A long-distance water delivery system pump-gate valve combined scheduling strategy optimization method

Abstract

This invention discloses an optimization method for the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system, relating to the fields of water transfer engineering and hydraulic control technology. Under given engineering conditions and safety constraints, the opening and closing timing and process of multiple controlled structures in the water conveyance system are uniformly parameterized and modeled. Through the unified vectorized expression of pump and gate valve opening and closing parameters and the construction of a comprehensive objective function based on a fixed normalized scale, a parallel optimization process combining one-dimensional transient simulation and constrained Bayesian optimization is used to carry out integrated iterative calculation of simulation and optimization. Within limited simulation resources and time, a safe, feasible, and comprehensively efficient opening and closing scheduling strategy is quickly approximated, achieving automatic generation of scheduling strategies and significantly improving optimization efficiency and the hit rate of feasible solutions.

Description

Technical Field

[0001] This invention relates to the field of water diversion engineering and hydraulic control technology, and in particular to an optimization method for the joint scheduling strategy of pumps, gates, and valves in long-distance water conveyance systems. Background Technology

[0002] Long-distance water conveyance projects are prone to transient problems such as negative pressure, excessive pressure, and significant fluctuations in the water level of shafts and forebays connected to tunnels under extreme disturbance conditions such as pump shutdown and rapid opening and closing of gate valves. These problems endanger the structural safety of the pipeline and connecting structures, and also affect the continuity and stability of the project operation. For cascade water conveyance systems with multiple pumping stations, control gates, and shafts, hydraulic disturbances generated by the operation of any upstream pumping station or gate valve can propagate along the system and superimpose between different stages, even causing the emptying or overflow of storage facilities with free water surfaces, such as the forebay of the pumping station, resulting in engineering safety risks. At the same time, from an operational economic perspective, the project needs to consider the cost and benefits of water conveyance and transfer, reduce unnecessary water wastage and discharge, and avoid increasing operating costs and wasting water resources.

[0003] In current engineering practice, the operation mode of pump gate valves is usually determined by relying on the engineering experience of dispatchers or existing dispatch rules of similar projects, using empirical formulas and offline calculation results of extreme conditions. Specifically, this often involves parametric calculations based on a one-dimensional transient flow simulation program, focusing on key parameters such as the timing, sequence, and rate of gate valve opening and closing. Several trial combinations are formed in the time and rate dimensions according to a given step size. One-dimensional transient calculations are performed on each combination, and indicators such as minimum pressure, maximum pressure, and vertical well water surface amplitude are statistically analyzed. Based on this, dispatch schemes are manually compared and selected.

[0004] However, for long-distance water conveyance systems with multiple hydraulic structures and cascade connections, the above-mentioned methods based on experience-based settings and parameterized exhaustive trial calculations have significant limitations: on the one hand, the scheduling rules rely heavily on the personal experience of the scheduling personnel and lack a systematic optimization mechanism that considers pressure safety constraints and engineering economic benefits in a unified manner; on the other hand, as the number of pumps and gate valves increases, the dimensions of parameters such as opening and closing timing and rate increase significantly, and the number of trial combinations formed under finite step size grows rapidly. Each scheme requires a complete one-dimensional transient simulation, resulting in high computational costs, low efficiency in finding feasible schemes that meet the constraints of pressure and water surface fluctuations along the entire line, and difficulty in globally weighing and optimizing operational benefits under strict safety constraints, making it impossible to achieve the global optimum of pump-gate-valve joint scheduling.

[0005] Therefore, it is necessary to propose a method for optimizing the joint scheduling of pumps, gates, and valves in long-distance water conveyance systems while balancing safety and efficiency under the constraints of limited simulation resources. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides an optimization method for the joint scheduling strategy of pumps, gates, and valves in long-distance water conveyance systems. By unifying the vectorized expression of pump and gate valve opening and closing parameters, constructing a comprehensive objective function based on a fixed normalized scale, and combining a parallel optimization process of one-dimensional transient simulation and constrained Bayesian optimization, a safe, feasible, and highly efficient scheduling strategy can be quickly approximated within limited simulation resources and time, thereby improving optimization efficiency and the hit rate of feasible solutions.

[0007] This invention is implemented as follows:

[0008] It includes two aspects, the first aspect:

[0009] An optimization method for the joint scheduling strategy of pumps and gate valves in a long-distance water conveyance system includes the following steps:

[0010] Step 1: Establish a one-dimensional transient flow model of the water conveyance system. The input of the model includes pipe geometric hydraulic parameters, hydraulic component parameters, boundary conditions and scheduling control trajectory. The output includes node piezometric head, pipe section flow rate, free liquid surface structure water level and overflow flow rate.

[0011] Step 2: Set up a set of controllable units participating in the joint scheduling, wherein the controllable units include water pumps, gates, and valves;

[0012] Construct control constraints for the controllable units, and define the waiting time of each controllable unit and the total system waiting time;

[0013] Different control methods are adopted for different types of adjustable units;

[0014] Integrate all scheduling parameters to be optimized to form a joint scheduling decision vector;

[0015] The gate or valve opening trajectory, pump speed trajectory or start / stop state trajectory formed after decoding the joint scheduling decision vector, together with the initial system conditions, boundary conditions, evaluation period and calculation time step, are used as the scheduling input of the one-dimensional transient flow model to generate the transient flow response under the corresponding conditions.

[0016] Extreme values ​​are extracted based on the transient flow response, and safety constraints are constructed, including lower limit constraint of piezometric head, upper limit constraint of piezometric head, lower limit constraint of free liquid level, upper limit constraint of free liquid level, and pump speed constraint.

[0017] Step 3: Construct a comprehensive objective function, which is obtained by normalizing and weighting the total water wastage, total action duration, and total waiting time.

[0018] Step 4: Use the Gaussian process proxy model to predict the comprehensive objective function value and security constraint function value of the unevaluated joint scheduling scheme, and select the candidate joint scheduling scheme that needs to be evaluated by calling the one-dimensional transient flow model for high-fidelity simulation through the acquisition function.

[0019] A data acquisition function is constructed that integrates the expected improvement amount and the joint feasible probability, and the optimization direction is guided by the data acquisition function under engineering safety constraints.

[0020] A batch candidate sample generation strategy with minimum distance constraints is adopted, and the joint scheduling scheme is iteratively optimized by combining multi-core parallel computing.

[0021] Step 5: After the optimization iteration terminates, select feasible samples that satisfy all security constraints from all evaluated samples, and take the feasible sample with the smallest comprehensive objective function value as the candidate optimal joint scheduling scheme. Decode the specific scheduling parameters of each controllable unit and generate the optimal joint scheduling parameter table. Call the one-dimensional transient flow model again to perform full-time verification calculation on the optimal joint scheduling scheme. If the verification passes, output the optimal joint scheduling scheme, full-time calculation data and final verification result table.

[0022] Furthermore, in step 2, the control unit controls the constraints to satisfy the formula:

[0023]

[0024] In the formula, Let m be the earliest allowed action time of the m-th adjustable unit. This is the actual start time of the action. For the duration of the action, The moment the action ends; The maximum allowable delay time for the m-th adjustable unit relative to the earliest allowed action time; and These are its shortest and longest allowed action durations, respectively;

[0025] The waiting time of the m-th adjustable unit is calculated using the following formula:

[0026]

[0027] The total waiting time of all adjustable units in the system is calculated using the following formula:

[0028]

[0029] In the formula, M represents the total number of adjustable units. Let x be the total waiting time, and let x be the joint scheduling decision vector.

[0030] Furthermore, in step 2, the control method is adapted to the different types of adjustable units. When the m-th adjustable unit is a gate or valve, its control quantity adopts the relative opening degree u. m (t)∈[0,1] represents the piecewise linear control method of "constant opening degree - linear change - constant opening degree", and the calculation formula is as follows:

[0031]

[0032] In the formula, u m,0 and u m,1 These represent the target opening before and after the action, respectively. The moment of the start of the action. The moment the action ends;

[0033] When the m-th adjustable unit is a water pump, its control process is represented by its speed or operating state; for a variable speed water pump, its speed control function n is defined. m (t) is:

[0034]

[0035] In the formula, and These are the rotational speeds before and after the action, respectively. The moment of the start of the action. The moment the action ends;

[0036] For shutdown conditions, the pump does not use the target speed after activation as the optimal control variable, but uses the shutdown command time, the start time of outlet valve closure, and the duration of outlet valve closure as scheduling parameters. During shutdown, the actual pump speed is obtained by solving the pump's full characteristic curve, unit inertial parameters, outlet valve closure process, and transient flow equations of upstream and downstream pipelines simultaneously. The actual speed is allowed to decay to zero during shutdown and may become negative under the backflow effect of water flow. A negative value indicates that the pump is running in reverse.

[0037] Furthermore, in step 2, the joint scheduling decision vector includes the start time of action of each controllable unit, the duration of action, and the normalized representation of the continuous control variables of each pump, the expression of which is:

[0038]

[0039] In the formula, M represents the total number of adjustable units. The number of continuous control variables for the water pump included in the optimization; t s,m This is the start time of the action of the m-th adjustable unit; The duration of the action of the m-th adjustable unit; Normalized representation of the continuous control variables for each water pump;

[0040] The continuous control variable of the water pump is mapped to the actual control quantity using the following formula:

[0041]

[0042] In the formula, , For the first The actual speed of each water pump The first The minimum and maximum speeds allowed for each pump, for pumps that only perform preset start-stop and do not participate in continuous regulation, their corresponding continuous control variables are not included in the joint scheduling decision vector. The formula is used for continuous speed regulation or start-up speed regulation conditions.

[0043] Furthermore, in step 2, the preset allowable values ​​of the lower limit constraint and the upper limit constraint of the piezometric head are calculated using the following formula:

[0044]

[0045]

[0046] In the formula, and These are the minimum and maximum permissible piezometric head for the j-th pressure monitoring section or node, respectively. This is the elevation of the top of the pipe at this cross-section; To prevent adverse negative pressure from occurring; This is the elevation of the pipeline centerline at this node; The allowable internal pressure of this cross section; ρ is the density of water; g is the acceleration due to gravity; This is a positive pressure safety margin;

[0047] The preset allowable values ​​for the lower and upper limits of the free liquid surface constraint are calculated using the following formula:

[0048]

[0049] In the formula, The maximum permissible water level, The minimum allowable water level, The control elevation for the top of the structure; For top safety margin; The elevation of the base slab of the structure; This is to ensure a safety margin at the bottom; The minimum operating water level specified in the operating procedures;

[0050] Free-floating structures must not experience uncontrolled overtopping exceeding the top control elevation; controlled water discharge through weir-type or pipe-type overflow facilities installed in the project shall be included in the total water discharge volume as an economic indicator.

[0051] Furthermore, in step 3, the construction method of the comprehensive objective function includes:

[0052] Calculate the total action duration and total waiting time based on the action timing parameters of each adjustable unit;

[0053] A normalized scale is determined for the total water discharge, total operation time, and total waiting time, and this normalized scale remains unchanged during the optimization process. The normalized scale for the total water discharge is determined as follows: it is determined by multiplying the 100% reference overflow capacity of each overflow facility at the highest permissible operating water level by the evaluation duration, with a non-zero lower limit set to avoid a zero denominator. The specific formula is as follows:

[0054]

[0055] In the formula, The lower limit value is non-zero. For the evaluation period, This represents the total number of overflow facilities in the system that may cause water spillage. This represents the reference overflow capacity of the k-th overflow facility;

[0056] Based on the normalized scale, the total water wastage, total action duration, and total waiting time are converted into normalized targets, and the calculation formula is as follows:

[0057]

[0058] In the formula, To normalize the total wastewater volume, To normalize the total motion duration, This represents the normalized total waiting time; a smaller value indicates better performance of the joint scheduling scheme. All are dimensionless targets;

[0059] in, The value of the dimensionless comprehensive objective function; , These are the weighting coefficients.

[0060] Furthermore, in step 4, the modeling process of the Gaussian process proxy model is as follows: based on the sample set that has completed the one-dimensional transient flow simulation evaluation, the normalized joint scheduling decision variables are used as inputs, and the comprehensive objective function value and each safety constraint function value are used as outputs to establish the objective function Gaussian process proxy model and the constraint function Gaussian process proxy model.

[0061] The objective function Gaussian process surrogate model is used to predict the mean and standard deviation of the comprehensive objective function of the candidate joint scheduling scheme, and the constraint function Gaussian process surrogate model is used to predict the mean and standard deviation of each security constraint function corresponding to the candidate joint scheduling scheme.

[0062] The predicted mean and standard deviation of the comprehensive objective function, as well as the predicted mean and standard deviation of each safety constraint function, are input into the acquisition function to select candidate joint scheduling schemes that require high-fidelity simulation evaluation using a one-dimensional transient flow model.

[0063] Furthermore, in step 4, the acquisition function is constructed as follows:

[0064] When there are feasible samples in the training dataset that satisfy all engineering safety constraints, the acquisition function is:

[0065]

[0066] When there are no feasible samples in the training dataset that satisfy all engineering safety constraints, the acquisition function is:

[0067]

[0068] in, , Let z be the expected improvement amount for candidate point z. Let z be the joint feasible probability of a candidate point z satisfying all engineering safety constraints. Let be the predicted mean and predicted standard deviation of the r-th constraint function at candidate point z, respectively; R be the total number of constraints; and Φ(·) be the cumulative distribution function of the standard normal distribution.

[0069] Furthermore, in step 4, the batch candidate sample generation strategy for the minimum distance constraint is as follows:

[0070] In each iteration, no more than q candidate samples are selected for parallel simulation evaluation; in the k-th iteration, Sobol sequences or other quasi-random sequences are used in the normalized variable space. Internally generated candidate point set Based on the collected function value Select the first candidate point ;

[0071] When selecting other candidate points sequentially, a triple mechanism of pseudo-observation update, local penalty function and minimum distance constraint is adopted;

[0072] Pseudo-observation update: The predicted mean of the selected candidate points is used as a temporary pseudo-observation value to construct a temporary surrogate model;

[0073] Local penalty function: Constructing a temporary acquisition function The collected function values ​​in the vicinity of the selected point are attenuated;

[0074] Minimum distance constraint: The candidate points must satisfy the normalized Euclidean distance constraint. ;

[0075] The b-th candidate point satisfies the constraints and makes... The largest; if no feasible point is found, then gradually relax the restrictions. False observations are not included in the training dataset.

[0076] in, To preset batch size, Let the dimension of the decision variables be... For the preset distance threshold, This is the candidate sample number.

[0077] Furthermore, step 5 also includes re-verifying the selected joint scheduling candidate optimal schemes using one-dimensional transient flow high-fidelity simulation. If the verification fails, the scheme is eliminated, and a second-best sample that satisfies the engineering safety constraints is selected as a new candidate optimal scheme. This verification is repeated until an optimal scheme that passes the verification is obtained. The selection of the optimal scheme satisfies the following calculation formula:

[0078]

[0079] in, For the final updated training dataset; This is the actual scheduling parameter vector corresponding to the i-th sample in the dataset; , These are the constraint function value and the comprehensive objective function value for the i-th sample, respectively.

[0080] The second aspect:

[0081] A joint scheduling strategy system for pumps, gates, and valves in a long-distance water conveyance system mainly consists of the following functional modules. These modules can be integrated into the same computing platform in software form, or they can be deployed in a distributed manner via a network or bus:

[0082] (1) Strategy and Sampling Module:

[0083] The strategy and sampling module is used to uniformly organize and initially set the control parameters such as the opening and closing time and opening and closing duration of the i-th gate and valve.

[0084] (2) Agent learning module:

[0085] The surrogate learning module is used to establish probabilistic surrogate models for the overall cost objective of long-distance water conveyance systems, as well as safety constraints such as piezometric head and water surface fluctuations.

[0086] (3) Constraint Acquisition Module:

[0087] The constraint acquisition module is used to calculate acquisition indicators and select the next batch of evaluation samples based on the surrogate model.

[0088] (4) Simulation model scheduling module:

[0089] The simulation model scheduling module is used to interface with the one-dimensional transient flow simulation engine and realize parallel computation and exception handling of multiple candidate scheduling strategies.

[0090] (5) Data and Visualization Module:

[0091] The data and visualization module is used to record optimization process data and support result analysis and display.

[0092] (6) Interface and configuration module:

[0093] The interface and configuration module is used to provide a unified parameter configuration and system call interface to the outside world, enabling the portability and integration of methods across different projects and simulation platforms.

[0094] The beneficial effects of this invention are:

[0095] 1. Balancing Safety and Efficiency: Pumps, gates, and valves are designated as jointly dispatchable and controllable units. Control constraints are constructed for each unit, and the waiting time of each unit and the total system waiting time are defined. Adaptive control methods are adopted for different types of units. Safety indicators such as minimum piezometric head, maximum piezometric head, highest water level, and free surface water level are incorporated into the constraints. Combined with parameters such as pipeline area and water hammer velocity, a constrained Bayesian optimization method is used to couple a Gaussian process surrogate model with one-dimensional transient flow simulation. Based on a normalized scale, the total water wastage, total action duration, and total waiting time are transformed into normalized objectives. A comprehensive objective function is constructed with fixed equal weights to quantify the overall merits of the scheme. Priority is given to searching for regions with high feasibility and low cost, while balancing system safety and operational efficiency.

[0096] 2. Improved optimization efficiency: The Gaussian process proxy model is used to predict the target and constraint responses of unevaluated schemes, reducing the number of high-fidelity simulation calls, predicting scheduling effects in advance, and simulating only candidate schemes with high potential to avoid redundant calculations. Within the limited simulation budget, the optimal feasible scheduling scheme can be quickly approximated.

[0097] 3. Adaptable to complex operating conditions: Pumps, gates, and valves are set as jointly dispatchable and adjustable units. The waiting time of each unit and the total waiting time of the system are defined. Corresponding and adapted control methods are adopted for different types of units. Piecewise linear control functions are constructed based on start-stop time, speed regulation duration, etc. The multi-pump and gate dispatching parameters are unified and parameterized. Combined with calculation parameters such as friction coefficient and pipe diameter, it can handle complex linkage operating conditions such as normal start-stop and power failure shutdown. The algorithm is resistant to nonlinearity and numerical fluctuations, supports parallel evaluation, and is adapted to long-distance multi-stage water conveyance systems.

[0098] 4. Intuitive Results and Facilitate Decision Making: Outputs the optimal solution and visualization results (time history curves, envelope diagrams, etc.), and simultaneously outputs parameter tables and final verification tables. It clarifies evaluation indicators such as minimum piezometric head, maximum piezometric head, highest water level, and total water wastage. Combined with the comprehensive objective function, it quantifies the overall merits and demerits of the solution, as well as the differences in safety margin and benefits, providing precise support for engineering decision making.

[0099] 5. Strong applicability and scalability: Based on a general model and standard framework, it can be connected to existing hydraulic calculation and monitoring systems; the modular decoupling design facilitates the expansion of constraints or indicators, adapts to new and existing projects, and supports result integration and process traceability.

[0100] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0101] Figure 1 This is a flowchart of the optimization method for the joint scheduling strategy of pumps and gate valves in long-distance water conveyance systems according to the present invention. Detailed Implementation

[0102] Example 1:

[0103] This embodiment presents an optimization method for the joint scheduling strategy of pumps and gate valves in a long-distance water conveyance system, such as... Figure 1 As shown, it includes the following steps:

[0104] Step S1, Data Acquisition and Preprocessing:

[0105] 1.1 Basic Data Collection

[0106] The basic data required for one-dimensional transient flow calculation and joint scheduling optimization of pumps, gates, and valves in long-distance water conveyance systems are collected, including parameters of pressurized pipe sections, parameters of hydraulic components, fixed and boundary parameters, operating constraint parameters of control units, and safety control parameters.

[0107] The parameters of the pressurized pipe section include the calculated length L of the pipe or steel-lined tunnel, the pipe diameter D, the pipe material roughness, the cross-sectional area of ​​the water passage A, the friction coefficient f, the water hammer velocity a, the elevation of the pipe centerline, and the initial flow rate.

[0108] The hydraulic component parameters include the geometric and performance parameters of pumps, valves, gates, surge tanks, forebays, outlet boundaries, and overflow facilities, such as pump head-flow characteristics, pump full characteristic curves, pump inertia parameters and allowable reverse rotation speed, valve flow coefficients, gate opening range, effective area of ​​surge tanks or forebays, weir crest elevation, weir width, weir flow coefficient, overflow pipe diameter, and flow coefficient.

[0109] The fixed and boundary parameters include gravitational acceleration g, upstream and downstream boundary water level or flow process, water outlet flow process, simulation time step, and the opening range, rotational speed range, allowable start and stop states, earliest allowable action time, and action duration range of water pumps, gates, and valves.

[0110] 1.2 Data Preprocessing

[0111] Units are standardized: the units for geometry, velocity, flow rate, and time are standardized to m, m / s, m³ / s, and s, respectively. Non-standard units are converted according to engineering conventions. If the collected data is a pressure value, it is converted to piezometric head.

[0112]

[0113] In the formula, H is the piezometric head, p is the relative pressure, ρ is the water density, and z is the elevation of the node or section.

[0114] Outlier handling: Remove outliers from parameters such as friction coefficient and water hammer velocity that deviate from the conventional range of values ​​for similar projects; missing parameters are supplemented with empirical values ​​from engineering projects.

[0115] Resistance conversion: The local resistance of non-controllable accessories such as elbows and reducers is converted into equivalent pipe length according to conventional methods in this field and incorporated into the adjacent pressurized pipe section;

[0116] Parameter calibration: Key hydraulic parameters such as friction coefficient, water hammer velocity, and cross-sectional area of ​​the pipeline are calibrated in conjunction with static hydraulic test data of the water conveyance system to ensure that the parameters match the actual project.

[0117] 1.3 Input Parameters Output

[0118] After preprocessing, standardized parameter tables are generated, including pipe segment parameter tables, node topology parameter tables, hydraulic component parameter tables, boundary condition parameter tables, initial operating condition parameter tables, control unit constraint parameter tables, overflow facility parameter tables, and safety constraint parameter tables. These parameter tables serve as inputs to the one-dimensional transient flow model and the joint scheduling optimization model, and are used for subsequent control equation solving, control trajectory generation, overflow statistics, evaluation of actual pump speed time history and maximum reverse speed, and safety constraint evaluation.

[0119] Step S2, one-dimensional modeling of the project:

[0120] A one-dimensional unsteady flow model is used to solve the hydraulic transient process. The model is based on the method of characteristics and supports components such as pipelines, shafts, elevated water tanks, surge tanks, forebays, valves, gates, and pumps. The model inputs are the geometric parameters of the tunnel pipeline and structures, roughness, wave velocity, and the control trajectories of each control unit generated by the decision variable φ. The outputs are the head, water level, flow rate, and overflow rate along the water conveyance system.

[0121] Based on the overall layout plan, longitudinal section, and equipment parameter table of the water conveyance system, the entire water conveyance system is abstracted as a one-dimensional hydraulic network composed of pressurized pipe sections, nodes, and local hydraulic components. The nodes include upstream boundary nodes, downstream boundary nodes, bifurcation nodes, confluence nodes, pump nodes, gate nodes, valve nodes, forebay nodes, and surge tank nodes. The pressurized pipe sections represent pressure pipelines, steel-lined tunnels, or connecting pipes. Elbows, reducers, joints, and other non-regulating accessories are incorporated into adjacent pipe sections after being calculated based on equivalent local resistance.

[0122] For the i-th pressurized pipe section, the piezometric head H is used. i (x,t) and flow Q i (x,t) are the state variables, where Q i =A i V i A i This refers to the cross-sectional area of ​​the pipeline through which water flows.

[0123] For long-distance pressurized water conveyance systems, under the condition that the flow velocity is much lower than the water hammer wave velocity, and neglecting the convection term, its one-dimensional unsteady flow governing equation can be written as:

[0124] (2.1)

[0125] (2.2)

[0126] In the formula, The piezometric head (m) is measured from the baseline. The cross-sectional average velocity is (m / s). This is the friction coefficient; It is the pipe diameter (m); The water hammer wave velocity (m / s); The acceleration due to gravity (taken as 9.81 m / s²) 2 ); , These represent distance (m) and time (s), respectively.

[0127] For a given pipeline, the water hammer wave velocity can be considered a constant. For steel pipes and steel-lined pressurized tunnels, the water hammer wave velocity is relatively high, usually around 1000 m / s, and its characteristic line is a straight line.

[0128] Equations (2.1) and (2.2) are a system of hyperbolic partial differential equations. Using the method of characteristics, the partial differential equations can be transformed into ordinary differential equations. The calculation formulas are as follows:

[0129] (2.3)

[0130] (2.4)

[0131] By taking the difference equation along the characteristic line, we can obtain the difference equation, and the calculation formula is as follows:

[0132] : (2.5)

[0133] : (2.6)

[0134] In the formula, Let Q = AV, representing the current node's undetermined flow. For the piezometric head to be determined, the coefficient is... , , and It is a known quantity from the previous moment, and its specific expression is: , , , ;constant , ; The area of ​​the pipe (m²) 2 ).

[0135] Solving equations (2.5) and (2.6) simultaneously yields...

[0136] (2.7)

[0137] After determining the flow rate, the piezometric head can be obtained from (2.5) or (2.6).

[0138] Discretize the i-th pipe segment into N. i The calculation unit is divided into several units, and the global calculation step size is taken as Δt. For each pipe segment, the number of segments and the spatial step size are determined, and the calculation formula is as follows:

[0139] (2.8)

[0140] in, Let be the length of the i-th pipe segment. This represents the actual water hammer wave velocity of the pipe section. For a globally uniform computation time step, The number of discrete computational units for this pipe section The discrete spatial step size of this pipe section.

[0141] To meet the calculation requirements of the method of characteristics, each pipe segment must satisfy the Courrant condition. The corrected calculated wave velocity was used for this pipe section. It participates in discrete calculations. This adjustment is only an optimization of the numerical calculation format and does not change the physical water hammer wave velocity of the pipe section, ensuring that the characteristic line trajectory accurately passes through the grid nodes and improving calculation stability.

[0142] To ensure the model can cover pumps, gates, valves, and free surface structures in long-distance water conveyance systems, the following node and component connection relationships are further established:

[0143] For a normal rigid connection node j, each connecting pipe segment satisfies equal head and continuous flow at this node;

[0144] For upstream reservoirs, downstream pools, or other constant water level boundaries, the method of giving boundary head is used;

[0145] For gate or valve components, they are uniformly regarded as local resistance components with variable opening degree, and a piecewise linear opening degree function is constructed from the start time of action, action duration, initial opening degree, and target opening degree.

[0146] For pump components, the flow rates before and after the pump are continuous and satisfy the pump head relationship. A piecewise linear control function is constructed based on the start-stop time, speed regulation time, target speed, or operating state. Under continuous speed regulation conditions, the pump control input is determined by the target speed or commanded speed process. Under normal shutdown or emergency shutdown conditions, the actual pump speed, flow rate, and torque time history are solved by the shutdown command, the outlet valve closing process, and the pump-pipeline coupling equation, and used to evaluate the shutdown reverse speed.

[0147] Free surface structures can be equipped with weir-type overflow facilities or pipe-type overflow facilities. The two types of overflow facilities correspond to different outflow relationships, but both output overflow processes. .

[0148] Discrete nodes are set at cross-sectional changes, hydraulic structure layouts, and abrupt changes in geometric parameters; the nodes are numbered consecutively from the system start point to the end point.

[0149] Before performing transient calculations, steady-state hydraulic calculations are first performed based on the initial boundary conditions, initial pump operating conditions, and initial gate valve opening to obtain the overall system performance. Initial water head at time Initial flow and the initial water level of the free surface The initial value serves as the initial condition for subsequent transient calculations.

[0150] At each time step The solution is performed in the following order:

[0151] First, update the control quantities of each pump, gate, and valve according to the scheduling scheme to be evaluated; second, calculate the characteristic constants of each pipe section; then, simultaneously solve the nodal continuity equation, boundary condition equation, pump head equation, valve / gate resistance equation, and forebay / surge well water balance equation to obtain the time... The simulation will then measure the head at each node, the flow rate in each pipe section, and the free surface water level. Finally, the simulation will continue until the preset termination time. The time step is determined based on the Courant number (CFL) condition to ensure numerical stability.

[0152] Based on the above one-dimensional transient flow model, the following response results can be obtained under any given scheduling scheme: head time histories of each node. Flow time history of each pipe section Time history of water levels in each forebay or surge tank Overflow flow time history Q of each overflow facility ov,k (t). Furthermore, evaluation indicators such as minimum head, maximum head, highest water level, and total water discharge can be extracted from the above time history results and used as inputs for subsequent safety constraint determination and objective function calculation.

[0153] Step S3, set the scheduling decision vector and security constraints:

[0154] 3.1 Controllable unit constraints:

[0155] The control constraints of the control unit are used to limit the search space of the joint scheduling decision vector, clarify the action time range, action duration boundary, and action end time calculation method of each controllable unit, and ensure that the unit actions meet the system operation requirements and avoid illegal operations.

[0156] Suppose there are M controllable units participating in the joint scheduling of a long-distance water conveyance system. These controllable units include pumps, gates, and valves. For the m-th controllable unit, its earliest allowed action time is defined as... The actual start time of the action is The action lasted for 10 minutes. The end time of the action is The control constraints of the regulation unit are constructed, and the calculation formula is as follows:

[0157] (3.1)

[0158] In the formula, The maximum allowable delay time for the m-th adjustable unit relative to the earliest allowed action time; and These are the shortest and longest allowed action durations, respectively.

[0159] The above constraints are used to limit the action sequence, action duration, and action end time of the adjustable unit, respectively.

[0160] The earliest allowed action time The timing of the accident, interlocking logic, minimum response delay, and operating procedures are predetermined.

[0161] 3.2 Define waiting time and total waiting time:

[0162] Waiting time is used to calculate the time cost term in the comprehensive objective function; it quantifies the delay time of each controllable unit from the earliest allowed action time to the actual action start time, as well as the total delay of all units, providing an indicator for time performance evaluation in scheduling optimization.

[0163] The waiting time of the m-th adjustable unit is defined as the difference between the actual start time of the action and the earliest allowed action time, and the calculation formula is as follows:

[0164] (3.2)

[0165] Therefore, the total waiting time of all controllable units in the entire system can be expressed as:

[0166] (3.3)

[0167] In the formula, The total waiting time will be used as one of the important evaluation parameters for scheduling optimization.

[0168] 3.3 Control methods for adjustable units:

[0169] 3.3.1 Control methods for gates / valve:

[0170] When the m-th adjustable unit is a gate or valve, its control quantity adopts the relative opening degree u. m (t)∈[0,1] represents the state and adopts a piecewise linear control method of "constant opening degree - linear change - constant opening degree":

[0171] (3.4)

[0172] In the formula, and These represent the target opening before and after the action, respectively.

[0173] When performing a gate or valve closing operation, take u m,0 =1,u m,1 =0; when performing a gate opening or valve opening operation, take u. m,0=0,u m,1 =1; When the unit remains in its original state and does not operate under a certain working condition, take u. m,0 =u m,1 .

[0174] 3.3.2 Water pump control method:

[0175] When the m-th adjustable unit is a water pump, its control process is represented by its speed or operating status.

[0176] (1) For a variable speed water pump, define its speed control function. for:

[0177] (3.5)

[0178] In the formula, and These are the rotational speeds before and after the action, respectively. , These are the start and end times of the action, respectively.

[0179] (2) For water pumps that only perform constant speed start-stop, discrete operating states s can be used. m (t) indicates that the operating state is either on or off. The start time of the action is... and the end time is .

[0180] (3) When the pump is in the shutdown evaluation condition, its target speed after action is not used as a joint scheduling decision variable; the variables related to the pump in the joint scheduling decision vector include the shutdown command time, the start time of outlet valve closure, the duration of outlet valve closure, and the waiting time parameters related to shutdown linkage. The actual pump speed time history and the maximum reverse speed are calculated by a one-dimensional transient flow model and used as the safety constraint evaluation result.

[0181] In a preferred embodiment, for pump shutdown conditions, the optimization method allows the actual pump speed to decrease from the rated speed to zero during shutdown, and to generate a negative speed (i.e., pump reversal) under the influence of upstream backflow. Pump reversal during shutdown is a common operating condition, and relevant national standards have clearly defined permissible limits for reversal speed. Unlike the practice of avoiding reversal as much as possible in general industrial pumping stations, this method adopts a controllable, short-term reversal within the standard limits under specific conditions of long-distance water transmission systems. The technical basis for this is that an appropriate reversal process can absorb some water hammer energy, reduce the peak positive pressure in the pipeline, and thus improve the hydraulic safety of the system. Meanwhile, the reversal is not unlimited; rather, the actual speed time history is obtained by simultaneously solving a one-dimensional transient flow model, and the maximum reversal speed is used as the safety constraint evaluation index. The occurrence of negative speed is an objective result of transient flow calculation, and its physical meaning is that the pump impeller rotates under the reverse drive of the water flow.

[0182] 3.4 Joint Scheduling Decision Vector:

[0183] The scheduling parameters to be optimized are uniformly integrated into decision vectors, simplifying the optimization calculation process, clarifying the scope and meaning of optimization variables, and providing a unified variable basis for subsequent scheduling optimization.

[0184] In a preferred embodiment, all scheduling parameters to be optimized are uniformly combined into a joint scheduling decision vector, and the calculation formula is as follows:

[0185] (3.6)

[0186] In the formula, M represents the total number of controllable units; The number of continuous control variables for the water pump included in the optimization; t s,m This is the start time of the action of the m-th adjustable unit; The duration of the action of the m-th adjustable unit; This is a normalized representation of the continuous control variables for each water pump.

[0187] The continuous control variables for the water pumps in Equation (3.6) are only applicable to water pumps participating in continuous speed regulation or start-up speed regulation. For the shutdown reverse evaluation condition, the continuous control variables for the water pumps may not be included in the decision vector, and the water pump scheduling scheme may be characterized by the shutdown command time and the outlet valve closing process parameters. The shutdown command time, the outlet valve closing start time, and the outlet valve closing duration are the calculation boundary conditions under the shutdown condition.

[0188] 3.5 Mapping relationship of continuous control variables for water pumps:

[0189] This section applies only to operating conditions where the pump participates in continuous speed regulation or start-up speed regulation. For operating conditions where the shutdown reverse speed is the key evaluation factor, the target pump speed is not considered as an optimization variable; the actual speed after shutdown is calculated using a one-dimensional transient flow model and the pump's inertial equations. The normalized continuous control variables of the pump are... This is mapped to the actual controllable quantity (speed) of the water pump, ensuring that the optimization variables correspond to the actual control requirements.

[0190] For the continuous control variable of the water pump , and actual speed The mapping relationship is as follows:

[0191] (3.7)

[0192] In the formula Let be the minimum and maximum permissible speeds of the l-th pump, respectively. This formula can be used to normalize the variables. Convert to a speed value that conforms to actual operating constraints.

[0193] When a water pump only performs preset start-stop operations and does not participate in continuous regulation, its continuous control variables are not included in the decision vector x.

[0194] 3.6 Correlation between scheduling input and transient flow model:

[0195] The connection between the aforementioned scheduling parameters and the one-dimensional transient flow model is clarified. The joint scheduling decision vector is decoded into the control trajectories of each controllable unit, and together with the system initial operating conditions, boundary conditions, evaluation period, and calculation time step, it serves as the input to the one-dimensional transient flow model to generate the transient flow response under the corresponding joint scheduling scheme.

[0196] The action timing parameters, action duration, and control trajectory determined by equations (3.1) to (3.7) are used as scheduling inputs for the one-dimensional transient flow model to generate the transient flow response under the corresponding operating conditions.

[0197] The scheduling input specifically includes: the earliest allowed action time for each adjustable unit. Actual start time of action Action duration End of action Maximum allowable delay time Shortest and longest motion duration , ;

[0198] For gates or valves, this also includes the opening degree before actuation. Target opening after action and opening trajectory ;

[0199] For variable speed water pumps, the initial operating speed is also included. Target speed upper and lower speed limits , and rotational speed trajectory .

[0200] For pump shutdown evaluation, the parameters also include shutdown condition type, shutdown command time, outlet valve closure start time, outlet valve closure duration, outlet valve closure trajectory, rated speed, unit inertia time constant, and pump full characteristic curve. These parameters, along with the system initial operating conditions, boundary conditions, evaluation period, and calculation time step, serve as inputs to a one-dimensional transient flow model for calculating the pump's actual speed, flow rate, torque, piezometric head, water level, and overflow time history.

[0201] 3.7 Constraints of the Water Conveyance System

[0202] Based on the calculation results of the one-dimensional transient flow model, the safe operation boundary of pressure pipelines, pressurized tunnels and free surface structures (such as forebays, surge tanks, and elevated water tanks) in the joint scheduling process is defined, and these safety requirements are uniformly transformed into mathematical constraint functions so that they can be directly connected with the constrained Bayesian optimization model to determine whether any joint scheduling scheme to be evaluated meets the engineering safety requirements.

[0203] 3.7.1 Obtaining the hydraulic response results:

[0204] For any joint scheduling scheme x to be evaluated, the established one-dimensional transient flow model is invoked during the evaluation period. The following hydraulic response results were obtained:

[0205] Piezometric head time history H at each monitoring node j (t;x);

[0206] Water level time history Z of each free liquid surface structure k (t;x);

[0207] The actual speed of the water pump and its maximum reverse speed;

[0208] Other necessary hydraulic response results

[0209] 3.7.2 Constraints on pressure pipelines and pressurized tunnels:

[0210] For piezometer head monitoring sections in pressure pipelines and pressurized tunnels, it is defined that they cannot have excessively low head (to prevent water column separation or pipeline instability caused by negative pressure) and excessively high head (to prevent pipe bursts or structural damage) during transient flow processes, and the minimum and maximum allowable piezometer head are given.

[0211] Suppose there are N in the system H There are j-th piezometric head monitoring sections. For the j-th piezometric head monitoring section, its minimum and maximum piezometric head and constraints during the evaluation period are defined as follows:

[0212] (3.8)

[0213] In the formula, H j,MIN and H j,MAX These are the minimum and maximum permissible piezometric head for the j-th head monitoring section or node, respectively. The actual piezometric head of the j-th monitoring section at time t and corresponding design variable x; and The monitoring section is for the entire calculation period. The minimum and maximum values ​​of the water head in the piezometer tube.

[0214] In a preferred embodiment, the minimum permissible piezometric head for the j-th cross-section is calculated using the following formula:

[0215] (3.9)

[0216] In the formula, This is the elevation of the top of the pipe at this cross-section; To prevent adverse negative pressure, a safety margin of 2m is preferred.

[0217] To prevent internal pressure from exceeding the structural bearing capacity of the pipeline or tunnel, and to avoid pipe bursts or lining damage, the maximum allowable piezometric head at the j-th cross-section is calculated using the following formula:

[0218] (3.10)

[0219] In the formula, The elevation of this node; The allowable internal pressure of this cross section; ρ is the density of water; g is the acceleration due to gravity; This represents the positive pressure safety margin.

[0220] 3.7.3 Constraints on the water pump:

[0221] To ensure that the pump control trajectory meets the equipment operation requirements, constraints are added to the pump speed. For variable speed pumps involved in continuous speed regulation, the speed control functions for both the pre-action speed and the target speed after action must meet the allowable speed range. This constraint does not apply to reverse operation.

[0222] (3.11)

[0223] In the formula, Let be the lower limit constraint function for the rotational speed of the m-th pump. Let x be the upper limit constraint function for the rotational speed of the m-th pump, and let x be the optimization decision vector. Let m be the minimum allowable speed of the m-th variable speed water pump. Let m be the maximum permissible speed of the m-th variable speed water pump. and These are the minimum and maximum speeds of the m-th variable speed water pump calculated by the model, respectively.

[0224] For pump units participating in the shutdown reverse speed evaluation, the reverse speed constraint function is defined as follows:

[0225] (3.11a)

[0226] In the formula, Let m be the maximum allowable reverse rotation speed of the m-th water pump.

[0227] 3.7.4 Constraints on free-floating structures:

[0228] For structures with free liquid surfaces, such as forebays, surge tanks, and elevated water tanks, it is defined that they must not overflow (above the top) or empty (below the bottom or the minimum water level in the operating procedure) during the scheduling process, and the highest and lowest allowable water levels are given.

[0229] Suppose there are N in the system Z There are three free surface monitoring structures. For the k-th free surface structure, its minimum and maximum water levels and safety constraints during the evaluation period are defined as follows:

[0230] (3.12)

[0231] In a preferred embodiment, to allow for safety margins at the top and bottom of the structure and to comply with permissible regulations, the permissible operating water level of the kth free surface structure can be determined by the following formula:

[0232] (3.13)

[0233] In the formula, The control elevation for the top of the structure; For top safety margin; The elevation of the base slab of the structure; This is to ensure a safety margin at the bottom; This is the minimum operating water level specified in the operating procedures.

[0234] The above water level limits are set to prevent uncontrolled overtopping of structures that exceed the top control elevation; among them, controlled water discharge through weir-type or pipeline-type overflow facilities installed in the project can be included in the total water discharge of the system as an economic indicator.

[0235] 3.7.5 Unified Expression and Feasibility Determination of Constraint Functions:

[0236] The aforementioned safety constraints are uniformly written in the form of constraint functions so that they can be directly applied to subsequent constraint Bayesian optimization model integration, thereby achieving automatic feasibility assessment of the solution.

[0237] (3.14)

[0238] In the formula: the first formula is the lower limit constraint of the piezometric head to prevent negative pressure conditions in the pipe network; the second formula is the upper limit constraint of the piezometric head to prevent the pipe network head from exceeding the limit; the third formula is the lower limit constraint of the free liquid level to avoid the water level being too low; the fourth formula is the upper limit constraint of the free liquid level to prevent the water tank or pool from overflowing; j is the piezometric head monitoring node number; and k is the free liquid level monitoring node number.

[0239] For the pump reverse speed constraint under shutdown conditions, the constraint functions defined by formulas (3.14), (3.11), and (3.11a) are used, and all constraint functions are uniformly numbered as follows:

[0240] (3.15)

[0241] in:

[0242]

[0243] In the formula, The number of piezometric head monitoring nodes, The number of structures on the free surface. The number of water pumps participating in the evaluation of the upper and lower limits of continuous speed regulation. The number of water pumps used in the shutdown reverse speed evaluation.

[0244] The feasibility judgment criterion is as follows: when the following constraints are all satisfied (not greater than 0): lower limit constraint of piezometric head, upper limit constraint of piezometric head, lower limit constraint of free liquid level, upper limit constraint of free liquid level, upper and lower limit constraints of speed of pumps participating in continuous speed regulation, and pump reverse speed constraint, the joint scheduling scheme is determined to be a scheme that meets the following conditions: A feasible solution is selected to meet the engineering safety requirements; otherwise, it is deemed infeasible. This includes lower and upper limits for piezometric head.

[0245] Step S4, Objective function normalization and construction of the comprehensive objective function:

[0246] Obtain the core operational data (hydraulic parameters, control action-related parameters) corresponding to the joint scheduling scheme to ensure that the evaluation results are consistent with the actual engineering operation conditions.

[0247] For any joint scheduling scheme to be evaluated, during the evaluation period The system internally calls the established one-dimensional transient flow model to perform transient calculations, obtaining the time history response of each hydraulic parameter of the system. Simultaneously, based on the action sequence parameters of each controllable unit in the joint scheduling scheme, the total action duration and total waiting time are calculated. The normalization scale is determined once before optimization begins based on equipment parameters, overflow facility parameters, and evaluation duration, and remains unchanged during subsequent optimization processes.

[0248] 4.1 Indicator Definition: Clarify the evaluation dimensions of the joint dispatch scheme (water loss, control workload, response speed), and ensure that the evaluation standards are unified and quantifiable through quantitative indicator definition.

[0249] Definition of total water wastage: For any joint dispatch scheme to be evaluated, its total water wastage V is defined as follows: over(x) is defined as the total overflow of all overflow facilities that may potentially release water during the entire evaluation period, calculated as follows:

[0250] (4.1)

[0251] In the formula, This represents the total number of overflow facilities in the system that may cause water spillage. Let x be the instantaneous overflow of the k-th overflow facility at time t and under scheduling scheme x. To evaluate the total duration.

[0252] Definition of total motion duration: Total motion duration T act (x) is defined as the sum of the action durations of all controllable units participating in the joint scheduling, and the calculation formula is as follows;

[0253] (4.2)

[0254] In the formula, M represents the total number of controllable units participating in joint scheduling. The duration of a single action of the m-th adjustable unit.

[0255] Total wait time definition: Total wait time Defined as the sum of the waiting times of all controllable units participating in joint scheduling relative to their earliest allowed action time, the calculation formula is as follows:

[0256] (4.3)

[0257] In the formula, Let m be the actual action moment of the m-th adjustable unit. This represents the earliest allowed action time for the m-th adjustable unit.

[0258] 4.2 Determination of Normalization Scale:

[0259] Eliminate the dimensional and order-of-magnitude differences among the three evaluation indicators (total water wastage, total action duration, and total waiting time), and avoid the cumbersome process of determining the scale through trial calculations of pre-assessment working conditions.

[0260] 4.2.1 Basis for constructing the normalization scale:

[0261] Assume the system has a total of For overflow facilities that may potentially release water, this invention uses the overflow capacity corresponding to the highest permissible operating water level as the basis for normalized scale construction. The structural parameters related to the calculation of water release volume include at least the type of each overflow facility, flow coefficient, weir crest elevation, weir width, submergence correction factor, overflow pipe inlet control elevation, overflow pipe outlet elevation, overflow pipe diameter, and number of parallel overflow pipes.

[0262] 4.2.2 Calculation of reference overflow capacity for different types of overflow facilities:

[0263] Weir-type overflow facility: For the k-th overflow facility, if it is a weir-type overflow facility, its weir crest elevation is... The maximum permissible operating water level is The flow coefficient is The effective weir width is The reference overflow capacity of the overflow facility is defined as follows: :

[0264] (4.4)

[0265] In the formula, The number of parallel overflow facilities of the weir flow type. is the flooding correction factor, and g is the gravitational acceleration.

[0266] Pipeline overflow facility: For the k-th overflow facility, if it is a pipeline overflow facility, then the reference overflow capacity of this overflow facility is defined as follows: :

[0267] (4.5)

[0268] In the formula, The number of parallel pipeline overflow facilities, For pipeline flow coefficient, The effective flow area of ​​a single overflow pipe, The elevation of the overflow pipe outlet. The elevation of the overflow pipe inlet is controlled.

[0269] 4.2.3 Specific calculation of the normalized scale of the three major evaluation indicators:

[0270] Based on the aforementioned reference overflow capacity and adjustable unit parameters, normalized scales for three evaluation indicators are defined to ensure that the scales are fixed, can be automatically calculated, and require no manual intervention, thereby improving the engineering feasibility and efficiency of the method.

[0271] The normalized scale S1 for total water discharge is determined by multiplying the 100% reference overflow capacity of each overflow facility at the highest permissible operating water level by the evaluation duration, and is defined as follows:

[0272] (4.6)

[0273] In the formula, This is a non-zero lower limit value, used to avoid the denominator being zero. .

[0274] Normalized scale of total motion duration The duration of the maximum single action based on the adjustable unit is defined as follows:

[0275] (4.7)

[0276] In the formula, The minimum non-zero value for the total action duration is 1 second, which is preferred. This represents the maximum duration of action for the m-th adjustable unit.

[0277] The normalized scale S3 for total waiting time: determined based on the maximum allowable waiting time difference of the adjustable unit, is defined as:

[0278] (4.8)

[0279] In the formula, The lower limit of the total waiting time is preferably 1 second. For the first The maximum allowable delay time for each adjustable unit.

[0280] 4.3 Construction of Normalized Objective and Integrated Objective Function:

[0281] 4.3.1 Normalization objective:

[0282] Using the normalization scale described above, the three core evaluation indicators are transformed into normalized targets, calculated using the following formula:

[0283] (4.9)

[0284] In the formula, The normalized total water wastage is calculated (the smaller the value, the smaller the water wastage loss and the better the economic efficiency). This is the normalized total action time (the smaller the value, the less the control workload). This is the normalized total waiting time (the smaller the value, the faster the control response).

[0285] 4.3.2 Construction of the comprehensive objective function:

[0286] Taking into account the water wastage loss, the workload of control actions, and the control response speed, the total water wastage, the total action duration, and the total waiting time were selected as the joint scheduling optimization targets.

[0287] Under the premise of meeting safety constraints, and taking into account the operational costs such as total water wastage, total action time, and total waiting time, it is believed that the smaller the water wastage, the better the economic efficiency, and the shorter the action time and waiting time, the faster the target water delivery state can be reached. In order to eliminate the differences in dimensions and orders of magnitude among the total water wastage, total action time, and total waiting time, and to avoid determining the normalization scale through trial calculations of pre-assessment working conditions, this step adopts a method of automatically assigning a normalization scale with fixed rules to construct a comprehensive objective function.

[0288] Define the comprehensive objective function as follows:

[0289] (4.10)

[0290] In the formula, The value is a dimensionless comprehensive objective function value. The smaller the value, the better the overall performance of the joint scheduling scheme.

[0291] 4.4 Data interface relationship with the one-dimensional transient flow model:

[0292] Clarify the data interaction logic between this step (objective function calculation) and the establishment of the one-dimensional transient flow model, forming a process of "scheduling scheme → model calculation → data feedback → objective function calculation".

[0293] The data interface relationship between this step and the one-dimensional transient flow model is as follows:

[0294] First, step S3 generates the control trajectories of each pump, gate, and valve based on the decision variable 𝑥, and inputs them into step S2 along with the system structural parameters, initial steady-state conditions, accident disturbance conditions, evaluation duration, and time step. Step S2 outputs the time histories of the piezometric head of each node, the flow time histories of each pipe section, the water level time histories of each free liquid surface structure, the actual speed time histories of the pumps, and the overflow flow time histories of each overflow facility. Step S4 calculates the total amount of water wasted, and calculates the total action duration and total waiting time based on the action sequence parameters in the joint scheduling scheme. Then, it calculates the comprehensive objective function value F(x) and provides this comprehensive objective function value to the constraint Bayesian optimization module in the subsequent step S5 for further optimization.

[0295] Through the above processing, the normalized scales of total water discharge, total action duration, and total waiting time can be automatically determined according to fixed rules before the optimization begins, without the need for pre-evaluation of working conditions. At the same time, the comprehensive objective function adopts a fixed equal weight form, and the calculation of total water discharge covers both weir flow type and pipeline type overflow facilities, thereby improving the clarity, repeatability, and engineering feasibility of the method.

[0296] Step S5, Model Training and Update:

[0297] After constructing the one-dimensional transient flow model, defining control variables and safety constraints, and constructing the normalized comprehensive objective function, a constrained Bayesian optimization method is used to iteratively optimize the joint scheduling scheme of pumps and gate valves. This constrained Bayesian optimization method does not directly replace high-fidelity simulation of the one-dimensional transient flow. Instead, it establishes a Gaussian process surrogate model of the objective and constraint functions based on samples that have undergone simulation evaluation. It then uses a data acquisition function to predict and screen candidate joint scheduling schemes with high improvement potential and a high probability of satisfying safety constraints. Finally, the one-dimensional transient flow model is invoked to perform high-fidelity simulation evaluation on these candidate schemes.

[0298] 5.1 Optimization Algorithm

[0299] After completing the construction of the one-dimensional transient flow model, the definition of control variables and safety constraints, and the construction of the normalized comprehensive objective function, the constrained Bayesian optimization method is used to iteratively optimize the joint scheduling scheme of pumps and gate valves.

[0300] The joint scheduling optimization problem of this invention is written as:

[0301] (5.1)

[0302] In the formula, x is the joint scheduling decision variable vector; Ω is the search space composed of the action start time, action duration, gate or valve opening boundary, pump speed boundary and start / stop state constraints; F(x) is the comprehensive objective function defined in step S4. The safety constraint function is defined in step S3, where R is the total number of constraints.

[0303] In a preferred embodiment, if only the lower limit of the piezometric head, the upper limit of the piezometric head, the pump speed, the lower limit of the free liquid level, and the upper limit of the free liquid level constraints are considered, then the total number of constraints is:

[0304] (5.2)

[0305] Where, N H N represents the number of piezometric head monitoring sections. Z To monitor the number of structures at the free surface level, The number of water pumps participating in the evaluation of the upper and lower limits of continuous speed regulation. The number of pumps participating in the reverse speed constraint evaluation.

[0306] For any candidate scheduling scheme x, its sample evaluation is carried out according to the following fixed procedure: First, according to step S3, the start time, duration, and control trajectory of each pump, gate, and valve are generated from x; second, the one-dimensional transient flow model of step S2 is called to calculate the piezometric head time history H of each node. j (t;x), Flow time history of each pipe section Q i(t;x), Time history of water level of each free liquid surface structure Z k (t;x) and the overflow flow time history Q of each overflow facility ov,k (t;x); Then, calculate the values ​​of each safety constraint function according to step S3, and calculate the comprehensive objective function value F(x) according to step S4; Finally, add the objective value and constraint value corresponding to the candidate scheduling scheme as a new observation sample to the training dataset.

[0307] 5.2 Initialize sample generation:

[0308] The decision variables are normalized to generate initial samples and complete simulation evaluation. An initial training dataset is constructed, and penalty rules for simulation failure samples are formulated.

[0309] To improve the numerical stability of the surrogate model training, the decision variables are first linearly mapped to the unit hypercube [0,1] according to their upper and lower bounds. d In the middle. Let the dimension of the decision variables be d, and the upper and lower bounds of the j-th decision variable be respectively... and Then the normalized variable and inverse normalization variables Defined as:

[0310] (5.3)

[0311] Where d is the dimension of the decision variable, Let j be the j-th normalized decision variable.

[0312] Before optimization, Latin hypercube sampling was used in the range [0,1]. d Generate an initial sample set. The initial sample size N. init The value is determined by the dimension d of the decision variables, the time required for a single one-dimensional transient flow simulation, available computational resources, and the optimization budget. To ensure the reproducibility of the examples, a fixed random seed can be set, but a fixed random seed is not a necessary limitation of this method.

[0313] Initial sample size N init Inverse normalization yields the corresponding And perform the following operations in sequence:

[0314] 1. Generate the control trajectory according to step S3;

[0315] 2. Call step S2 to perform a one-dimensional transient flow simulation;

[0316] 3. Calculate the values ​​of each constraint function;

[0317] 4. Calculate the comprehensive objective function value

[0318] This forms the initial training dataset:

[0319] (5.4)

[0320] in, Let be the normalized decision variable vector and the objective function value for the i-th sample. Let be the original decision variable vector for the i-th sample; represents the constraint function values ​​for the i-th sample; n is the number of samples in the training dataset.

[0321] Initial stage sample size n=N init N init Determined based on the dimension d of the decision variable, the time consumed in a single one-dimensional transient flow simulation, the computational resources used, and the total simulation budget;

[0322] In one embodiment, take .

[0323] In the formula, d is the dimension of the joint scheduling decision variables; 20 is the minimum number of samples to ensure the initial training stability of the Gaussian process surrogate model; 5d is used to ensure that the initial samples have basic spatial coverage in the normalized variable space. If the total high-fidelity simulation budget is small, then No more than 30% of the total high-fidelity simulation budget.

[0324] In the initial iteration phase, the number of training set samples satisfies n= After each subsequent iteration completes the real simulation evaluation and backfills the samples, n increases according to the actual number of samples added to the training dataset. For example, when the dimension of the decision variable d=12, take... =60; when d=3, take =20.

[0325] In this embodiment, the sample index is divided into two categories: the global general index is labeled with superscript (i) (e.g., ...). / / ), which represents the consecutive numbers of all samples in the dataset; the local index of the batch samples in the k-th iteration is labeled with the indices k, b (e.g., / After batches of samples are added to the dataset, they are mapped sequentially to the global index (i).

[0326] If a candidate solution exhibits numerical divergence, fails to solve, or returns missing values ​​during the simulation in step S2, it will be classified as a failed sample and penalized as follows:

[0327] (5.5)

[0328] in, , The penalty value for the objective function and constraint function of the simulation failure samples (in one embodiment, it can be taken as a sufficiently large positive number, for example) Penalty value (This value is based on experience and can be adjusted according to the magnitude of the comprehensive objective function).

[0329] 5.3 Gaussian process surrogate modeling with objective and constraint functions:

[0330] Based on the current training dataset D n The normalized joint scheduling decision variable z is taken as input, and the comprehensive objective function value F and the values ​​of each security constraint function are taken as input. For the output, a Gaussian process surrogate model for the objective function and a Gaussian process surrogate model for the constraint functions are established. The Gaussian process surrogate model for the objective function is used to predict the mean and standard deviation of the comprehensive objective function of the candidate joint scheduling scheme; the Gaussian process surrogate model for the constraint functions is used to predict the mean and standard deviation of each security constraint function corresponding to the candidate joint scheduling scheme.

[0331] Suppose the observed response satisfies:

[0332] (5.6)

[0333] In the formula, For the first Observe the response value of each proxy object; This represents the true response function to be learned. Represents the Gaussian white noise term. For the first The noise variance of each proxy object allows different proxy objects to have different noise levels.

[0334] For the objective function proxy model, we have:

[0335] (5.7)

[0336] In the formula, The true response function of the objective function; The mean function of the surrogate model for the objective function; Let z be the covariance function of two distinct candidate points z and z′.

[0337] For the r-th constraint function surrogate model, we have:

[0338]

[0339]

[0340] (5.8)

[0341] In the formula, Let r be the true response function of the r-th constraint function; when hour = ,when hour = ; The variance of the signal; For the first The feature length scale of each input dimension; The total dimension of the decision variables; This represents the total number of constraint functions. Spatial distance parameter; Let j be the j-th normalized component of the two candidate points.

[0342] The kernel function hyperparameters can be determined by maximizing the log-marginal likelihood and can be solved using gradient optimization algorithms such as L-BFGS-B. To reduce the influence of local optima, repeated optimization at multiple initial points can be used, selecting the hyperparameter combination with the highest log-marginal likelihood. The specific convergence threshold and maximum number of iterations are determined based on the training sample size, computational resources, and the stability requirements of the surrogate model.

[0343] (5.9)

[0344] In the formula, Let be the optimal hyperparameters for the l-th proxy model; Let be the training response vector of the l-th surrogate model; Let be the value vector of the mean function of the l-th surrogate model at the training point; The training sample covariance matrix is ​​composed of kernel functions; Let l be the noise variance of the l-th surrogate model; It is the identity matrix; This represents the number of training samples.

[0345] Based on the current training dataset, the predicted mean and predicted variance for any candidate point z are as follows:

[0346] (5.10)

[0347] in, ) represents the covariance vector between the candidate point and all training samples. The kernel function value (prior variance) of the candidate point itself. For noise variance, Let y be the identity matrix and y be the target value vector of the training set.

[0348] When the sample size meets the evaluation requirements, indicators such as leave-one-out cross-validation error, standardized residuals, negative logarithmic prediction density, or prediction interval coverage can be used to monitor the quality of the surrogate model. The above monitoring indicators are used to help determine whether to increase the sample size or adjust the kernel function, and are not necessary conditions for the implementation of this method.

[0349] 5.4 Construction of the acquisition function:

[0350] Define a feasible sample set and the current optimal value, construct a collection function that integrates the expected improvement amount and the joint feasible probability, and realize the optimization direction guidance under constraints.

[0351] Let the set of sample indices in the current training dataset that satisfy all constraints be:

[0352] (5.11)

[0353] in, The set of sample indices that satisfy all constraints; i is the index of the training sample.

[0354] when When the optimal feasible objective function value is defined as:

[0355] Based on this, for any candidate point z, the expected improvement is defined as:

[0356]

[0357] (5.13)

[0358] In the formula, Let z be the predicted mean and predicted standard deviation of the objective function at candidate point z. This represents the standardized improvement amount of the objective function; Let z be the expected improvement amount for candidate point z; and These are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0359] (5.12)

[0360] in, This represents the current optimal feasible objective function value.

[0361] Meanwhile, the joint feasible probability of a candidate point satisfying all security constraints is defined as:

[0362] (5.14)

[0363] in, Let z be the joint feasible probability of a candidate point satisfying all safety constraints. , Let z be the predicted mean and predicted standard deviation of the constraint function at the r-th candidate point z. This is the cumulative distribution function of the standard normal distribution.

[0364] Therefore, the acquisition function for constrained Bayesian optimization is defined as:

[0365] (5.15)

[0366] in, To constrain the Bayesian optimization of the acquisition function value.

[0367] When there are no feasible samples in the current training dataset, i.e. If the feasible region is searched first, the acquisition function can be rewritten as follows:

[0368] (5.16)

[0369] 5.5 Batch candidate sample generation:

[0370] It adapts to multi-core parallel computing resources, generates batches of candidate samples with good dispersion, and improves the optimization efficiency of each iteration.

[0371] To utilize multi-core parallel computing resources, a maximum of q candidate samples are selected for parallel simulation evaluation in each iteration (the batch sample number q is determined based on available parallel computing resources and the time consumed in a single one-dimensional transient flow simulation. In one embodiment, when the computing device has 8 to 16 cores of parallel computing resources, q is 4-8, preferably q=6).

[0372] In the k-th iteration, first in [0,1] d Internally generated N cand Sobol candidate points

[0373]

[0374] In the formula, 1000 is used to ensure sufficient coverage of the candidate pool in the low-dimensional decision variable space, and 100d is used to increase the number of candidate points as the dimension of the decision variables increases. For example, when d=12, N is taken as... cand =1200; when d=30, take N. cand =3000.

[0375] First, calculate C. k Collect the function value of each candidate point and select the first candidate point:

[0376] (5.17)

[0377] Among them, Ck This is the set of candidate Sobol points generated in this round; To collect function values.

[0378] To obtain the remaining q-1 candidate points in the same batch, a sequential pseudo-observation update strategy is adopted. That is, after the b-1th candidate point has been selected, the predicted mean of that point is used as the temporary pseudo-observation value within the batch, which is used to construct the temporary acquisition function and the local penalty term.

[0379]

[0380] (5.18)

[0381] in, , For the pseudo-observation target value and pseudo-observation constraint value of the candidate point of the k-th round (b-1); This is the (b-1)th normalized candidate point selected in the k-th round.

[0382] The aforementioned pseudo-observations are only used to construct temporary surrogate models or temporary acquisition functions within the same batch. To reduce the clustering of candidate points in the same area, the pseudo-observations are not considered as real simulation results and are not added to the final training dataset. Only samples evaluated through actual simulation using the one-dimensional transient flow model are used as real samples to fill the training dataset. After temporarily adding pseudo-observations, a local penalty function is used to attenuate the acquisition function values ​​in the vicinity of the selected candidate points, constructing a temporary acquisition function corresponding to the b-th candidate point. :

[0383] (5.19)

[0384] in, This is the original acquisition function calculated based on the current real training dataset; This refers to the b'-th normalized candidate point selected in the k-th round; This is a local penalty function used to reduce the probability that a neighboring region of a selected candidate point will be selected again. The local penalty function can be:

[0385] (5.20)

[0386] In the formula, The local penalty radius is determined by the spatial scale of the normalized variables and the batch sample dispersion requirements. This varies depending on the candidate points. Approaching the selected candidate point , This reduces the temporary acquisition function value in the area; when the two are far apart... A value close to 1 has little impact on the acquisition function.

[0387] To further avoid excessive concentration of batch samples, a normalized Euclidean distance constraint is introduced:

[0388] (5.21)

[0389] In the formula, d is the dimension of the normalized decision variable; The minimum distance threshold in the normalized variable space can be taken as follows in one embodiment. Preferred selection If no candidate point in the candidate pool satisfies the minimum distance constraint, the constraint is gradually relaxed. Alternatively, the candidate point with the highest temporary acquisition function value can be selected from the never-repeated candidate points.

[0390] Therefore, the b-th candidate point is selected according to the following formula:

[0391] (5.22)

[0392] After all batch candidate points are selected, a one-dimensional transient flow model is called for parallel simulation evaluation. After the simulation is completed, only the objective function value and constraint function value obtained from the actual simulation are backfilled into the training dataset for the next round of Gaussian process surrogate model update.

[0393] 5.6 Parallel Simulation Evaluation and Data Backfilling

[0394] Complete the simulation evaluation and data backfilling of batch candidate samples, update the training dataset and retrain the surrogate model to improve the model's prediction accuracy.

[0395] The {z} selected in this round k,b} b=1 q Inverse normalization to the actual scheduling parameters {x k,b} b=1 q Then, for each candidate scheduling scheme, perform the following steps:

[0396] 1. Generate the control trajectory according to step S3;

[0397] 2. Call step S2 to perform a one-dimensional transient flow simulation;

[0398] 3. Calculate the constraint function values;

[0399] 4. Calculate the comprehensive objective function value;

[0400] 5. If the simulation fails, the objective function and constraint function will be uniformly penalized according to formula (5.5).

[0401] After evaluating q candidate points in this round, samples that duplicate the normalized decision variables of existing samples in the training dataset are removed, and the results of the remaining samples are added to the training dataset.

[0402] (5.23)

[0403] in, The actual objective function value of the b-th candidate point in the k-th round; The constraint function values ​​for the b-th candidate point in the k-th round; The updated training dataset is stored with normalized decision variables as unique identifiers. This is the set of sample indices that have completed the real simulation evaluation in the kth round and have not duplicated existing samples.

[0404] Based on the updated training dataset, the Gaussian process surrogate model of the objective function and constraint function is retrained, and the next iteration begins.

[0405] 5.7 Termination Conditions and Result Output:

[0406] Set quantitative iteration termination conditions, select the optimal feasible scheduling scheme and complete the final verification, output key engineering indicators, and form a complete optimization closed loop.

[0407] The constrained Bayesian optimization loop terminates when any of the following conditions are met:

[0408] 1. The cumulative number of high-fidelity simulations reaches the maximum number of evaluations, N. max ;

[0409] 2. Continuous N stall In each iteration, the relative improvement rate of the current optimal feasible objective function value is less than the threshold ε.

[0410] The relative improvement rate is defined as:

[0411] (5.24)

[0412] in, Let the relative improvement rate of the objective function in the k-th round be denoted as . represents the optimal feasible objective function values ​​for round k and round (k-1).

[0413] In a preferred embodiment, the maximum number of high-fidelity simulations N max =100, N is the number of consecutive iterations without effective improvement. stall =10, relative improvement rate threshold ε=10 -3 (The above termination parameters are adjusted according to the optimization accuracy requirements).

[0414] After optimization terminates, if feasible samples exist, the sample that satisfies all safety constraints and minimizes the comprehensive objective function value will be selected as the optimal joint scheduling scheme.

[0415] (5.25)

[0416] in, To constrain the final updated training dataset after the termination of the Bayesian optimization iteration; This is the actual scheduling parameter vector corresponding to the i-th sample in the dataset; , These are the constraint function value and the comprehensive objective function value for the i-th sample, respectively.

[0417] If multiple feasible samples have the same or approximately the same comprehensive objective function value, a second screening is performed according to the project priority; samples with smaller total water waste are selected first. If the total water waste is the same or approximately the same, samples with shorter total action time are further selected. If they are still the same, samples with shorter total waiting time are selected.

[0418] And on Step S2 is called again for final verification. The verification standard is to verify whether the scheme meets all the engineering safety constraints defined in step S3 (constraints such as piezometric head, free liquid level, and pump reverse speed). The corresponding minimum piezometric head, maximum piezometric head, lowest water level, highest water level, total water wastage, total action time, total waiting time, and comprehensive objective function value are output.

[0419] If the verification fails, then remove data from the training dataset. The proposed solution is removed from the feasible sample, and the suboptimal sample that satisfies all safety constraints and minimizes the comprehensive objective function value is selected as the new solution. The above final verification process is repeated until the optimal joint scheduling scheme that has passed verification is obtained.

[0420] If no feasible sample satisfies all safety constraints when optimization terminates, the candidate solution with the least degree of constraint violation, the corresponding violation constraint item, and suggested adjustment information will be output. The maximum number of evaluations can be automatically increased, the initial sample can be regenerated, or a suggestion to relax non-safety operation constraints can be made. Infeasible samples must not be output as the optimal executable solution.

[0421] Step S6, Result Generation and Output:

[0422] After step S5 is terminated, based on the complete set of evaluated samples formed during the optimization process, feasible samples that satisfy all security constraints are extracted, and the verification results and visualization results of the optimal joint scheduling scheme are output.

[0423] For the optimal joint scheduling scheme Decoding allows us to obtain the specific scheduling parameters for each pump, gate, and valve.

[0424] For the m-th adjustable unit, the output content should include at least:

[0425] Unit number and type;

[0426] Earliest allowed action time t m ear ;

[0427] Actual start time of action ;

[0428] Action duration ;

[0429] For gates or valves, the opening degree u before output action m,0 Target opening degree u m,1 and control trajectory ;

[0430] For a water pump, the operating state or speed n before the output action. m,0 Target operating state or rotational speed n m,1 and control trajectory .

[0431] This results in an optimal joint scheduling parameter table, which can be directly used as input for scheduling execution instructions or automatic control systems.

[0432] After outputting the optimal joint scheduling scheme, the one-dimensional transient flow model from step S2 is called again to... Perform full-time final verification calculations to obtain the time history of the piezometer head at each node. Flow time history of each pipe section Time history of water levels in various free liquid surface structures The actual pump speed time history and the overflow flow time history of each overflow facility. .

[0433] This results in the final verification table of the optimal joint scheduling scheme, which includes at least the minimum piezometric head, the maximum piezometric head, the minimum water level, the maximum water level, the total water wastage, the total action time, the total waiting time, the comprehensive objective function value, and various safety margins.

[0434] To facilitate engineers in comparing and analyzing different solutions, this step also outputs visualization results corresponding to the optimal joint scheduling scheme and the set of alternative schemes. The visualization results include at least one or more of the following:

[0435] 1. Time history curves: time history of piezometric head at nodes, time history of free liquid level, time history of overflow flow, and time history of flow in key pipe sections;

[0436] 2. Envelope diagrams: minimum piezometric head envelope diagram along the route, maximum piezometric head envelope diagram along the route, and water level fluctuation envelope diagram for key structures;

[0437] 3. Parameter-Response Relationship Diagram: A scatter plot or contour plot showing the relationship between scheduling parameters and key indicators such as total water discharge, minimum piezometric head, and highest water level;

[0438] 4. Feasible region distribution map: Represents the distribution of regions that meet safety constraints and infeasible regions;

[0439] The output format of this step includes, but is not limited to:

[0440] Structured data files: used to store full-time data such as piezometric head, water level, flow rate, and overflow.

[0441] Parameter table and verification table: used to store the scheduling parameters and safety verification results of the optimal and alternative schemes;

[0442] Graphical files: used for results presentation, project reporting, and solution comparison;

[0443] List of dispatch instructions: used to directly guide the execution of manual dispatching or automatic control systems.

[0444] The effects of the above results are as follows:

[0445] 1. Direct execution function: The optimal joint scheduling parameter table and scheduling instruction list can be directly used to guide the start-up and shutdown of water pumps, the opening and closing of gates, and the regulation of valves;

[0446] 2. Safety verification function: The final verification table and safety margin results can be used to confirm whether the plan meets the engineering safety requirements;

[0447] 3. Process traceability: Structured data files and graphical files can be used for process optimization review, operational event analysis, and emergency response drills;

[0448] 4. Engineering integration role: The results can be output to existing scheduling decision support platforms, monitoring platforms or digital twin platforms through file interfaces or programming interfaces.

[0449] Example 2:

[0450] This embodiment presents a pump-gate-valve joint scheduling strategy system for a long-distance water conveyance system. Based on the optimization method described in Embodiment 1, it is used for a constrained Bayesian optimization system for scheduling strategies in long-distance water conveyance tunnels. It mainly consists of the following functional modules, which can be integrated into the same computing platform in software form or deployed in a distributed manner via a network or bus:

[0451] (1) Strategy and Sampling Module:

[0452] The strategy and sampling module is used to uniformly organize and initially set the control parameters such as the opening and closing times and durations of the i-th gate and valve. Based on the parameter value range and dispersion length given by the engineering project, this module automatically generates candidate sets of time and opening parameters, constructs a parameter space for optimization, and generates an initial sample point set. This module outputs a candidate scheduling strategy vector and a list of samples that need to be evaluated in parallel in each iteration to the simulation scheduling module.

[0453] (2) Agent learning module:

[0454] The surrogate learning module is used to establish a probabilistic surrogate model for the comprehensive cost objective and safety constraints such as piezometric head and water surface fluctuations in long-distance water conveyance systems. This module receives observational data such as the objective function value, minimum piezometric head, maximum piezometric head, and vertical shaft water surface amplitude under various sample operating conditions returned by the simulation scheduling module. It automatically completes kernel function type selection, hyperparameter estimation, and model updates using methods such as Gaussian process regression, and outputs the predicted mean and uncertainty estimate for unevaluated strategy points. This module provides the constraint acquisition module with probability distribution information for both the objective and constraints.

[0455] (3) Constraint Acquisition Module:

[0456] The constraint acquisition module is used to calculate acquisition indicators and select the next batch of evaluation samples based on the surrogate model. This module receives the objective function and the predicted distributions of each constraint function output by the surrogate learning module. Based on the feasibility-weighted expected improvement concept, it calculates the expected improvement amount of each candidate point and the joint probability of satisfying all safety constraints, and couples the two to obtain a comprehensive acquisition value. The module searches for several points with high acquisition values ​​in the parameter space, uses them as the gate valve scheduling strategy for the next round of parallel evaluation, and outputs them to the strategy and sampling module and the simulation scheduling module, enabling focused exploration of high-value regions within the feasible domain under a limited simulation budget.

[0457] (4) Simulation model scheduling module:

[0458] The simulation model scheduling module interfaces with the one-dimensional transient flow simulation engine and enables parallel computation and anomaly handling for multiple candidate scheduling strategies. This module receives scheduling strategy vectors, system boundary conditions, and initial hydraulic states from the strategy and sampling module. It automatically generates input files for each candidate strategy, calls the one-dimensional unsteady flow numerical model for transient calculations, and obtains indicators such as minimum and maximum piezometric head along the flow path, water surface fluctuations in each shaft and forebay, and water discharge. The module supports batch initiation of simulation tasks at a preset parallelism, records and controls retrying for samples that fail or have missing results, and uniformly returns the target and constraint observations of valid samples to the agent learning module and the data and visualization module.

[0459] (5) Data and Visualization Module:

[0460] The data and visualization module records optimization process data and supports result analysis and display. This module comprehensively records the scheduling strategy, objective function value, safety constraints, and computation time for each evaluation, forming a traceable test log. It can also generate, as needed, minimum piezometric head distribution maps, wastewater volume contour maps, 3D scatter plots of comprehensive costs, and feasible region distribution maps on the gate closure time parameter plane. Through the visualization interface, operators can intuitively understand the distribution of scheduling strategies near the safety constraint boundaries and the optimization iteration convergence process, providing auxiliary decision-making basis for engineering applications and scheme comparison.

[0461] (6) Interface and configuration module:

[0462] The interface and configuration module provides a unified parameter configuration and system call interface, enabling the portability and integration of the method of this invention across different engineering projects and simulation platforms. This module includes an engineering parameter configuration interface, a model coupling interface, and an operation control interface. Users can set the geometric parameters, roughness coefficient, pump and gate valve layout information, and safety thresholds of the long-distance water conveyance system, configure optimization target weights, total number of samples, single-round parallel scale, and other operating parameters, and interface with existing one-dimensional transient flow simulation programs through file interfaces or programming interfaces. This module enables flexible deployment and engineering application of the entire system, facilitating the integration of this invention into existing scheduling decision support platforms.

[0463] The above description is only used to illustrate the technical solutions of the present invention and is not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention (such as the application of various formulas, the order of steps, etc.) without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing the joint scheduling strategy of pumps and gate valves in a long-distance water conveyance system, characterized in that, Includes the following steps: Step 1: Establish a one-dimensional transient flow model of the water conveyance system. The input of the model includes pipe geometric hydraulic parameters, hydraulic component parameters, boundary conditions and scheduling control trajectory. The output includes node piezometric head, pipe section flow rate, free liquid surface structure water level and overflow flow rate. Step 2: Set up a set of controllable units participating in the joint scheduling, wherein the controllable units include water pumps, gates, and valves; Construct control constraints for the controllable units, and define the waiting time of each controllable unit and the total system waiting time; Different control methods are adopted for different types of adjustable units; Integrate all scheduling parameters to be optimized to form a joint scheduling decision vector; The gate or valve opening trajectory, pump speed trajectory or start / stop state trajectory formed after decoding the joint scheduling decision vector, together with the initial system conditions, boundary conditions, evaluation period and calculation time step, are used as the scheduling input of the one-dimensional transient flow model to generate the transient flow response under the corresponding conditions. Extreme values ​​are extracted based on the transient flow response, and safety constraints are constructed, including lower limit constraint of piezometric head, upper limit constraint of piezometric head, lower limit constraint of free liquid level, upper limit constraint of free liquid level, and pump speed constraint. Step 3: Construct a comprehensive objective function, which is obtained by normalizing and weighting the total water wastage, total action duration, and total waiting time. Step 4: Use the Gaussian process proxy model to predict the comprehensive objective function value and security constraint function value of the unevaluated joint scheduling scheme, and select the candidate joint scheduling scheme that needs to be evaluated by calling the one-dimensional transient flow model for high-fidelity simulation through the acquisition function. A data acquisition function is constructed that integrates the expected improvement amount and the joint feasible probability, and the optimization direction is guided by the data acquisition function under engineering safety constraints. A batch candidate sample generation strategy with minimum distance constraints is adopted, and the joint scheduling scheme is iteratively optimized by combining multi-core parallel computing. Step 5: After the optimization iteration terminates, select feasible samples that satisfy all security constraints from all evaluated samples, and take the feasible sample with the smallest comprehensive objective function value as the candidate optimal joint scheduling scheme. Decode the specific scheduling parameters of each controllable unit and generate the optimal joint scheduling parameter table. Call the one-dimensional transient flow model again to perform full-time verification calculation on the optimal joint scheduling scheme. If the verification passes, output the optimal joint scheduling scheme, full-time calculation data and final verification result table.

2. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 2, the control unit controls the constraints to satisfy the formula: In the formula, Let m be the earliest allowed action time of the m-th adjustable unit. This is the actual start time of the action. For the duration of the action, The moment the action ends; The maximum allowable delay time for the m-th adjustable unit relative to the earliest allowed action time; and These are its shortest and longest allowed action durations, respectively; The waiting time of the m-th adjustable unit is calculated using the following formula: The total waiting time of all adjustable units in the system is calculated using the following formula: In the formula, M represents the total number of adjustable units. Let x be the total waiting time, and let x be the joint scheduling decision vector.

3. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 2, a corresponding control method is adopted for different types of adjustable units. When the m-th adjustable unit is a gate or valve, its control quantity adopts the relative opening degree u. m (t)∈[0,1] represents the piecewise linear control method of "constant opening degree - linear change - constant opening degree", and the calculation formula is as follows: In the formula, u m,0 and u m,1 These represent the target opening before and after the action, respectively. The moment of the start of the action. The moment the action ends; When the m-th adjustable unit is a water pump, its control process is represented by its speed or operating state; for a variable speed water pump, its speed control function n is defined. m (t) is: In the formula, and These are the rotational speeds before and after the action, respectively. The moment of the start of the action. The moment the action ends; For shutdown conditions, the pump does not use the target speed after activation as the optimal control variable, but uses the shutdown command time, the start time of outlet valve closure, and the duration of outlet valve closure as scheduling parameters. During shutdown, the actual pump speed is obtained by solving the pump's full characteristic curve, unit inertial parameters, outlet valve closure process, and transient flow equations of upstream and downstream pipelines simultaneously. The actual speed is allowed to decay to zero during shutdown and may become negative under the backflow effect of water flow. A negative value indicates that the pump is running in reverse.

4. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 2, the joint scheduling decision vector includes the start time of action of each controllable unit, the duration of action, and the normalized representation of the continuous control variables of each pump, and its expression is: In the formula, M represents the total number of adjustable units. The number of continuous control variables for the water pump included in the optimization; t s,m This is the start time of the action of the m-th adjustable unit; The duration of the action of the m-th adjustable unit; Normalized representation of the continuous control variables for each water pump; The continuous control variable of the water pump is mapped to the actual control quantity using the following formula: In the formula, , For the first The actual speed of each water pump The first The minimum and maximum speeds allowed for each pump, for pumps that only perform preset start-stop and do not participate in continuous regulation, their corresponding continuous control variables are not included in the joint scheduling decision vector. The formula is used for continuous speed regulation or start-up speed regulation conditions.

5. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 2, the preset allowable values ​​for the lower limit constraint and the upper limit constraint of the piezometric head are calculated using the following formula: In the formula, and These are the minimum and maximum permissible piezometric head for the j-th pressure monitoring section or node, respectively. This is the elevation of the top of the pipe at this cross-section; To prevent adverse negative pressure from occurring; This is the elevation of the pipeline centerline at this node; The allowable internal pressure of this cross section; ρ is the density of water; g is the acceleration due to gravity; This is a positive pressure safety margin; The preset allowable values ​​for the lower and upper limits of the free liquid surface constraint are calculated using the following formula: In the formula, The maximum permissible water level, The minimum allowable water level, The control elevation for the top of the structure; For top safety margin; The elevation of the base slab of the structure; This is to ensure a safety margin at the bottom; This is the minimum operating water level specified in the operating procedures.

6. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 3, the construction method of the comprehensive objective function includes: Calculate the total action duration and total waiting time based on the action timing parameters of each adjustable unit; A normalized scale is determined for the total water discharge, total operation time, and total waiting time, and this normalized scale remains unchanged during the optimization process. The normalized scale for the total water discharge is determined as follows: it is determined by multiplying the 100% reference overflow capacity of each overflow facility at the highest permissible operating water level by the evaluation duration, with a non-zero lower limit set to avoid a zero denominator. The specific formula is as follows: In the formula, The lower limit value is non-zero. For the evaluation period, This represents the total number of overflow facilities in the system that may cause water spillage. This represents the reference overflow capacity of the k-th overflow facility; Based on the normalized scale, the total water wastage, total action duration, and total waiting time are converted into normalized targets, and the calculation formula is as follows: In the formula, To normalize the total wastewater volume, To normalize the total motion duration, This represents the normalized total waiting time; a smaller value indicates better performance of the joint scheduling scheme. All are dimensionless targets; in, The value of the dimensionless comprehensive objective function; , These are the weighting coefficients.

7. The method for optimizing the joint scheduling strategy of pumps and gate valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 4, the modeling process of the Gaussian process proxy model is as follows: based on the sample set that has completed the one-dimensional transient flow simulation evaluation, the normalized joint scheduling decision variables are used as inputs, and the comprehensive objective function value and each safety constraint function value are used as outputs to establish the objective function Gaussian process proxy model and the constraint function Gaussian process proxy model. The objective function Gaussian process surrogate model is used to predict the mean and standard deviation of the comprehensive objective function of the candidate joint scheduling scheme, and the constraint function Gaussian process surrogate model is used to predict the mean and standard deviation of each security constraint function corresponding to the candidate joint scheduling scheme. The predicted mean and standard deviation of the comprehensive objective function, as well as the predicted mean and standard deviation of each safety constraint function, are input into the acquisition function to select candidate joint scheduling schemes that require high-fidelity simulation evaluation using a one-dimensional transient flow model.

8. The method for optimizing the joint scheduling strategy of pumps and gate valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 4, the acquisition function is constructed as follows: When there are feasible samples in the training dataset that satisfy all engineering safety constraints, the acquisition function is: When there are no feasible samples in the training dataset that satisfy all engineering safety constraints, the acquisition function is: in, , Let z be the expected improvement amount for candidate point z. Let z be the joint feasible probability of a candidate point z satisfying all engineering safety constraints. Let be the predicted mean and predicted standard deviation of the r-th constraint function at candidate point z, respectively; R be the total number of constraints; and Φ(·) be the cumulative distribution function of the standard normal distribution.

9. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, In step 4, the batch candidate sample generation strategy for the minimum distance constraint is as follows: In each iteration, no more than q candidate samples are selected for parallel simulation evaluation; in the k-th iteration, Sobol sequences or other quasi-random sequences are used in the normalized variable space. Internally generated candidate point set Based on the collected function value Select the first candidate point ; When selecting other candidate points sequentially, a triple mechanism of pseudo-observation update, local penalty function and minimum distance constraint is adopted; Pseudo-observation update: The predicted mean of the selected candidate points is used as a temporary pseudo-observation value to construct a temporary surrogate model; Local penalty function: Constructing a temporary acquisition function The collected function values ​​in the vicinity of the selected point are attenuated; Minimum distance constraint: The candidate points must satisfy the normalized Euclidean distance constraint. ; The b-th candidate point satisfies the constraints and makes... The largest; If no feasible points are found, the restrictions should be gradually relaxed. False observations are not included in the training dataset. in, To preset batch size, Let the dimension of the decision variables be... For the preset distance threshold, This is the candidate sample number.

10. The method for optimizing the joint scheduling strategy of pumps, gates, and valves in a long-distance water conveyance system according to claim 1, characterized in that, Step 5 also includes re-verifying the selected joint scheduling candidate optimal schemes using one-dimensional transient flow high-fidelity simulation. If the verification fails, the scheme is eliminated, and a suboptimal sample that meets the engineering safety constraints is selected as a new candidate optimal scheme. The verification is repeated until the optimal scheme that passes the verification is obtained. The selection of the optimal scheme satisfies the following calculation formula: in, For the final updated training dataset; This is the actual scheduling parameter vector corresponding to the i-th sample in the dataset; , These are the constraint function value and the comprehensive objective function value for the i-th sample, respectively.

Images