating condition to a set of operating conditions including multiple sets of flow data allows the design to cover various typical scenarios of actual operation of the hydraulic project, avoiding local verification failures caused by designing for preset operating conditions.

[0033] Step 102: Construct physical conservation constraints based on the load parameters.

[0034] In the actual operation of water conservancy projects, the physical law of water mass conservation must be satisfied. Specifically, a flow balance constraint is constructed based on the inflow rate. That is, under the current operating conditions, after deducting the power station's power generation flow from the total inflow, the remaining water volume must be discharged entirely by the spillway gates. By establishing this flow balance equation, boundary constraints can be imposed on the gate scheduling parameter space.

[0035] Step 103: For the joint parameter combination of structural parameters and scheduling parameters that satisfy the physical conservation constraints, predict the first security constraint response and the second security constraint response through a segmented proxy model.

[0036] In other words, for the combined parameter combinations that satisfy physical conservation in the engineering structure parameter space and equipment scheduling parameter space, the first safety constraint response and the second safety constraint response are predicted using a pre-built performance response proxy model or a flow field proxy model.

[0037] For example, the joint parameter combination is a high-dimensional input vector formed by splicing the geometric dimensions of the floating dike entity and the gate operation scheduling dimensions. The velocity field proxy model is a mathematical regression model pre-trained using computational fluid dynamics numerical simulation data. The navigation velocity response is used to characterize the maximum longitudinal and transverse velocities within the pilot channel gate area that determine the safety of ship navigation. The outlet velocity response characterizes the maximum velocity at the end of the stilling basin, which is directly related to the scour prevention safety of the downstream riverbed.

[0038] Correspondingly, by utilizing a pre-built velocity field proxy model, the time-consuming physical simulation is transformed into an efficient mathematical mapping operation. Based on the input joint parameter combination, the corresponding key hydrodynamic velocity indicators can be output quickly.

[0039] Optionally, the velocity field proxy model can employ a piecewise polynomial response surface or other fitting algorithms, such as Gaussian process regression, which can simultaneously output the velocity prediction value and its prediction uncertainty based on a Bayesian framework.

[0040] Step 104: Divide the feasible scheduling domain using the safety performance threshold, and extract scheduling flexibility evaluation indicators based on the feasible scheduling domain, including volume indicators for quantifying the proportion of scheduling schemes and spatial interaction indicators for quantifying the regional impact distribution of each device.

[0041] In other words, the feasible scheduling domain is divided based on the first security constraint response, the second security constraint response, and the preset security performance threshold, and the scheduling flexibility evaluation index is extracted based on the feasible scheduling domain. The scheduling flexibility evaluation index includes a volume index for quantifying the proportion of scheduling schemes and a spatial interaction index for quantifying the distribution of equipment regional influence.

[0042] In this embodiment, a preset flow velocity safety threshold is used as the upper limit benchmark for flow velocity required by navigation regulations and hydraulic structure design specifications. The obtained flow velocity response is compared with the flow velocity safety threshold to determine whether the current combination of parameters meets the engineering navigation and energy dissipation requirements under preset operating conditions.

[0043] The feasible scheduling domain is the set of all gate scheduling schemes that, given the floating dike geometry, ensure the flow velocity response does not exceed the corresponding threshold and satisfies the flow balance constraint. Based on this, the extracted volume index reflects the mathematical proportion of feasible scheduling schemes in the gate scheduling parameter space; its value corresponds to the available margin for gate scheduling during operation. The spatial interaction index represents the uniformity or decoupling degree of the influence of each gate opening state on the overall flow field.

[0044] Furthermore, by extracting volume indicators and spatial interaction indicators in parallel, this embodiment transforms the design of the floating dike from the absolute minimization of flow velocity values ​​to the maximization of scheduling space capacity and structural rationalization, in order to quantify the cross-domain enhancement effect of the floating dike as a physical water flow barrier on the flexibility of electromechanical scheduling.

[0045] Step 105: In the engineering structural parameter space, with the goal of minimizing the structural engineering quantity and the constraint that the volume index and spatial interaction index meet the performance threshold, the optimal structural parameters are solved. That is, with the goal of minimizing the structural engineering quantity in the engineering structural parameter space and the constraint that the volume index and spatial interaction index meet the preset performance threshold, the optimal structural parameters are obtained in the engineering structural parameter space.

[0046] This embodiment describes the outer-layer decision-making process of a two-layer nested collaborative optimization framework. The preset performance thresholds can be determined based on the navigation design specifications of the water conservancy project, the engineering requirements for gate scheduling flexibility, and the safety redundancy requirements under various operating conditions. For example, the performance threshold for the volume index can be set to ensure that the proportion of feasible scheduling schemes in the set of effective candidate scheduling schemes is not less than a preset minimum proportion; the performance threshold for the spatial interaction index can be set to ensure that the interval decoupling degree or sensitivity uniformity is not less than a preset minimum standard value. The quantity of the floating dike can be expressed as the product of the geometric feature dimensions of the floating dike, reflecting the actual engineering cost and construction expenses. Accordingly, the core logic of the outer-layer optimization lies in finding the solution of continuous geometric variables that minimizes the quantity of the floating dike in the multi-dimensional space, provided that the volume index and spatial interaction index are higher than the preset performance thresholds; this solution represents the optimal floating dike parameters.

[0047] Specifically, the formula for calculating the objective function of minimizing the amount of floating breakwater work can be expressed as:

[0048] C _cost =ρ×L×W×D;

[0049] Among them, C _cost Let ρ be the target value of the floating breakwater project quantity to be optimized, ρ be the unit volume cost coefficient of the floating breakwater, L be the length of the floating breakwater, W be the width of the floating breakwater, and D be the draft of the floating breakwater.

[0050] Furthermore, to optimize the objective function, gradient-free global optimization algorithms such as differential evolution or particle swarm optimization can be employed. For example, the algorithm initializes a candidate population of a preset size. During iterative evaluation, if the evaluation index corresponding to a certain set of levee parameters fails to reach a preset performance threshold, a numerical penalty is applied to its fitness function, driving the population to converge towards the feasible solution region that satisfies scheduling flexibility requirements and minimizes engineering costs. If the preset performance threshold is set too high, causing the algorithm to determine that there is no feasible solution for all operating conditions after complete traversal, a fault-tolerant alarm mechanism will be triggered, indicating that adjusting the levee's geometric parameters alone has reached the physical extreme value, and that other hydraulic structure modification measures are needed to jointly satisfy the constraints.

[0051] Step 106: Within the feasible scheduling domain corresponding to the optimal structural parameters, with the goal of minimizing the response to the second security constraint, output the recommended equipment operation strategy.

[0052] After completing the global optimization of the outer floating embankment structural parameters, this embodiment constitutes the inner decision-making process of a two-layer nested collaborative optimization framework. Specifically, for the determined optimal floating embankment parameters, the local outflow velocity performance of each scheme is evaluated in the corresponding set of feasible scheduling schemes. By traversing and comparing all schemes in the set, the gate opening combination that minimizes the global outflow velocity response is selected, and the output is the recommended gate scheduling strategy.

[0053] Accordingly, this inner layer optimization, while ensuring navigation safety and overall scheduling flexibility, selects specific execution instructions with the highest scour prevention safety redundancy for the water conservancy hub operation personnel.

[0054] For example, this application also proposes a feasible scheduling domain volume calculation method based on equality dimensionality reduction and smoothness margin weighting, used to calculate volume indicators. Taking a spillway gate as the implementation object, it specifically includes:

[0055] Step 201, when dividing the feasible scheduling domain and calculating the volume index, includes:

[0056] In the equipment scheduling parameter space, based on the total number of gate openings in each equipment area, discrete combinations of opening numbers limited by the total number of gate openings are extracted. That is, discrete combinations of opening numbers limited by the total number of gate openings are extracted from the gate scheduling parameter space.

[0057] Specifically, the gate scheduling parameter space includes integer variables and continuous variables. In actual engineering, the number of physical orifices of the spillway is fixed, and the possible combinations of orifices that can be opened in each section constitute a finite discrete set. A full traversal of this set is performed to extract all possible discrete combinations of orifice numbers. This discrete traversal restricts the calculation range of the volume integral to within the permissible limits of the equipment's physical hardware.

[0058] Step 202: Under the assumption of uniform aperture, based on physical conservation, analytically solve for the continuous aperture ratio corresponding to each discrete aperture number combination.

[0059] Specifically, for the extracted preset discrete orifice number combinations, there exist countless numerical arrangements of continuous opening ratios to satisfy the flow equation. To avoid the curse of infinite-dimensional integral calculations, this embodiment introduces a preset uniform opening assumption, forcibly setting the continuous opening ratio of each section of the spillway to the same value. Based on this, the theoretical discharge flow is calculated using the pressurized orifice flow formula combined with the submergence coefficient correction, and a single-variable algebraic equation is constructed using flow balance constraints. Through this algebraic equation, the uniquely matching continuous opening ratio value of the discrete orifice number combination is analytically solved. This master-slave variable dimensionality reduction and deterministic balancing mechanism, by eliminating the degrees of freedom of continuous variables using physical equations, can achieve effective collapse of the high-dimensional scheduling space.

[0060] Step 203: Select combinations whose continuous opening ratios satisfy the preset opening boundary conditions to form a set of effective candidate scheduling schemes.

[0061] The process involves physically validating the continuous opening ratios obtained from the analytical inverse kinematics solution. The preset boundary condition for the opening ratio is a continuous interval greater than 0 and less than or equal to 1. If the continuous opening ratio obtained from the inverse kinematics solution of a set of discrete opening combinations exceeds this boundary condition, the configuration is deemed to have physically failed in the current discharge demand and is directly eliminated. The retained combination data sequence constitutes the set of valid candidate scheduling schemes. This out-of-bounds filtering mechanism ensures that all schemes input to the proxy model comply with the laws of hydrodynamic conservation.

[0062] Step 204: The set of operating conditions includes multiple operating conditions under different hydrological scenarios. The volumetric indices are calculated, including:

[0063] Based on the annual frequency distribution characteristics of each operating condition, configure the corresponding operating condition frequency weights.

[0064] To ensure that the optimal design of the floating breakwater parameters is not dependent on a single hydrological scenario, the probability distribution of each operating condition in the hydrological sequence within a complete hydrological cycle is read, and corresponding constant scalars are assigned. The normal flow condition is assigned a larger weight value, while the extreme flood condition is assigned a smaller weight value. The sum of the frequency weight vectors for all operating conditions is fixed at 1.

[0065] Step 205: Based on the first and second security constraint responses and security performance thresholds corresponding to each scheme in the effective candidate scheduling scheme set, the feasibility of each scheme is determined, and the volume index is calculated accordingly. That is, the volume index is calculated based on the results of determining the feasibility of each scheme.

[0066] For example, the parameter matrices of each scheme in the set of effective candidate scheduling schemes are input into the flow field proxy model to obtain the flow field prediction results for each scheme. The prediction results are compared and verified with the preset flow velocity safety threshold to perform a quantitative evaluation of the physical scheme feasibility, and the global volume value reflecting the operational flexibility is calculated.

[0067] In some optional implementations, this application also provides a basic implementation method based on a hard boundary indicator function for feasibility determination and volume index calculation, including:

[0068] Step 205a: Determine the feasibility of each scheme and calculate the volume index accordingly. Specifically, this includes calling the hard indicator function to perform a binary determination on whether the first safety constraint response and the second safety constraint response exceed the safety performance threshold.

[0069] Specifically, the preset hard indicator function mathematically exhibits a step shape: it outputs 1 when the predicted flow rate response is less than or equal to a preset flow rate safety threshold, and outputs 0 when the predicted flow rate response is greater than the threshold. Using this hard indicator function, each candidate scheduling scheme is divided into 0 or 1 Boolean values, and there is no transition interval in the evaluation state of the scheme.

[0070] Step 205b: Count the number of schemes that simultaneously meet the safety performance threshold, and determine the proportion of the number of schemes in the set of effective candidate scheduling schemes as the volume index.

[0071] Accordingly, the scheme entries whose hard indicator function is determined to be 1 are accumulated to obtain the total number of valid schemes that simultaneously meet the dual safety indicators of navigation and scour prevention under the preset operating conditions. Next, the total number of valid schemes is divided by the total number of cardinal items in the set of valid candidate scheduling schemes to output the local volume percentage under this single operating condition.

[0072] In another optional implementation, this application also provides a weighted approach based on a smooth margin function for feasibility assessment and volume index calculation, specifically including:

[0073] Obtain the margin sensitivity parameters that are pre-calibrated by combining the predicted standard deviation of the segmented surrogate model.

[0074] For example, the velocity field proxy model has regression residuals during the construction phase. The inherent prediction standard deviation data of this underlying model is obtained, and combined with a set safety factor, a margin sensitivity parameter is obtained through division. This parameter is used to quantitatively characterize the uncertainty level of the underlying mathematical regression surface when evaluating the hydrodynamic flow field.

[0075] In other words, extract the inherent prediction standard deviation data σ of the underlying model. _pred Combined with the set safety factor c _safe Margin sensitivity parameter = 1 / (c _safe ×σ _pred The transition band width of the smooth margin function is calibrated to a safe multiple of the surrogate model's prediction standard deviation, so that the decay of the margin score occurs within the velocity range covered by the model's prediction uncertainty.

[0076] Using a smooth margin function that includes a margin sensitivity parameter, the degree to which the first safety constraint response and the second safety constraint response deviate from the safety performance threshold are transformed into continuous margin scores.

[0077] Accordingly, the smooth margin function is used to replace the discontinuous hard boundary step decision logic. The formula for calculating the smooth margin function for a single flow field safety constraint is:

[0078] φ _margin =1 / (1+exp(τ×(v _pred -v _th )));

[0079] Where, φ _margin The output is the continuous margin score, where exp is the natural exponential function, τ is the margin sensitivity parameter, and v _pred The velocity response during navigation or the velocity response at the outlet of the pool is predicted by the velocity field surrogate model, v _th This corresponds to the preset flow rate safety threshold value.

[0080] Calculate the average of the products of the margin scores for each scheme in the set of effective candidate scheduling schemes to obtain the margin-weighted volume index.

[0081] For each candidate vector in the set of effective candidate scheduling schemes, obtain the corresponding navigation flow field margin score scalar and outflow flow field margin score scalar, and obtain the comprehensive margin feature value through dot product operation.

[0082] Accordingly, the arithmetic mean of all comprehensive margin characteristic values ​​within the global set is calculated to obtain the final scalar. This final scalar maps the scale of the safety scheme in the scheduling parameter space and also measures the distribution of safety margins of each scheduling combination far from the flow velocity safety critical boundary, or in other words, measures the distribution characteristics of the safety margins of each scheduling combination relative to the disaster-causing physical boundary.

[0083] Step 206: According to the operating condition frequency weight, the volume indicators calculated separately for each operating condition are weighted and summed to obtain a volume indicator representing the cross-scenario adaptability across all operating conditions. Alternatively, according to the operating condition frequency weight, the local volume proportions calculated separately for each operating condition are weighted and summed to obtain a volume indicator representing the cross-scenario adaptability across all operating conditions.

[0084] Specifically, the volume calculation results for all single operating conditions, including extreme floods and normal receding water, are multiplied by their corresponding matching operating condition frequency weights, and then all products are linearly summed across the entire set. Through this multi-scenario aggregation calculation strategy, the physical design task of the floating embankment geometry can be transformed into a comprehensive search process that maximizes the flexibility of electromechanical scheduling, covering the annual inflow frequency cycle of the reservoir.

[0085] As an achievable approach, this application also proposes a spatial interaction index extraction method based on cross-regional response field isolation and physical decoupling mechanism, used to extract spatial interaction indicators in the above embodiments, with each area of ​​the spillway gate as the implementation object, specifically including:

[0086] Step 301: Under the premise of satisfying physical conservation, obtain the total number of gate openings contained in each equipment area as the capacity parameter.

[0087] For preset operating conditions, the physical structure of the spillway gate determines its upper limit of discharge capacity. Specifically, physical equipment parameters are read to obtain the number of gate openings configured in each independently divided gate control area, and these are defined as the capacity parameters of the corresponding areas. This capacity parameter reflects the hydrodynamic load distribution weight that each area can bear when fully open.

[0088] Step 302: Based on the principle of evenly distributing the target discharge flow determined by physical conservation according to the proportion of capacity parameters in each equipment area, calculate the uniform distribution scheme and lock the uniform distribution scheme as the preset baseline scheduling state.

[0089] Specifically, based on pre-set allocation principles, the target discharge flow rate to be released under the current operating conditions is linearly divided according to the capacity parameter proportion of each region. For the k-th gate region, its foundation opening value is calculated using the following formula:

[0090] n _k_0 =round(N _gate_k ×n _total / N _gate );

[0091] Where, n _k_0 N represents the number of basic openings in the k-th gate region, where round is the nearest integer function. _gate_kLet n be the capacity parameter corresponding to the k-th gate region. _total The minimum total number of gates that need to be opened to meet the current total discharge demand is calculated. This minimum number of gates can be obtained by rounding up based on the target discharge under current operating conditions and the theoretical discharge capacity of a single gate when fully open. N _gate This represents the total number of gate openings across all regions.

[0092] Accordingly, after determining the basic number of orifices in each region, the uniform orifice ratio that satisfies the target discharge flow is uniformly solved using the pressurized orifice flow formula, forming a uniform distribution scheme. The parameter vector of this scheme is defined as a preset baseline scheduling state. This processing mechanism provides the same physical reference system for the marginal impact of navigation flow velocity when measuring the deviation of scheduling parameters in each region from the uniform state.

[0093] Based on this, this embodiment also provides an optional implementation method for calculating spatial interaction indicators that reflect physical isolation characteristics, specifically including:

[0094] Step 303: Introduce a continuous relaxation mechanism to the discrete number of openings variable in the equipment scheduling parameter space, and map the discrete number of openings variable to a continuous number of openings variable on a continuous real closed interval.

[0095] Specifically, the actual number of open orifices in each section of the spillway is a discrete integer variable, which cannot be directly calculated using partial derivatives in advanced mathematics. Therefore, to eliminate this mathematical constraint, a topological mapping is performed on the variables in the internal logic of the velocity field proxy model. This extends the orifice number variable, which originally belonged to the discrete integer domain, to a continuous real closed interval including the upper and lower boundaries, forming a continuous orifice number variable.

[0096] Step 304: Based on the master-slave variable separation criterion, calculate the second-order cross partial derivatives of the first safety constraint response only for each continuous opening number variable with the same dimension.

[0097] In this embodiment, the gate scheduling parameter space contains two types of variables: the number of open orifices and the opening ratio. At the hydrodynamic scale, the macroscopic spatial collision interference of the transverse jet is dominated by the physical topological distribution of the number of open orifices. The opening ratio serves only as a local, subordinate fine-tuning variable for vertical flow balancing. Mixing differential terms of variables with different dimensions would distort the dimensions of the characteristic space.

[0098] Therefore, dimensionality reduction and isolation are performed on the feature extraction dimensions. That is, the aperture ratio variable is removed from the derivative matrix, and only the continuous aperture number variable with the same dimensions corresponding to different control regions is selected to form the master control subspace. Based on the mapped velocity field surrogate model, the corresponding second-order partial differential coefficients are extracted for each master control variable. This coefficient characterizes the nonlinear control effect of the change in the opening state of a certain region on the marginal velocity of the outflow from another region.

[0099] Step 305: Construct a cross-regional interaction matrix based on the second-order cross partial derivatives, and extract the proportion of the sum of the absolute values ​​of the cross-regional interaction terms relative to the sum of the absolute values ​​of all elements in the cross-regional interaction matrix.

[0100] In this embodiment, all extracted partial differential coefficients are arranged according to the correspondence between regions to form a symmetrical square matrix, namely the cross-regional interaction matrix. In this matrix, the elements on the main diagonal reflect the secondary interaction intensity of the parameters of each region, while the elements off the main diagonal reflect the interaction interference intensity between different regions. The sum of the absolute values ​​of the off-diagonal elements and the sum of the absolute values ​​of all elements including the main diagonal are calculated separately, and the two are divided to obtain the ratio.

[0101] Step 306: Calculate the interval decoupling degree, which reflects the physical isolation effect, based on the proportion, and use the interval decoupling degree as a spatial interaction index.

[0102] Based on the obtained matrix ratios, they are further mapped to scalar indices to characterize the degree of isolation between independent regions. The calculation formula is as follows:

[0103] γ=1-(I _cross / (I _cross +I _self +ε _0 ));

[0104] Where γ is the output interval decoupling degree, I _cross I is the sum of the absolute values ​​of all off-diagonal cross-regional interaction terms in the cross-regional interaction matrix. _self ε is the sum of the absolute values ​​of all terms on the main diagonal of the cross-regional interaction matrix. _0 This is a preset minimum positive number to prevent division by zero anomalies.

[0105] In this embodiment, the cross-regional interaction matrix can also be read and automatically parsed. If the cross-interference values ​​of two preset regions in the matrix exceed the preset alarm line, an inter-regional joint adjustment constraint table will be automatically generated. This constraint table indicates that, in actual engineering operation, the physical opening of the associated region should be avoided by synchronously and significantly adjusting it, and safety control rules should be directly output for the operation of the water conservancy project.

[0106] Based on this, this embodiment also provides another optional implementation method, namely, calculating the spatial interaction index by extracting first-order features without calculating the second-order features of the matrix, specifically including:

[0107] Step 307: Obtain the preset baseline scheduling state, and use the segmented agent model to calculate the first-order sensitivity of the scheduling parameters of each device area relative to the first security constraint response.

[0108] Specifically, a uniform distribution scheme is used as the initial calculation starting point, and a unit activation increment is independently applied to each control region. The increased state vector is then re-inputted into the surrogate model to obtain the partial derivative of the flow velocity output variable with respect to a single parameter input, forming a first-order sensitivity feature vector containing multiple components.

[0109] Optionally, when calculating the feature vector, a reduced zone-level dimension is used, treating different regions as independent evaluation objects to limit the computational load. If the gates on both sides of the same zone in a water conservancy project exhibit asymmetric flow field interference characteristics, the extraction mode is switched to a gate-by-gate level, and the first-order sensitivity feature values ​​are calculated independently for all gates. This extraction operation proportionally expands the dimension of the feature matrix.

[0110] Step 308: Extract the absolute range of each first-order sensitivity and the maximum absolute sensitivity.

[0111] Iterate through all first-order sensitivity numerical components and calculate their absolute values. Accordingly, sort and compare them to obtain the maximum and minimum values ​​in the absolute value sequence, and obtain the absolute range of the entire sequence through subtraction. Retain the maximum absolute sensitivity in the sequence for use in balancing calculations.

[0112] Step 309: Use the absolute value range and the maximum absolute sensitivity to perform zero-normalization calculation to obtain the sensitivity uniformity index, and use the sensitivity uniformity index as the spatial interaction index.

[0113] Due to the redistribution effect of the physical flow field, negative characteristic components may be obtained during the calculation, which would lead to numerical failure if conventional calculation methods are used. Therefore, a correction calculation is performed using the range characteristic, and the calculation formula is as follows.

[0114] η=1-R _abs / (S _max +ε _0 );

[0115] Where η is the calculated sensitivity uniformity index, R _abs For the absolute range, S _max For maximum absolute sensitivity, ε _0 It is a preset minimum positive number.

[0116] Through corrected calculations, the output eigenvalues ​​are mapped to a closed interval between 0 and 1. When the eigenvalue approaches 1, it indicates that the opening operations of each region have a similar impact on the hydrodynamics of the waterway, and the scheduling instructions have equivalent substitution characteristics. This eigenvalue is returned to the main program as a spatial interaction index to participate in the convergence evaluation of the outer structure parameters.

[0117] For example, this application also proposes a gradient analytical transfer and optimization solution method based on the chain rule of composite functions, used to solve for the optimal structural parameters in the above embodiments, taking the geometric parameters of the floating embankment and the physical water resistance ratio as the implementation objects, specifically including:

[0118] Step 401: Based on the preset channel size parameters and the current engineering structure, the physical water resistance ratio is analyzed and constructed; the gradient descent optimization algorithm is used to perform optimization iterative calculation in the engineering structure parameter space.

[0119] Specifically, when making outer-layer decisions within a two-layer nested collaborative optimization framework, if the volume index of the feasible scheduling domain is evaluated using a smooth margin function for continuous calculation, the entire parameter search space possesses the property of being globally continuous and differentiable. Based on this property, gradient descent optimization algorithms, such as sequential quadratic programming, are employed for local or global optimization. Compared to heuristic algorithms, gradient descent optimization algorithms can utilize the first-order partial derivatives of the objective function and constraint functions to update the levee parameter vector along the steepest gradient descent direction, thereby improving convergence speed and computational accuracy.

[0120] Accordingly, during the iterative calculation process, the input first derivative matrix is ​​used to evaluate the marginal contribution of fine-tuning the current breakwater size in various directions to reduce engineering costs and increase scheduling flexibility.

[0121] Step 402: When calculating the target gradient, the chain rule of composite functions is applied. The chain rule of composite functions includes the physical water resistance ratio. The partial derivatives of the floating embankment parameters in each dimension of the engineering structure parameter space constitute the Jacobian matrix.

[0122] Specifically, the underlying velocity field proxy model employs a partitioning and splicing strategy based on the physical water resistance ratio. The boundary drift effect of the fluid state abrupt change interval has a coupling effect on the gradient direction. Therefore, when calculating the total partial derivative of the predicted velocity with respect to the floating embankment parameter vector, it is necessary to perform a composite calculation of the rate of change of the transition weights and the rate of change of the local response surface. The calculation formula of the chain rule for composite functions is as follows:

[0123] v _nav_grad =sum((w _grad ×α _grad )×v _nav_k +w _k ×v _nav_k_grad );

[0124] Among them, v _nav_grad Let w be the total partial derivative of the predicted flow velocity with respect to the floating breakwater geometry parameter vector, sum be the summation operation over all physical zones, and w be the summation operation. _grad To determine the partial derivative of the smooth transition weighting function with respect to the physical water resistance ratio, α _grad v is the Jacobian matrix of the physical water resistance ratio to the geometric parameter vector of the floating embankment._nav_k w represents the velocity prediction value of the polynomial response surface model within the k-th physical partition. _k Let v be the smooth transition weight function corresponding to the k-th physical partition. _nav_k_grad Let be the local partial derivative of the polynomial response surface model with respect to the floating embankment geometric parameter vector within the k-th physical partition.

[0125] Furthermore, at the boundary of adjacent flow regimes, the derivative of the smooth transition weight function is calculated using the following formula:

[0126] w _left_grad =-(1 / ε)×w _left ×(1-w _left );

[0127] Among them, w _left_grad ε is the partial derivative of the weighting function for the smooth transition on the left side of adjacent physical zones with respect to the physical water resistance ratio, ε is the preset transition half-width parameter, and w _left The weight function is used for the smooth transition to the left side of adjacent physical partitions.

[0128] In this embodiment, the chain rule is embedded in the optimizer, so that the gradient information obtained by the optimization algorithm when crossing the physical partition boundary has a seamless continuity. Therefore, it can avoid numerical oscillation or falling into a pseudo-convergence state at the critical threshold of water resistance ratio.

[0129] Step 403: The physical water resistance ratio is obtained by analyzing the preset channel size parameters and the engineering structure in the current optimization iteration.

[0130] Specifically, based on the ratio of the cross-sectional area of ​​the floating breakwater to the cross-sectional area of ​​the approach channel, the physical water resistance ratio under the current geometric parameters of the floating breakwater is calculated. The calculation formula is as follows:

[0131] α=(D×W) / (H _0 ×B);

[0132] Where α is the physical water resistance ratio, D is the draft included in the floating breakwater geometry, W is the width included in the floating breakwater geometry, and H... _0 B is the upstream water depth included in the preset channel size parameters, and B is the pilot channel width included in the preset channel size parameters.

[0133] Furthermore, the partial derivatives of this physical water-blocking ratio with respect to four independent dimensions in the geometric parameter space of the floating embankment are extracted to construct the Jacobian matrix. The formula for calculating this matrix is ​​as follows.

[0134] α _grad =(0,W / (H _0 ×B),D / (H _0 ×B),0);

[0135] Where, α _grad The Jacobian matrix contains four components that correspond to the partial derivatives of the physical water resistance ratio with respect to the length, draft, width, and distance from the centerline of the pilot channel.

[0136] This matrix maps minute perturbations in the draft and width of the floating breakwater into variations in the flow field's water resistance ratio; and through the chain rule, this variation is transmitted to the flow velocity output.

[0137] In some alternative implementations, if a hard-boundary indicator function logic is used when partitioning the feasible scheduling domain, the calculated volume index, relative to the levee geometry, exhibits a non-differentiable and discontinuous non-smooth step function characteristic. Therefore, a gradient-free global optimization algorithm is switched on and invoked.

[0138] Specifically, a differential evolution algorithm or a particle swarm optimization algorithm is configured to solve the outer embankment parameter optimization problem. That is, the dimensional values ​​of the embankment's geometric parameter space are obtained, multiplied by the corresponding multiplier coefficient, and the population size parameters are generated.

[0139] Accordingly, historical optimal objective function values ​​are extracted and cached in real time. For example, the improvement rate of the global optimal solution within a preset 50 generations is monitored, and when the improvement rate is detected to be lower than a preset 10, the system will take action. -4 When the time is reached, the termination criterion is triggered, the iterative search process stops, and the currently converged optimal individual is output as the optimal embankment parameters. This embodiment can solve for objectives in both continuous and non-smooth spaces by configuring two sets of parallel optimization paths.

[0140] For example, this application also proposes a method for constructing a segmented proxy model based on physical segmentation and boundary enrichment, using a flow velocity field proxy model as the implementation object, specifically including:

[0141] Step 501, pre-construction of the segmented proxy model, including:

[0142] Based on the engineering structure parameter space and equipment scheduling parameter space, we perform master-slave variable dimensionality reduction and equation balancing sampling to obtain the initial sample set.

[0143] As an example, uniformly distributed initial discrete points are generated in a multidimensional joint parameter space. To ensure that the acquired samples all conform to the laws of physical conservation, a pre-defined constrained Latin hypercube sampling algorithm is used to extract the geometric variables of the floating embankment and the discrete number of openings of the spillway as the main variables.

[0144] Accordingly, based on the flow balance equation, the continuous opening ratio is analytically solved as a variable, and physical failure combinations are eliminated. For each joint parameter combination in the initial sample set, the corresponding flow velocity response data is obtained through computational fluid dynamics numerical simulation. Based on this, the retained joint parameter combinations constitute the initial sample set. Regarding the number of dimensions in the parameter space, the amount of data contained in the initial sample is set to a preset multiple of the input dimension, used to obtain basic flow field characteristics when controlling the computational load.

[0145] Step 502: Based on the initial sample set, fit and construct an initial version of the response surface model, and generate candidate parameter combinations;

[0146] The Gaussian kernel criticality score is calculated based on the candidate parameter combinations, the preliminary prediction results of the initial sample set, and the safety performance threshold.

[0147] In other words, the initial response surface model, which is constructed based on the initial sample set, generates candidate parameter combinations in the engineering structure parameter space and equipment scheduling parameter space; and calculates the Gaussian kernel criticality score based on the candidate parameter combinations, the preliminary prediction results of the initial sample set of the initial response surface model, and the preset safety performance threshold.

[0148] In the joint parameter space, a large number of candidate parameter combinations are generated. Since the region affecting the accuracy of the feasible scheduling domain volume is concentrated in a narrow area where the predicted velocity values ​​are close to the constraint boundary, an initial response surface model fitted with the initial sample set is used to predict velocity for all candidate parameter combinations. A pre-defined kernel function is then used to evaluate the critical value of each point for boundary positioning. The criticality score calculation formula for the navigation constraint setting is as follows:

[0149] ψ _nav =exp(-(v _nav_pred -v _th ) 2 / (2×σ _band 2 ));

[0150] Where, ψ _nav For the criticality score relative to navigation constraints, exp is the natural exponential function, v _nav_pred The numerical values ​​of the navigation current velocity response predicted based on the initial response surface model, v _th σ is the upper limit of the permissible flow rate for navigation within the preset flow rate safety threshold. _band The boundary band half-width parameter is used to control the width of the enriched region.

[0151] Furthermore, for the outflow velocity response, the corresponding criticality score is calculated, and the maximum value of the two is extracted as the Gaussian kernel criticality score of the candidate parameter combination.

[0152] In this embodiment, the flow velocity can also be predicted using an initial response surface model for candidate parameter combinations, and a Gaussian kernel criticality score can be calculated based on the predicted flow velocity and a safe flow velocity threshold. Specifically, the initial response surface model, fitted to an initial sample set, generates candidate parameter combinations in the embankment geometric parameter space and the gate scheduling parameter space; the initial response surface model is used to predict the flow velocity using these candidate parameter combinations, and a Gaussian kernel criticality score is calculated based on the predicted flow velocity and a preset safe flow velocity threshold.

[0153] Step 503: Based on the Gaussian kernel criticality score, adaptive enrichment sampling is performed within the boundary zone where the predicted flow velocity is close to the safety performance threshold. The sampled points are added to the initial sample set to obtain the expanded sample set.

[0154] Specifically, the Gaussian kernel criticality scores of all candidate parameter combinations are sorted in descending order, and a predetermined number of candidate parameter combinations with the highest scores are selected. To avoid excessive local clustering, a minimum Euclidean distance constraint is applied simultaneously during selection. The selected candidate parameter combinations are submitted to a computational fluid dynamics numerical simulation platform to obtain high-fidelity measured flow velocity data, which is then supplemented and merged into the original data sequence to output an expanded sample set.

[0155] Step 504: Based on the expanded sample set, a segmented proxy model is fitted and constructed.

[0156] Using the augmented data sequence, a mathematical regression surface was fitted. Since the fluid dynamics exhibit nonlinear evolution under different embankment sizes, a partitioning strategy based on physical parameters was employed for specific fitting.

[0157] Step 505: Calculate the physical water resistance ratio corresponding to each sample point in the expanded sample set.

[0158] Based on pre-set channel dimensions and the width and draft of the floating breakwater at each sample point, the physical water resistance ratio is obtained through division. This value characterizes the intensity of obstruction and disturbance of the floating breakwater structure on the water passage section of the approach channel.

[0159] Step 506: Using the physical water resistance ratio as the basis for spatial division, the parameter joint space is divided into multiple physical partitions with different flow characteristics.

[0160] Correspondingly, as the physical flow barrier ratio increases, the flow state evolves from linear flow around the body to nonlinear backwater and even abrupt flow transitions. Based on this characteristic, multiple physical threshold parameters are set. The specific number of physical zones and the boundary thresholds of the zones can be determined by observing the abrupt slope of the response surface or by combining empirical criteria for hydrodynamic flow transitions, based on the nonlinear characteristics of the velocity response in the initial sample set as a function of the physical flow barrier ratio. The number of physical zones and the boundary thresholds can be adjusted according to the hydraulic characteristics of the actual project and the fitting requirements of the surrogate model.

[0161] As an optional implementation, the joint space is divided into three regions. For example, the region with a physical water resistance ratio of less than 0.15 is a weak disturbance zone, the region with a physical water resistance ratio between 0.15 and 0.40 is a medium disturbance zone, and the region with a physical water resistance ratio greater than 0.40 is a strong disturbance zone.

[0162] Step 507: Within each physical partition, independently construct a polynomial response surface model using sample points falling into that partition.

[0163] Specifically, for each physical partition after segmentation, a separate set of sample points belonging to that space is extracted. A standard second-order polynomial response surface model is then used to fit regression coefficients to the local data clusters within each physical partition. Since the flow regime changes within each partition are relatively smooth, this local modeling strategy can reduce the requirement for the polynomial order and avoid the local underfitting or overfitting phenomena caused by using a global single model.

[0164] Step 508: At the boundaries of adjacent physical partitions, use a smooth transition function to continuously fuse and stitch adjacent polynomial response surface models to output a globally continuously differentiable segmented proxy model.

[0165] Accordingly, independent modeling of each partition is completed, and transition weights are used to switch the numerical values ​​of the model output at the partition boundaries, so that the global proxy model has continuous differentiability. The formula for the smooth transition function is:

[0166] w _left =1 / (1+exp((α-α _t ) / ε));

[0167] Among them, w _left Here, exp represents the activation weight value for the physical partition to the left of the adjacent boundary, α is the natural exponential function, and α is the physical water resistance ratio obtained from the analysis of the current input parameters. _t ε is the preset boundary threshold at the intersection of adjacent physical partitions, and ε is the preset smooth transition half-width parameter.

[0168] Further, the activation weight value of the physical partition to the right of the adjacent boundary is calculated, which is 1 minus the activation weight value of the physical partition to the left. Next, the velocity prediction output of the polynomial response surface model of each physical partition is multiplied by its corresponding activation weight value and accumulated to output the smoothed and fused global velocity prediction value.

[0169] Step 509: Independently calculate the leave-one-out cross-validation determination coefficients of the polynomial response surface model within each physical partition, and calculate the global normalized root mean square error over the entire parameter space.

[0170] To verify the generalization ability and prediction accuracy of the splicing model, offline error statistics were performed. For local partitions, a leave-one-out cross-validation mechanism was adopted, that is, after successively removing individual sample points, the prediction bias of the remaining samples for that point was calculated, and the leave-one-out cross-validation determination coefficient was output to evaluate the local goodness of fit of that partition. Correspondingly, the measured flow velocity and predicted flow velocity of all samples in the full parameter space were compared, and the sample range was combined to output the global normalized root mean square error to evaluate the overall bias of the model.

[0171] Step 510: When the leave-one-out cross-validation determination coefficient of any physical partition is lower than the preset single-partition accuracy lower limit, or the global normalized root mean square error is higher than the preset global error upper limit, the model accuracy is determined to not meet the acceptance criteria. When the leave-one-out cross-validation determination coefficients of all physical partitions are not lower than the preset single-partition accuracy lower limit, and the global normalized root mean square error is not higher than the preset global error upper limit, the current segmented surrogate model is output as the final model.

[0172] The two statistical values ​​are input into the discriminator. For example, a lower limit for single-zone accuracy and a higher limit for global error are preset. When the local coefficient of any partition fails to meet the standard, or the overall error exceeds the standard, the current version of the proxy base is determined to be invalid, and this is output to the online optimization module.

[0173] Step 511: Trigger local supplementary sampling in physical partitions where the accuracy does not meet the acceptance criteria, and return to refit the polynomial response surface model.

[0174] For velocity field proxy models that fail to meet the acceptance criteria, a targeted closed-loop error correction mechanism is implemented. Specifically, a preset physical partition where the leave-one-out cross-validation coefficient of determination is insufficient is located. Supplementary sample points are generated only within this weak space, and simulation data is acquired. Accordingly, the reconstructing instruction is executed again until all statistical indicators meet the accuracy threshold requirements. In this embodiment, this targeted resampling logic prevents computational resource overload within the accurate physical partitions, thereby improving the resource conversion efficiency of proxy model construction.

[0175] For example, this application also proposes a targeted verification closed-loop method based on cross-validation of digital space and physical models, used for physical verification of optimal structural parameters and recommended equipment operation strategies. This embodiment takes hydraulic model tests as the implementation object, specifically including:

[0176] Step 601, after outputting the recommended device operation strategy, also includes:

[0177] For the recommended scheme consisting of optimal structural parameters and recommended equipment operation strategies, obtain the corresponding measured flow velocity values ​​of the physical model.

[0178] Specifically, the optimal floating breakwater geometry obtained from the outer layer optimization and the recommended gate opening combination obtained from the inner layer optimization guide the actual physical hydraulic model test. Flow velocity sensors are placed in the water tank of the physical model to measure the actual flow characteristics within 300 meters of the approach channel entrance and at the end of the stilling basin. The collected time-averaged flow velocity data is scaled down to generate the measured flow velocity values ​​of the physical model. This process introduces the optimization results from the purely digital computational space into a real physical flow field environment for verification.

[0179] Correspondingly, several perturbation schemes can be generated in the local neighborhood space of the optimal parameter combination, and the physical model test data of the neighborhood scheme can be obtained to verify the stability of the local flow field gradient.

[0180] Step 602: Obtain the measured flow velocity value of the physical model and the first and second safety constraint responses output by the segmented proxy model, and calculate the prediction deviation.

[0181] Alternatively, it can be described as the prediction deviation between the measured flow velocity values ​​of the physical model and the corresponding first and second safety constraint responses output by the segmented proxy model or performance response proxy model.

[0182] Measured data and predicted data from segmented surrogate models using the same set of combined parameters in physical fluid dynamics experiments are extracted, and numerical differences and ratio calculations are performed. Specifically, the measured and predicted values ​​of the maximum confluence velocity at the navigation channel entrance are extracted simultaneously, as are the measured and predicted values ​​of the maximum longitudinal velocity exiting the stilling basin. For each monitoring dimension, the absolute value of the difference between the measured and predicted values ​​is calculated and defined as the absolute deviation characteristic; the ratio of this absolute difference to the measured value is also calculated and defined as the relative deviation characteristic. Through this dual deviation extraction method, the prediction residual level of the underlying velocity field surrogate model near the current optimal solution region can be comprehensively quantified.

[0183] Step 603: When the prediction deviation does not exceed the convergence allowable range, the recommended scheme is output as the final output scheme; when the prediction deviation exceeds the convergence allowable range, the recommended scheme and its corresponding physical model measured flow velocity value are included as supplementary sample points, triggering the refitting of the segmented surrogate model and the re-solution of the optimal structural parameters.

[0184] For example, the preset convergence allowable range is defined by both absolute deviation and relative deviation conditions. The absolute deviation threshold and relative deviation threshold are configured. The joint criterion formula for determining the convergence of the prediction model is:

[0185] abs(v _test -v _pred ) / v _test ≤e _rel ;

[0186] abs(v _test -v _pred )≤e _abs ;

[0187] Among them, v _test v represents the measured flow velocity value from the physical model. _pred The velocity response prediction is the output of the velocity field surrogate model, where abs is the absolute value function, and e _rel e is the preset relative deviation threshold. _abs This is the preset absolute deviation threshold.

[0188] In this embodiment, to ensure engineering safety and prediction reliability, a relative deviation threshold and an absolute deviation threshold are preset. When both inequalities are true, the model is deemed to have passed convergence verification. If either inequality is false, the prediction deviation is deemed to have exceeded a preset convergence allowable range.

[0189] Insufficient accuracy in the optimal solution region prediction may distort the original output of the levee parameters. Accordingly, real flow velocity data obtained from physical model experiments are extracted and merged into the original training database as high-fidelity sample points. Based on the updated database, the polynomial response surface fitting command is re-executed, and the entire process of optimizing outer geometric parameters and inner equipment scheduling is restarted.

[0190] In this embodiment, the targeted error correction mechanism can achieve adaptive evolution of the digital twin model by continuously absorbing data feedback from the real physical world, so that the output water conservancy project design scheme has physical reliability.

[0191] For example, when this invention is applied to the collaborative design of floating dike parameters and gate scheduling in hydraulic engineering projects, compared with traditional fixed single-condition design methods, the feasible scheduling domain volume index corresponding to the optimal floating dike parameters output by this invention is improved, the number of selectable scheduling schemes during operation increases, and the adaptability of cross-condition scheduling is enhanced. Simultaneously, spatial interaction indicators show that the degree of flow field coupling interference between different gate regions is reduced, and the ability for independent zone control is improved. The specific improvement in these indicators varies depending on the project scale, hydrological conditions, and gate zoning method.

[0192] This application also provides a scheduling collaborative optimization system based on segmented agent and feasible region evaluation. The system includes a processor, which is coupled to a memory. The memory stores instructions, and when the instructions are executed by the processor, the above-mentioned scheduling collaborative optimization method based on segmented agent and feasible region evaluation is implemented.

[0193] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, a scheduling collaborative optimization method based on segmented agents and feasible domain evaluation is implemented.

[0194] This application also provides a computer program product comprising instructions which, when executed by a computer, implement the methods performed in the above-described method embodiments.

[0195] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A scheduling collaborative optimization method based on segmented agency and feasible region evaluation, characterized in that, include: Obtain the engineering structure parameter space, equipment scheduling parameter space, and operating condition set including operating condition load parameters; Physical conservation constraints are constructed based on load parameters. For the joint parameter combination of structural parameters and scheduling parameters that satisfy physical conservation constraints, the first security constraint response and the second security constraint response are predicted by a segmented surrogate model. Feasible scheduling domains are divided using safety performance thresholds, and scheduling flexibility evaluation indicators are extracted based on the feasible scheduling domains, including volume indicators for quantifying the proportion of feasible solutions and spatial interaction indicators for quantifying the regional impact distribution of each device. In the engineering structural parameter space, with the goal of minimizing the structural engineering volume and the constraint that the volume index and the spatial interaction index meet the performance threshold, the optimal structural parameters are solved. Within the feasible scheduling domain corresponding to the optimal structural parameters, a recommended equipment operation strategy is output with the goal of minimizing the response to the second security constraint.

2. The method according to claim 1, characterized in that, The feasible scheduling domain is divided and volume metrics are calculated, including: In the equipment scheduling parameter space, based on the total number of gate openings in each equipment area, discrete combinations of opening numbers limited by the total number of gate openings are extracted. Under the assumption of uniform aperture, and based on physical conservation, the continuous aperture ratio corresponding to each discrete aperture number combination is analytically solved; Combinations whose continuous opening ratios satisfy the preset opening boundary conditions are selected to form a set of effective candidate scheduling schemes; Based on the first and second security constraint responses of each scheme in the effective candidate scheduling scheme set, as well as the security performance threshold, the feasibility of each scheme is determined, and the volume index is calculated accordingly.

3. The method according to claim 2, characterized in that, The feasibility of each scheme is determined, and the volumetric parameters are calculated accordingly, including: Call the hard indicator function to perform a binary determination on whether the first safety constraint response and the second safety constraint response exceed the safety performance threshold; The number of schemes that simultaneously meet the safety performance threshold is counted, and the proportion of this number in the set of effective candidate scheduling schemes is determined as the volume index.

4. The method according to claim 2, characterized in that, When determining the feasibility of each scheme and calculating the volumetric indices accordingly, the specific steps include: Obtain the margin sensitivity parameters that are pre-calibrated by combining the predicted standard deviation of the segmented surrogate model; Using a smooth margin function that includes a margin sensitivity parameter, the degree to which the first safety constraint response and the second safety constraint response deviate from the safety performance threshold are converted into continuous margin scores. Calculate the average of the products of the margin scores for each scheme in the set of effective candidate scheduling schemes to obtain the margin-weighted volume index.

5. The method according to claim 1, characterized in that, Extract spatial interaction metrics, specifically including: Obtain the preset baseline scheduling state, and use the segmented agent model to calculate the first-order sensitivity of the scheduling parameters of each device area relative to the response of the first security constraint. Extract the absolute range of each first-order sensitivity and the maximum absolute sensitivity; The sensitivity uniformity index is obtained by using the absolute value range and the maximum absolute sensitivity for zero-normalization calculation, and then used as the spatial interaction index.

6. The method according to claim 1, characterized in that, Extract spatial interaction metrics, specifically including: A continuous relaxation mechanism is introduced into the discrete number of openings variable in the equipment scheduling parameter space to map the discrete number of openings variable to a continuous number of openings variable on a continuous real closed interval. According to the master-slave variable separation criterion, the second-order cross partial derivatives of the first safety constraint response are calculated only for each continuous opening number variable of the same dimension. A cross-regional interaction matrix is ​​constructed based on the second-order cross partial derivatives, and the proportion of the sum of the absolute values ​​of the cross-regional interaction terms relative to the sum of the absolute values ​​of all elements in the cross-regional interaction matrix is ​​extracted. The interval decoupling degree, which reflects the physical isolation effect, is calculated based on the proportion, and is used as a spatial interaction indicator.

7. The method according to claim 1, characterized in that, The pre-construction of the segmented proxy model specifically includes: Based on the engineering structure parameter space and equipment scheduling parameter space, we perform master-slave variable dimensionality reduction and equation balancing sampling to obtain the initial sample set. Based on the initial sample set, a preliminary response surface model is fitted and constructed, generating candidate parameter combinations; Based on the candidate parameter combinations, the preliminary prediction results of the initial sample set, and the safety performance threshold, the Gaussian kernel criticality score is calculated. Based on the Gaussian kernel criticality score, adaptive enrichment sampling is performed within the boundary zone where the predicted flow velocity is close to the safety performance threshold. The sampled points are then added to the initial sample set to obtain the expanded sample set. Based on the expanded sample set, a segmented proxy model is constructed.

8. The method according to claim 1, characterized in that, After outputting the recommended device operation strategy, it also includes: For the recommended scheme consisting of optimal structural parameters and recommended equipment operation strategies, obtain the corresponding measured flow velocity values ​​of the physical model; The measured flow velocity values ​​of the physical model and the first and second safety constraint responses output by the segmented proxy model are obtained respectively, and the prediction deviation is calculated. When the prediction deviation does not exceed the convergence allowable range, the recommended output scheme is the final output scheme; When the prediction deviation exceeds the convergence allowable range, the recommended scheme and its corresponding physical model measured flow velocity values ​​are included as supplementary sample points, triggering the refitting of the segmented surrogate model and the re-solution of the optimal structural parameters.

9. The method according to claim 3 or 4, characterized in that, The set of operating conditions includes multiple operating conditions under different hydrological scenarios, and the calculated volumetric indices include: Configure corresponding operating condition frequency weights based on the annual frequency distribution characteristics of each operating condition. Based on the operating condition frequency weight, the single-condition volume indexes calculated separately under all operating conditions are weighted and summed to obtain the volume index that characterizes the cross-scenario adaptability under all operating conditions.

10. A scheduling collaborative optimization system based on segmented proxy and feasible region evaluation, characterized in that, The system includes a processor coupled to a memory storing instructions which, when executed by the processor, implement the method as described in any one of claims 1 to 9.

Images