A dam flood discharge emptying intelligent decision method for multi-objective optimization

Abstract

A smart decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization includes: collecting multi-dimensional data streams from the dam reservoir, preprocessing them to generate standard multi-dimensional data streams, and establishing a database; loading a distributed hydrological model to output the inflow process line of the dam; loading a one-dimensional unsteady flow dynamics model to output the predicted water level, flow velocity evolution process, and key results at each cross-section along the dam; defining decision variables X and objective functions, inputting the decision variables X into the objective functions to solve for a set of flood discharge and emptying scheduling schemes with different weights among the objectives; setting the relative weight vector W of the core objectives according to the current scheduling scenario, calculating the relative proximity C, and obtaining the final flood discharge and emptying scheduling scheme; this invention achieves the prediction of future hydrological conditions by constructing an intelligent decision-making framework that integrates multi-source data fusion and preprocessing, spatiotemporal evolution prediction, multi-objective Pareto front optimization, and collaborative decision-making and closed-loop execution.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for water conservancy projects, specifically to an intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization. Background Technology

[0002] As the core hub of water conservancy and hydropower projects, dams play an irreplaceable role in the safe and efficient operation of watershed flood control and disaster reduction, optimal allocation of water resources, supply of clean energy, and regional ecological balance. Within this macro-system, flood discharge and air release mechanisms are key links in regulating reservoir capacity and responding to extreme flood events.

[0003] In current technological practices, the decision support system for dam flood discharge largely relies on scheduling procedures or charts solidified from historical hydrological data and engineering experience. Specifically, technicians statistically analyze the probability of floods of different magnitudes based on years of hydrological observation data and take appropriate action in conjunction with the safe flood discharge capacity of the downstream river channel. For example, when the upstream inflow and reservoir water level reach a certain preset flood control threshold, the scheduling system will trigger the corresponding flood discharge plan, instructing operators or the automatic control system to open the designated spillway structures.

[0004] However, with the increasing frequency of extreme hydrological events due to global climate change, the inherent limitations of the aforementioned static rule-based scheduling methods are becoming increasingly apparent in addressing the complex and ever-changing scheduling scenarios. Modern reservoir scheduling deals with a continuously changing, uncertain, and multi-objective coupled complex dynamic system. Pre-set scheduling rules prioritize flood control safety, but sacrifice the optimization space for other objectives, failing to dynamically and precisely balance and optimize among multiple continuously changing and conflicting objective functions. Furthermore, existing technologies primarily rely on threshold judgments based on current monitoring data, representing a passive response mechanism lacking the ability to effectively predict and integrate future hydrological processes.

[0005] Therefore, it is necessary to design a dynamic and continuous multi-objective optimization scheduling decision-making technology based on the fusion of multi-dimensional dam data to achieve effective prediction and integration of future hydrological processes of the dam. Summary of the Invention

[0006] To address the technical problems of existing technologies that rely on static and discrete scheduling procedures, thus failing to dynamically and collaboratively optimize multiple conflicting objectives such as flood control safety, power generation benefits, and ecological needs under continuously changing and uncertain hydrological conditions, resulting in delayed decision-making response and limited optimization space, this invention provides an intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization, comprising: Collect multidimensional data streams from dams and reservoirs, preprocess them, generate standard multidimensional data streams, and establish a database. The multidimensional data streams include real-time monitoring data of the dam itself, real-time observation data from various rain gauges and hydrological stations within the basin, meteorological forecast data within the basin, electricity price prediction curves, power grid load demand curves, and flood discharge and release schedules to be optimized. Load a distributed hydrological model, which is used to predict the inflow process curve of the dam; input the meteorological forecast data into the distributed hydrological model and output the inflow process curve of the dam; A one-dimensional unsteady flow dynamics model is loaded, which is used to simulate and predict the evolution of water level and velocity at various cross-sections along the dam under a given scheduling scheme, and to obtain key results. The inflow process line of the dam and the flood discharge scheduling scheme to be optimized are input into the one-dimensional unsteady flow dynamics model, and the predicted evolution of water level and velocity at various cross-sections along the dam and key results are output. The key results include the upstream water level process line, the total outflow process line, and the total power generation process line. Define decision variable X and objective function. Based on the predicted evolution process of water level and flow velocity at each section along the dam and the key results, input the decision variable X into the objective function to solve and generate a set of flood discharge and release schedule schemes with different weights among the objectives. Based on the current scheduling scenario, a relative weight vector W for the core objectives is set, the relative proximity C is calculated, and the flood discharge scheduling scheme set is sorted to obtain the final flood discharge scheduling scheme; the core objectives include flood control safety, power generation benefits, and ecological protection.

[0007] Furthermore, a closed-loop correction mechanism is set up during the execution cycle of the final flood discharge and air release schedule; the closed-loop correction mechanism includes: Obtain the actual measured hydrograph WL of the water level upstream of the dam. actual Total discharge flow process line Q out,actual Compared with the predicted upstream water level hydrograph WL pred Total discharge flow process line Q out，pred The prediction error vector e is calculated, and the distributed hydrological model and the one-dimensional unsteady flow dynamics model are updated to form a closed-loop adaptive model; the execution cycle is 1 hour.

[0008] Furthermore, the distributed hydrological model is a digital elevation model, which is composed of square grid cells formed by discretizing the dam based on physical mechanisms; The process of independently calculating the inflow hydrograph for each grid cell and outputting the inflow hydrograph for the dam includes: The meteorological forecast data is input into the distributed hydrological model, and evapotranspiration and soil infiltration are calculated separately for each network unit; the meteorological forecast data includes air temperature, humidity, wind speed, and solar radiation; Based on the evaporation and transpiration and soil infiltration, it is determined whether the surface rainfall intensity exceeds the infiltration capacity. When the surface rainfall intensity exceeds the infiltration capacity, a two-dimensional diffused wave surface runoff simulation is performed based on the generated excess runoff to predict the surface runoff of each network unit. The runoff is calculated step by step to output the inflow process line of the dam.

[0009] Furthermore, the expression for the decision variable X is: ; in, Let be the percentage of the i-th gate that is open. Let T be the reference traffic of the j-th unit, and T be the optimization time domain. The decision step size is defined as n, where n represents the total number of gates, m represents the total number of generator sets, and t represents the time period.

[0010] Furthermore, the predicted evolution of water level and flow velocity at various cross-sections along the dam, and the output process of key results, include: The inflow process line output by the distributed hydrological model is used as the upstream boundary condition of the one-dimensional unsteady flow dynamics model; the flood discharge and air release scheduling scheme to be optimized is used as the downstream boundary condition of the one-dimensional unsteady flow dynamics model. The upstream and downstream boundary conditions are input into the one-dimensional unsteady flow dynamics model, and the flow evolution is carried out through the one-dimensional Saint-Venant equations. The predicted evolution process of water level and flow velocity at each section along the dam and the key results are output.

[0011] Furthermore, the distributed hydrological model and the one-dimensional unsteady flow dynamics model are cascaded to form a spatiotemporal evolution prediction method from meteorological forecasting to watershed runoff generation to reservoir scheduling and state response.

[0012] Furthermore, the objective function includes: a flood control safety objective function. Power generation benefit objective function Ecological protection objective function and engineering health objective function ; The flood control safety objective function The expression is: ; Where WL(t) is the predicted hydrograph of the water level upstream of the dam during time period t, WL flood WL is the flood control limit water level process line. checkTo verify the flood level hydrograph, w1 and w2 are weighting coefficients, and w1 >> w2; The power generation efficiency objective function The expression is: ; Where Price(t) is the predicted time-of-use electricity price for period t, Power(t) is the predicted total power generation for period t, and Cop is the gate operation cost coefficient. Let be the percentage of the i-th gate that is open. For decision-making step size; The ecological protection objective function The expression is: ; Among them, Q out (t) represents the predicted total discharge flow during time period t, Q eco_min (t) represents the minimum ecological water demand flow rate specified downstream during time period t; The project health objective function The expression is: ; Where, ΔQ max This represents the maximum permissible rate of change in downstream flow rate per unit time.

[0013] Furthermore, the method for calculating the relative proximity C includes: Obtain the solution value of the objective function, construct the decision matrix, perform normalization processing, and calculate the weighted normalized decision matrix; Determine the positive and negative ideal solutions of the weighted normalized decision matrix, and calculate the Euclidean distance D+ from the solution of the objective function to the positive ideal solution and the Euclidean distance D- to the negative ideal solution; the positive ideal solution is the optimal solution for each indicator; the negative ideal solution is the worst solution for each indicator. Obtain the relative closeness C, and the expression for calculating the relative closeness is: .

[0014] Furthermore, the prediction error vector e includes the prediction error e of the upstream water level hydrograph. WL Total discharge flow process curve prediction error e Q ; The prediction error of the upstream water level hydrograph e WL The calculation expression is: ; The total outflow process curve prediction error e Q The calculation expression is: .

[0015] Furthermore, the closed-loop adaptive formation process includes the following: In the closed-loop correction mechanism, the prediction error vector e is input into the Kalman filter, and the parameters of the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model are corrected by modifying the estimated values ​​of the state variables of the Kalman filter, thereby updating the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model; the state variables of the Kalman filter are the key physical parameters in the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model.

[0016] The beneficial effects of this invention are as follows: By constructing an intelligent decision-making framework that integrates multi-source data fusion and preprocessing, spatiotemporal evolution prediction, multi-objective Pareto front optimization, and collaborative decision-making and closed-loop execution, this invention enables the prediction of future hydrological conditions and seeks a dynamic optimal balance across multiple dimensions such as flood control and power generation. This breaks through the fundamental limitations of traditional scheduling procedures in handling complex, dynamic, and multi-objective problems, enabling dam flood discharge scheduling to leap from a passive response mode based on historical experience to an active optimization mode based on future prediction. Under the premise of ensuring absolute safety, this invention can finely balance multiple interests such as power generation and ecology, significantly improving the comprehensive operational efficiency and intelligent management level of reservoirs. Attached Figure Description

[0017] Figure 1 This is a flowchart of the intelligent decision-making method for dam flood discharge and emptying oriented to multi-objective optimization provided by the present invention; Figure 2 This is a logical schematic diagram of the spatiotemporal evolution prediction provided by the present invention; Figure 3 This is a flowchart of the hybrid adaptive multi-objective particle swarm optimization algorithm provided by the present invention. Detailed Implementation

[0018] The technical solution of the present invention is further described below, but the scope of protection is not limited to what is described.

[0019] This invention provides an intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization, such as... Figure 1 As shown, it includes: Step S100: Collect multidimensional data streams from the dam and reservoir, preprocess them, generate standard multidimensional data streams, and establish a database; the multidimensional data streams include real-time monitoring data of the dam itself, real-time observation data from various rain gauges and hydrological stations within the basin, meteorological forecast data within the basin, electricity price prediction curves and power grid load demand curves, and flood discharge and release schedules to be optimized. The system connects to the data server of the Supervisory Control and Data Acquisition (SCADA) system via an industrial Ethernet interface, following the Modbus / TCP protocol, and collects real-time monitoring data of the dam body at a pre-set frequency. This data includes: the reservoir water level measured by an ultrasonic level gauge; the opening height of each spillway gate and surface gate measured by an absolute rotary encoder installed on the gate hoist drive shaft; and the actual total discharge flow measured by an acoustic Doppler current profiler (ADCP) installed on the tailrace channel cross-section.

[0020] Through a wide area network fiber optic link and a data encryption gateway, a connection is established with the central station of the upstream basin hydrological telemetry system. Following relevant standards, real-time observation data from various rain gauges and hydrological stations within the basin are acquired at a set frequency.

[0021] By connecting to the Internet and using the HTTP protocol GET request method, the numerical weather forecast product publishing API of the meteorological data center is accessed periodically to obtain gridded meteorological forecast data covering the basin where the dam is located for the next 72 hours, with a spatial resolution of 3 kilometers and a temporal resolution of 1 hour. The meteorological forecast data includes precipitation, temperature, wind speed, relative humidity and total solar radiation.

[0022] The system connects to the data exchange front-end unit of the regional power grid dispatch center via a communication interface based on the IEC 60870-5-104 protocol, and obtains the time-of-use electricity price forecast curve and power grid load demand curve for the next 24 hours at a set frequency.

[0023] By reading the flood discharge and air release schedule plan to be optimized, which is issued by relevant departments and clearly stipulates the minimum ecological flow threshold that the downstream section needs to maintain within a specific period.

[0024] For the time series data in the standard multidimensional data stream, an algorithm based on the three-standard-deviation (3σ) criterion is applied to identify and mark outliers. For the marked outliers and missing data points, cubic spline interpolation is used to fill in the gaps to ensure the integrity of the data sequence. Subsequently, to eliminate the misalignment problem of timestamps between different data sources, all time series data are resampled to a unified, fixed 15-minute time base using linear averaging. Finally, to facilitate subsequent processing by neural network and other models, all numerical data are linearly scaled to the closed interval [0,1] using min-max normalization. The processed data is structured and stored in a time series database, with metadata tags attached to identify the data source, type, and processing time, thus completing the database establishment. The database also includes discharge capacity curves and unit output curves.

[0025] Step S200: Load the distributed hydrological model, which is used to predict the inflow process of the dam; input the meteorological forecast data into the distributed hydrological model and output the inflow process of the dam. The distributed hydrological model is a digital elevation model, which is composed of square grid cells formed by discretizing the dam based on physical mechanisms. The process of independently calculating the inflow process curve for each grid cell and outputting the inflow process curve for the dam includes: inputting the meteorological forecast data into the distributed hydrological model, and calculating evapotranspiration and soil infiltration for each network cell separately; the meteorological forecast data includes air temperature, humidity, wind speed, and solar radiation; the meteorological forecast data is such as NWP meteorological forecast data; Based on the evaporation and transpiration and soil infiltration, it is determined whether the surface rainfall intensity exceeds the infiltration capacity. When the surface rainfall intensity exceeds the infiltration capacity, a two-dimensional diffused wave surface runoff simulation is performed based on the generated excess runoff to predict the surface runoff of each network unit. The runoff is calculated step by step to output the inflow process line of the dam.

[0026] The square grid unit is 1 km × 1 km; The calculation of evaporation and transpiration was performed using the Penman-Monteith formula. The calculation of soil infiltration adopts the Green-Ampt model that takes into account the soil's early moisture content, and the required soil parameters (such as saturated hydraulic conductivity and matrix potential) are obtained from existing basic soil databases.

[0027] When surface rainfall intensity exceeds infiltration capacity, the resulting excess runoff is calculated using a two-dimensional diffusion wave equation on a slope field defined by a digital energy model (DEM) to simulate surface runoff formation. The runoff generated by each grid cell is channeled along a pre-extracted digital river network using a river channel calculation model based on the Muskingum-Cunge method, ultimately forming an inflow process line at the reservoir inlet section that predicts the inflow over the next 72 hours with a time resolution of 1 hour. The entire calculation process is driven by meteorological forecast data, such as... Figure 2 As shown.

[0028] Step S300: Load a one-dimensional unsteady flow dynamics model. This model is used to simulate and predict the evolution of water level and velocity at various cross-sections along the dam under a given scheduling scheme, as well as key results. Input the inflow process curve of the dam and the flood discharge scheduling scheme to be optimized into the one-dimensional unsteady flow dynamics model, and output the predicted evolution of water level and velocity at various cross-sections along the dam, as well as key results. The key results include the upstream water level process curve, the total outflow process curve, and the total power generation process curve. The predicted evolution of water level and flow velocity at various cross-sections along the dam, and the output process of key results, include: The inflow process line output by the distributed hydrological model is used as the upstream boundary condition of the one-dimensional unsteady flow dynamics model; the flood discharge and air release scheduling scheme to be optimized is used as the downstream boundary condition of the one-dimensional unsteady flow dynamics model. The upstream and downstream boundary conditions are input into the one-dimensional unsteady flow dynamics model, and the flow evolution is carried out through the one-dimensional Saint-Venant equations. The predicted evolution process of water level and flow velocity at each section along the dam and the key results are output.

[0029] The distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model are cascaded to form a spatiotemporal evolution prediction method from meteorological forecasting to watershed runoff generation to reservoir scheduling and state response. The data stream of the spatiotemporal evolution prediction is as follows: Figure 2 As shown.

[0030] The one-dimensional unsteady flow hydrodynamic model longitudinally divides the reservoir area from the dam's front to the reservoir's tail along the centerline of the main channel into 500 computational cross-sections. On these cross-sections, the evolution of the flow follows the one-dimensional Saint-Venant equations, namely the continuity equation and the momentum equation. These equations are numerically solved using the Preissmann implicit finite difference method with a four-point central difference scheme. The upstream boundary condition of the one-dimensional unsteady flow hydrodynamic model is the inflow process line output by the distributed hydrological model. The downstream boundary condition of the one-dimensional unsteady flow hydrodynamic model is a discharge flow or dam-front water level process defined by the flood discharge scheduling scheme to be optimized. The flood discharge scheduling scheme to be optimized is represented as a time series vector, specifically defining the percentage of gate opening of each flood discharge structure (such as deep discharge orifices, surface orifices, and power generation diversion tunnels) or the flow rate referenced by the generator units at each 1-hour step Δt within the next 72 hours.

[0031] For flood discharge structures, the functional relationship between their discharge capacity and the water level in front of the dam and the gate opening is determined by the discharge capacity curve that has been calibrated in advance through physical model tests and stored in the database.

[0032] For generator units, the relationship between power generation and water head and flow rate is determined by the unit's output curve. By iteratively solving the Saint-Venant equations, the system accurately simulates the evolution of water level and flow velocity at various cross-sections along the reservoir over the next 72 hours under any given scheduling scheme, and outputs key results, including the upstream water level hydrograph, the total outflow hydrograph, and the total power generation hydrograph. The distributed hydrological model and the one-dimensional unsteady flow dynamics model are cascaded to form a complete system, realizing spatiotemporal evolution prediction, as follows: Figure 2As shown, it can perform predictive simulations from weather forecasts to watershed runoff, reservoir scheduling, and state response.

[0033] Step S400: Define decision variable X and objective function. Based on the predicted evolution of water level and flow velocity at each section along the dam and the key results, input the decision variable X into the objective function for optimization and solution, and generate a set of flood discharge scheduling schemes with different weights among the objectives (i.e., Pareto optimal frontier). The expression for the decision variable X is: (1) in, Let be the percentage of the i-th gate that is open. Let T be the reference traffic of the j-th unit, and T be the optimization time domain. The decision step size is defined as n, where n represents the total number of gates, m represents the total number of generator sets, and t represents the time period.

[0034] The objective function includes: flood control safety objective function. Power generation benefit objective function Ecological protection objective function and engineering health objective function ; The flood control safety objective function The expression is: (2) Where WL(t) is the predicted hydrograph of the water level upstream of the dam during time period t, WL flood WL is the flood control limit water level process line. check To verify the flood level hydrograph, w1 and w2 are weighting coefficients, and w1 >> w2; The flood control safety objective function It punishes any dispatching behavior that exceeds the flood control limit level and imposes penalties on situations approaching the check flood level.

[0035] The power generation efficiency objective function The expression is: (3) Where Price(t) is the predicted time-of-use electricity price for period t, Power(t) is the predicted total power generation for period t, and Cop is the gate operation cost coefficient, which is used to penalize frequent gate adjustments in order to extend equipment life. Let be the percentage of the i-th gate that is open. For decision-making step size; The ecological protection objective function The expression is: (4) Among them, Q out (t) represents the predicted total discharge flow during time period t, Q eco_min (t) represents the minimum ecological water demand flow rate specified downstream during time period t; The ecological protection objective function This is to ensure that the downstream ecological base flow is met.

[0036] The project health objective function The expression is: (5) Where, ΔQ max It is the maximum allowable rate of change in downstream flow rate per unit time, used to avoid downstream bank erosion and hydraulic structure vibration caused by drastic flow changes.

[0037] The optimization solution employs a Hybrid Adaptive Multi-Objective Particle Swarm Optimization (HA-MOPSO) algorithm, the algorithm flow of which is as follows: Figure 3 As shown, the details are as follows: a. Initialization: Randomly generate a population containing N (e.g., N=200) particles (i.e., the complete 72-hour scheduling scheme X). Each particle has a position vector x. i and a velocity vector v i .

[0038] b. Evaluation: For each particle, call the spatiotemporal evolution prediction module to calculate its corresponding four objective function values.

[0039] c. Non-dominated sorting and archive update: Based on the calculated objective function value, perform fast non-dominated sorting and crowding calculation on the current population and all solutions. Add new non-dominated solutions to the archive and remove dominated solutions from the archive. If the archive size exceeds the upper limit, remove solutions from the most crowded regions based on crowding to maintain solution diversity.

[0040] d. Global Optimal Guide (gbest) Selection: For each particle i, a globally optimal particle g is selected from the external archive using a binary tournament selection method (based on crowding comparison) to guide its flight. b i.

[0041] e. Particle Update: Update the velocity and position of each particle according to the following formula: (6) (7) Among them, pbest i Let c1 and c2 be the historical best position of particle i, c1 and c2 be learning factors, and r1 and r2 be random numbers between [0,1]. The inertia weight w(k) adopts an adaptive adjustment strategy: (8) Where k is the current iteration number, K max This represents the maximum number of iterations; this strategy encourages global search in the early stages and promotes local convergence in the later stages.

[0042] f. Hybrid Operation: To avoid getting trapped in local optima, every G (e.g., G=10) generations, a simulated annealing-based local search is performed on the 10% of solutions with the lowest crowding in the outer archive (i.e., solutions in sparse regions). Specifically, a Gaussian perturbation is applied to the decision variables of these solutions, and new solutions are accepted or rejected according to the Metropolis criterion, thereby enhancing the algorithm's neighborhood exploration capability.

[0043] g. Termination condition: When the number of iterations reaches K. max (For example, K) max The algorithm terminates when the change in the hypervolume metric of the external archive is less than a minimum threshold ε for M consecutive generations (e.g., M=50) or when the change is less than a minimum threshold ε.

[0044] Ultimately, all the non-dominated solutions stored in the external archive constitute the desired Pareto optimal frontier.

[0045] The solution set of the Pareto optimal front can be visualized, and the specific visualization process is as follows: The solution set of the Pareto optimal frontier is transmitted to a visualization interface for display. The main display area of ​​the visualization interface is a three-dimensional scatter plot, with its three coordinate axes corresponding to the normalized flood control safety target (inversely, the larger the value, the safer), power generation benefit target, and ecological protection target, respectively. Each scatter point represents a complete 72-hour dispatch plan. Users (dispatchers) can rotate, zoom, and pan the three-dimensional view using the mouse. When the mouse hovers over any scatter point, a floating window on the visualization interface displays the specific values ​​of the four objective functions of the plan. After clicking on any scatter point in the scatter plot, the detailed information panel on the right side of the visualization interface will immediately draw the predicted process lines of the dam front water level, inflow, total outflow, and total power generation for the next 72 hours, as well as the process diagram of the opening changes of the key flood discharge gates.

[0046] Step S500: Based on the current scheduling scenario, set the relative weight vector W of the core objectives, calculate the relative proximity C, sort the flood discharge scheduling scheme set, and obtain the final flood discharge scheduling scheme; the core objectives include flood control safety, power generation benefits, and ecological protection. The final flood discharge and air release schedule is obtained based on the TOPSIS method. A relative weight vector W is defined for the three core objectives of flood control safety, power generation efficiency, and ecological protection under the current scheduling scenario, where W = [w s , w e , w p ], where w s For flood control safety, w e w is the weight vector of power generation benefits. p The weight vector is used for ecological protection; for example, during the flood season, the weight can be set to [0.7, 0.2, 0.1]; during the dry season, it can be set to [0.2, 0.6, 0.2]. The relative closeness C is calculated based on the relative weight vector, and the calculation method includes: Obtain the solution value of the objective function, construct the decision matrix, perform normalization processing, and calculate the weighted normalized decision matrix; Determine the positive and negative ideal solutions of the weighted normalized decision matrix, and calculate the Euclidean distance D+ from the solution of the objective function to the positive ideal solution and the Euclidean distance D- to the negative ideal solution; the positive ideal solution is the optimal solution for each indicator; the negative ideal solution is the worst solution for each indicator. The relative closeness C is obtained, and the expression for calculating the relative closeness C is: (9) The top three scheduling schemes are ranked in descending order based on the relative proximity C value, and the final flood discharge and air release scheduling scheme is determined from the recommended schemes.

[0047] In this embodiment, after determining the final flood discharge and release schedule, the operation instructions predicted for the first decision step (i.e., the next hour) in the final flood discharge and release schedule are extracted, such as "Open the No. 1 deep hole gate to 35.0%, and set the flow rate of the No. 2 generator set to 150 m³". 3 After the command is encapsulated into a data packet conforming to the DNP3 protocol standard, the final flood discharge and air release schedule is executed; A closed-loop correction mechanism is set within the execution cycle (i.e., decision step) of the final flood discharge and air release scheduling scheme; the closed-loop correction mechanism includes: Obtain the actual measured hydrograph WL of the water level upstream of the dam. actual Total discharge flow process line Q out,actual Compared with the predicted upstream water level hydrograph WL pred Total discharge flow process line Q out，predThe prediction error vector e is calculated, and the distributed hydrological model and the one-dimensional unsteady flow dynamics model are updated to form a closed-loop adaptive model; the execution cycle is 1 hour.

[0048] The prediction error vector e includes the prediction error of the upstream water level hydrograph e. WL Total discharge flow process curve prediction error e Q ; The prediction error of the upstream water level hydrograph e WL The calculation expression is: (10) The total outflow process curve prediction error e Q The calculation expression is: (11) The closed-loop adaptive formation process includes the following: In the closed-loop correction mechanism, the prediction error vector e is input into the Kalman filter. The parameters of the distributed hydrological model and the one-dimensional unsteady current hydrodynamic model are corrected by adjusting the estimated values ​​of the Kalman filter's state variables, thereby updating these models. The Kalman filter's state variables are the key physical parameters in the distributed hydrological model and the one-dimensional unsteady current hydrodynamic model. The corrected parameters will be used for a new round of prediction and optimization calculations in the next decision cycle (i.e., at the start of the next hour). This ensures the long-term accuracy of the entire decision-making process and its adaptability to environmental changes.

[0049] The present invention also provides an intelligent decision-making system for dam flood discharge and emptying oriented to multi-objective optimization, which is implemented by the intelligent decision-making method for dam flood discharge and emptying oriented to multi-objective optimization. The system includes: a multi-source data fusion and preprocessing module P100, a spatiotemporal evolution prediction module P200, a multi-objective optimization module P300, a collaborative decision-making and closed-loop execution module P400, and a database P500. Among them, the underlying data acquisition interface layer of the multi-source data fusion and preprocessing module P100, through diverse physical and logical connection methods, realizes real-time connection with multiple key information systems inside and outside the dam, completes the acquisition of multi-dimensional data streams of the dam reservoir, and generates standard multi-dimensional data streams to input into the spatiotemporal evolution prediction module P200.

[0050] The spatiotemporal evolution prediction module P200 constructs a hybrid physics model based on the standard multidimensional data stream to simulate the virtual response process of future hydrological conditions and reservoir response, and outputs the predicted water level and flow velocity evolution processes and key results at each cross section along the dam; the spatiotemporal evolution prediction module P200 includes a watershed prediction module P210 and a reservoir evolution module P220, as follows: Figure 2 As shown; The watershed prediction module P210 is based on a distributed hydrological model with a defined physical mechanism. Based on high-precision digital elevation model (DEM) data, the entire catchment area upstream of the dam is discretized into a computational domain consisting of 1 km × 1 km square grid cells. For each independent grid cell, a complete hydrological process simulation calculation is performed, and the inflow process line of the dam is output to the reservoir evolution module P220.

[0051] The reservoir evolution module P220 is based on a one-dimensional unsteady flow dynamics model. It receives the inflow process line of the dam, obtains the flood discharge and release schedule to be optimized, outputs the predicted water level and flow velocity evolution process and key results of each section along the dam, and transmits them to the multi-objective optimization module P300. The multi-objective optimization module P300 is used to receive the predicted evolution process of water level and flow velocity at each section along the dam and key results. In a complex decision space containing multiple conflicting objectives, it systematically searches and generates a set of flood discharge and release schemes (Pareto optimal frontier) with different weights among the objectives and inputs them into the collaborative decision-making and closed-loop execution module P400. The collaborative decision-making and closed-loop execution module P400 is a graphical user interface for interaction, which visualizes the flood discharge and air release scheduling scheme set generated by the multi-objective optimization module P300. Based on the current scheduling scenario, it sets the relative weight vector W of the core objective, calculates the relative proximity C, sorts the flood discharge and air release scheduling scheme set, and obtains the final flood discharge and air release scheduling scheme. It also sets a closed-loop correction mechanism to update the spatiotemporal evolution prediction module P200, forming a closed-loop adaptive mechanism.

[0052] The P500 database is a shared time-series database that enables data interaction between modules. It is coordinated by a unified scheduling and task management service to ensure smooth data flow and orderly execution of computing tasks.

[0053] Specific example 1 is as follows: This study uses a large cascade reservoir as an example. The reservoir has a total capacity of 5 billion cubic meters, a flood control limit level of 145.0 meters, and a check flood level of 155.0 meters. It is equipped with four 250 MW mixed-flow turbine generator units, five deep-slot flood discharge gates, and three surface gates. Currently, the reservoir is in its main flood season, with an initial upstream water level of 144.5 meters.

[0054] At time T0, the method of the present invention initiates a decision-making process.

[0055] Step 1: Collect multidimensional data streams from the dam and reservoir, preprocess them, generate standard multidimensional data streams, and establish a database.

[0056] The NWP weather forecast indicates that the basin will experience a period of heavy rainfall within the next 72 hours, with a flood peak expected between T+24 and T+48 hours.

[0057] Step 2: Load the cascaded distributed hydrological model and the one-dimensional unsteady flow dynamics model to achieve spatiotemporal evolution prediction.

[0058] Based on the data displayed by the NWP weather forecast, a distributed hydrological model is driven to predict the inflow process line of the reservoir in the next 72 hours, with the peak flow expected to reach 8000 cubic meters per second.

[0059] Step 3: Generate a set of flood discharge and release schedules with different weights among the objectives (Pareto optimal frontier): The HA-MOPSO algorithm is initiated for optimization. The decision variable X represents the hourly gate opening and unit flow rate for the next 72 hours.

[0060] The objective function is set as follows: F safety WL flood It is 145.0 meters; F econ The electricity price (Price(t)) exhibits a clear peak-valley characteristic, with the price during peak daytime hours being three times that during off-peak nighttime hours; F eco The minimum downstream ecological flow rate stipulated in the regulations is 150 cubic meters per second; F h ΔQ max Set to 500 (m) 3 / s) / h. After approximately 30 minutes of computation (500 iterations), the algorithm generates a Pareto front containing 200 non-dominated solutions.

[0061] Step four: Visualize the Pareto front. Based on the current urgency of flood control, the dispatcher sets the TOPSIS weights to W = [0.7, 0.2, 0.1]. The set of flood discharge scheduling schemes (Pareto optimal front) generated in Step three is sorted, and the scheme ranked first is determined as the final flood discharge scheduling scheme. The scheduling strategy characteristics of the final flood discharge scheduling scheme are: 1. Pre-discharge to free up reservoir capacity: During the period from T0 to T+12 hours before the flood arrives, taking advantage of the low electricity price at night, some deep flood discharge outlets are opened to lower the water level from 144.5 meters to 143.0 meters in advance, freeing up 200 million cubic meters of flood control capacity.

[0062] 2. Peak Shaving and Flood Storage & Staggered Power Generation: During the main flood phase from T+24 to T+48 hours, the gate opening was precisely controlled, limiting the maximum outflow to 6500 cubic meters per second. This resulted in a slow rise in the water level in front of the dam, reaching a peak of 144.95 meters, successfully reducing the flood peak from 8000 cubic meters per second to 1500 cubic meters per second. Simultaneously, the scheduling plan fully utilized the opportunity of rising water head, operating all four generating units at full capacity during the daytime when electricity prices are high, while reducing power generation at night to store water in the reservoir.

[0063] 3. Later-stage decline: After the flood process ends, the discharge flow will be gradually increased so that the reservoir water level will steadily decline to below the flood control limit level at the end of the optimized time domain.

[0064] Ultimately, the final flood discharge scheduling scheme achieved a maximum water level of 144.95 meters within the 72-hour optimization time domain, generating a total power generation benefit equivalent to RMB 28.5 million, meeting ecological flow requirements throughout the entire process, and the maximum discharge flow variation rate did not exceed 480 (m³). 3 / s) / h.

[0065] The comparison examples are as follows: Under the exact same flood scenario and initial conditions as in Example 1, scheduling is performed using a traditional method based on static scheduling procedures. These procedures stipulate: 1. The floodgates may only be opened when the reservoir water level exceeds 145.0 meters.

[0066] 2. When the inflow exceeds 5,000 cubic meters per second, all generator sets shall operate at their rated output.

[0067] 3. When the reservoir water level reaches 146.0 meters, open 2 deep-hole gates; when it reaches 147.0 meters, open all 5 deep-hole gates.

[0068] Under the guidance of this procedure, the scheduling process is as follows: In the initial stage of the flood, the reservoir did not conduct any pre-discharge operations as the water level was not exceeded, and the water level remained around 144.5 meters. When the flood peak began to flow in at T+24 hours, the inflow exceeded 5,000 cubic meters per second, and the generating units began to operate at full capacity. However, due to the rapid increase in inflow, the reservoir water level rose rapidly. At T+30 hours, the water level exceeded 145.0 meters and quickly reached 146.0 meters. At this point, the dispatcher opened two deep-hole gates according to procedures, resulting in a sudden increase in the discharge flow. However, due to insufficient discharge, the water level continued to rise, reaching 147.0 meters at T+34 hours, forcing the opening of all five deep-hole gates, causing the discharge flow to increase sharply to 7,500 cubic meters per second. During the entire flood, the highest water level in the reservoir reached 147.25 meters, far exceeding the flood control limit. Since most of the floodwater was discharged through the floodgates, and the timing of power generation did not match the peak electricity price, the total power generation benefit was only 19.2 million yuan. In addition, due to the delayed and violent gate opening operation, the discharge flow suddenly increased from the power generation flow to 7,500 cubic meters per second in a short period of time, causing a huge impact on the downstream river channel.

[0069] Table 1 shows a comparison of the effects of specific example 1 and the comparison example: Table 1

[0070] As can be seen from the detailed comparison of the specific example 1 and the comparative example above, the intelligent decision-making method for dam flood discharge and emptying oriented to multi-objective optimization provided by the present invention, compared with traditional technology, can significantly improve the comprehensive operational efficiency of reservoirs through forward-looking prediction and global optimization while ensuring absolute flood control safety. At the same time, it takes into account multiple needs such as ecological and engineering safety, realizes intelligent and refined dam scheduling decision-making, and has extremely high engineering application value.

[0071] The above-disclosed embodiments are merely specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A smart decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization, characterized in that, include: Collect multidimensional data streams from dams and reservoirs, preprocess them, generate standard multidimensional data streams, and establish a database; The multidimensional data stream includes real-time monitoring data of the dam body, real-time observation data of various rain gauges and hydrological stations in the basin, meteorological forecast data in the basin, electricity price forecast curves and power grid load demand curves, and flood discharge and release schedules to be optimized. Load a distributed hydrological model, which is used to predict the inflow process curve of the dam; input the meteorological forecast data into the distributed hydrological model and output the inflow process curve of the dam; A one-dimensional unsteady flow dynamics model is loaded, which is used to simulate and predict the evolution of water level and velocity at various cross-sections along the dam under a given scheduling scheme, and to obtain key results. The inflow process line of the dam and the flood discharge scheduling scheme to be optimized are input into the one-dimensional unsteady flow dynamics model, and the predicted evolution of water level and velocity at various cross-sections along the dam and key results are output. The key results include the upstream water level process line, the total outflow process line, and the total power generation process line. Define decision variable X and objective function. Based on the predicted evolution process of water level and flow velocity at each section along the dam and the key results, input the decision variable X into the objective function to solve and generate a set of flood discharge and release schedule schemes with different weights among the objectives. Based on the current scheduling scenario, a relative weight vector W for the core objectives is set, the relative proximity C is calculated, and the flood discharge scheduling scheme set is sorted to obtain the final flood discharge scheduling scheme; the core objectives include flood control safety, power generation benefits, and ecological protection.

2. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, A closed-loop correction mechanism is set up during the execution cycle of the final flood discharge and air release schedule; the closed-loop correction mechanism includes: Obtain the actual measured hydrograph WL of the water level upstream of the dam. actual Total discharge flow process line Q out,actual Compared with the predicted upstream water level hydrograph WL pred Total discharge flow process line Q out，pred The prediction error vector e is calculated, and the distributed hydrological model and the one-dimensional unsteady flow dynamics model are updated to form a closed-loop adaptive model; the execution cycle is 1 hour.

3. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, The distributed hydrological model is a digital elevation model, which is composed of square grid cells formed by discretizing the dam based on physical mechanisms; The process of independently calculating the inflow hydrograph for each grid cell and outputting the inflow hydrograph for the dam includes: The meteorological forecast data is input into the distributed hydrological model, and the evapotranspiration and soil infiltration are calculated separately for each network unit. The meteorological forecast data includes temperature, humidity, wind speed, and solar radiation; Based on the evaporation and transpiration and soil infiltration, it is determined whether the surface rainfall intensity exceeds the infiltration capacity. When the surface rainfall intensity exceeds the infiltration capacity, a two-dimensional diffused wave surface runoff simulation is performed based on the generated excess runoff to predict the surface runoff of each network unit. The runoff is calculated step by step to output the inflow process line of the dam.

4. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, The expression for the decision variable X is: ； in, Let be the percentage of the i-th gate that is open. Let T be the reference traffic of the j-th unit, and T be the optimization time domain. The decision step size is defined as n, where n represents the total number of gates, m represents the total number of generator sets, and t represents the time period.

5. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, The predicted evolution of water level and flow velocity at various cross-sections along the dam, and the output process of key results, include: The inflow process line output by the distributed hydrological model is used as the upstream boundary condition of the one-dimensional unsteady flow dynamics model; the flood discharge and air release scheduling scheme to be optimized is used as the downstream boundary condition of the one-dimensional unsteady flow dynamics model. The upstream and downstream boundary conditions are input into the one-dimensional unsteady flow dynamics model, and the flow evolution is carried out through the one-dimensional Saint-Venant equations. The predicted evolution process of water level and flow velocity at each section along the dam and the key results are output.

6. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, The distributed hydrological model and the one-dimensional unsteady flow dynamics model are cascaded to form a spatiotemporal evolution prediction method from meteorological forecasting to watershed runoff generation to reservoir scheduling and state response.

7. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 1, characterized in that, The objective function includes: flood control safety objective function. Power generation benefit objective function Ecological protection objective function and engineering health objective function ; The flood control safety objective function The expression is: ； Where WL(t) is the predicted hydrograph of the water level upstream of the dam during time period t, WL flood WL is the flood control limit water level process line. check To verify the flood level hydrograph, w1 and w2 are weighting coefficients, and w1 >> w2; The power generation efficiency objective function The expression is: ； Where Price(t) is the predicted time-of-use electricity price for period t, Power(t) is the predicted total power generation for period t, and Cop is the gate operation cost coefficient. Let be the percentage of the i-th gate that is open. For decision-making step size; The ecological protection objective function The expression is: ； Among them, Q out (t) represents the predicted total discharge flow during time period t, Q eco_min (t) represents the minimum ecological water demand flow rate specified downstream during time period t; The project health objective function The expression is: ； Where, ΔQ max This represents the maximum permissible rate of change in downstream flow rate per unit time.

8. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 7, characterized in that, The method for calculating the relative closeness C includes: Obtain the solution value of the objective function, construct the decision matrix, perform normalization processing, and calculate the weighted normalized decision matrix; Determine the positive and negative ideal solutions of the weighted normalized decision matrix, and calculate the Euclidean distance D+ from the solution of the objective function to the positive ideal solution and the Euclidean distance D- to the negative ideal solution; the positive ideal solution is the optimal solution for each indicator; the negative ideal solution is the worst solution for each indicator. Obtain the relative closeness C, and the expression for calculating the relative closeness is: 。 9. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 2, characterized in that, The prediction error vector e includes the prediction error of the upstream water level hydrograph e. WL Total discharge flow process curve prediction error e Q ; The prediction error of the upstream water level hydrograph e WL The calculation expression is: ； The total outflow process curve prediction error e Q The calculation expression is: 。 10. The intelligent decision-making method for dam flood discharge and emptying oriented towards multi-objective optimization as described in claim 9, characterized in that, The closed-loop adaptive formation process includes the following: In the closed-loop correction mechanism, the prediction error vector e is input into the Kalman filter, and the parameters of the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model are corrected by modifying the estimated values ​​of the state variables of the Kalman filter, thereby updating the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model; the state variables of the Kalman filter are the key physical parameters in the distributed hydrological model and the one-dimensional unsteady flow hydrodynamic model.

Images