Intelligent acceleration-heterogeneous parallel river basin water engineering group joint scheduling solution method

Abstract

This invention discloses an intelligent acceleration-heterogeneous parallel joint scheduling solution method for watershed water conservancy engineering groups, belonging to the field of numerical calculation and intelligent scheduling technology for water conservancy engineering. This method first divides the hydrological control intervals and implements differentiated scheduling optimization that maximizes conventional and remaining flood control capacity for different control sections. It then uses a heterogeneous parallel architecture to split tasks and achieve synchronous computation across multiple nodes, employing the XGBoost algorithm to replace traditional river channel calculation methods. This invention can significantly improve scheduling solution efficiency, overcome computational bottlenecks, ensure computational accuracy, enhance adaptability to complex watersheds, optimize water resource allocation, strengthen flood control, and has low deployment costs and strong compatibility, meeting the needs of real-time watershed scheduling and scientific management.

Description

Technical Field

[0001] This invention relates to the field of numerical calculation and scheduling technology for water conservancy projects, and in particular to a smart acceleration-heterogeneous parallel joint scheduling solution method for watershed water conservancy projects. Background Technology

[0002] Watershed water conservancy projects are the core carriers for water resource regulation and flood control in watersheds. The rapid and accurate solution of their joint scheduling schemes is crucial for achieving efficient utilization of water resources and improving flood control safety. With the deepening of watershed management, the scale of water conservancy projects continues to expand and the types of projects become increasingly diverse. The hydraulic coupling relationships between water conservancy projects centered on reservoirs are becoming increasingly complex. Traditional serial solution methods are no longer suitable for the real-time solution requirements of joint scheduling of complex watershed water conservancy projects. In particular, the problems of low solution efficiency and long time consumption of core calculation links are especially prominent in multi-reservoir cluster scheduling scenarios that include heterogeneous features of conventional scheduling and optimized scheduling.

[0003] Traditional joint scheduling solutions for watershed water conservancy projects often employ a serial iterative approach, calculating each water conservancy project and hydrological unit in a fixed order, with the result of the previous unit serving as the input for the next. This approach has significant limitations: First, it is inefficient, with the solution cycle increasing linearly with the number of water conservancy projects, making it difficult to respond to real-time scheduling needs in emergency scenarios such as extreme floods and easily leading to delayed scheduling decisions. Second, it is highly dependent on the computing power of a single computing node, and hardware resource expansion cannot effectively improve the solution capability, and it has poor fault tolerance, requiring the entire solution process to be re-executed if a single link fails. Third, it does not incorporate intelligent acceleration technology, resulting in limited improvement in the solution speed for core tasks such as river evolution and scheduling optimization, making it difficult to match the computational needs of highly coupled and complex water conservancy projects.

[0004] River evolution calculation is a core component of joint scheduling solutions. Traditional methods, such as the Muskingan method, require numerous iterative calculations, resulting in extremely low efficiency and becoming a key bottleneck restricting the overall scheduling solution speed. While intelligent acceleration algorithms like XGBoost, with their efficient gradient boosting characteristics, have a natural advantage in complex time-series prediction tasks, they have not been effectively applied to rapid river evolution calculations. Simultaneously, although parallel computing technology has been implemented in multiple engineering fields, its integrated application in watershed water conservancy project scheduling faces challenges due to the significant differences in project types and complex hydraulic coupling relationships. These challenges include difficulty in task decomposition, mismatched computing power requirements between different task types, poor synergy between intelligent acceleration and parallel computing, and difficulties in data synchronization between nodes. This is particularly problematic for the heterogeneous characteristics of two conventional scheduling tasks and two optimized scheduling tasks in multi-reservoir clusters, where a suitable parallel computing architecture and task scheduling strategy are lacking, hindering efficient collaborative solutions for heterogeneous scheduling tasks.

[0005] Currently, the integrated application of XGBoost intelligent acceleration and heterogeneous parallel computing has become an important direction for improving the computational efficiency of complex engineering projects. However, in the field of joint scheduling of watershed water conservancy projects, there are still two major technical challenges: First, how to break through the application bottleneck of XGBoost in river evolution calculation and replace the traditional Muskingum method to achieve fast and high-precision solution of river evolution; Second, how to reasonably split the computational tasks, build an adapted heterogeneous parallel architecture, and design an "intelligent acceleration-heterogeneous parallel" collaborative strategy to achieve collaborative solution of reservoirs with different scheduling strategies and other water conservancy projects and accurate data synchronization between nodes, taking into account the heterogeneous scheduling characteristics of multiple reservoirs.

[0006] Therefore, in view of the shortcomings of existing technologies such as low efficiency of serial solution, long calculation time of river evolution, poor adaptability of parallel solution of heterogeneous water conservancy groups, and insufficient integration of intelligent acceleration and parallel computing, it is urgent to propose a joint scheduling acceleration solution method for water conservancy groups of water conservancy groups that combines "intelligent acceleration-heterogeneous parallelism". By integrating XGBoost intelligent acceleration and heterogeneous parallelism technology, the core problems of intelligent acceleration calculation of river evolution and heterogeneous parallel scheduling of multiple reservoirs can be solved, thereby improving the overall solution efficiency of joint scheduling and meeting the needs of real-time scheduling and scientific management of water conservancy groups of water conservancy groups. Summary of the Invention

[0007] To address the aforementioned issues, this invention provides an intelligent acceleration-heterogeneous parallel joint scheduling solution method for watershed water conservancy projects. This method improves model solution efficiency while ensuring scheduling accuracy, specifically addressing the problems of "long time consumption and difficulty in real-time operation" in traditional model solutions. It provides scientific support for the engineering application of joint optimization scheduling for flood control in watershed water conservancy projects.

[0008] This invention is implemented as follows:

[0009] A smart acceleration-heterogeneous parallel joint scheduling solution method for watershed water conservancy projects includes the following steps:

[0010] Collect and organize basic watershed data, including engineering and hydrological data, to ensure data consistency in time sequence and perform preprocessing.

[0011] Based on the natural water system network, spatial distribution of water conservancy projects, and hydraulic coupling relationships of the basin, and combined with the pre-set control section boundaries, the division of the basin's hydrological control zone is completed;

[0012] Determine the strength of hydraulic connections between water projects, ensure close hydraulic connections between water projects in each section, and use the control section as the only coupling node in each section to avoid direct hydraulic coupling across sections.

[0013] Scheduling models are constructed for both conventional and optimized scheduling intervals. The conventional scheduling model follows the established engineering rules, while the optimized scheduling model reconstructs the objective function and constraints.

[0014] The Muskingu method was used to back-calculate flood data for the optimized scheduling interval, providing hydrological boundary conditions for the optimized scheduling model.

[0015] The optimization objective is to maximize the remaining flood control capacity, while respecting engineering safety and hydraulic constraints.

[0016] A heterogeneous parallel computing framework is built, which allocates the conventional scheduling model and the optimized scheduling model to two independent computing nodes to achieve synchronous parallel solution, avoids the sequential dependency of traditional serial computing, and improves the solution efficiency of the scheduling model.

[0017] Define the functions of the main control node, computing node 1, and computing node 2. The main control node is responsible for parameter initialization, task distribution, status monitoring, and result summarization. Computing node 1 is responsible for solving the conventional scheduling model, and computing node 2 is responsible for solving the optimized scheduling model.

[0018] The main control node receives the results of the division of the watershed hydrological control zone, constraint parameters, and scheduling target parameters. It splits the tasks according to the differentiated scheduling requirements, distributes the routine scheduling solution tasks to computing node 1, distributes the optimized scheduling solution tasks to computing node 2, and clarifies the calculation scope and data interaction frequency of the two computing nodes.

[0019] The two computing nodes receive task instructions and parameters, independently start the corresponding scheduling model to solve the problem, and report the running status back to the main control node.

[0020] The master control node monitors the running status of the two computing nodes in real time and performs corresponding processing operations for abnormal situations to ensure computing synchronization;

[0021] After the two computing nodes complete their calculations, they upload the results, and the main control node merges and outputs the calculation results.

[0022] After the main control node aggregates the parallel computing results, it uses the XGBoost algorithm to perform flood evolution calculations across the entire basin, enabling the prediction of water level and flow at river sections and completing the coupling of hydraulic results between the conventional and optimized scheduling intervals.

[0023] Furthermore, the division of the watershed hydrological control zone is based on control section 1 as the conventional scheduling control boundary, with its upstream area classified as the conventional scheduling zone, including all reservoirs, sluices, and supporting flood storage and detention areas within the zone; and control section 2 as the optimized scheduling control boundary, with its upstream area including the region from downstream of control section 1 to upstream of control section 2, classified as the optimized scheduling zone, including the water conservancy projects within the zone.

[0024] Furthermore, the determination of the strength of hydraulic correlation between water projects is based on the hydraulic influence weight, and the calculation formula is as follows:

[0025]

[0026] In the formula, The weight of the hydraulic influence of upstream water project i on downstream water project j; The outflow from the upstream water project is the time interval T after the upstream water project lags behind. The flood evolution time lag; K is the river flow propagation coefficient, 0 ; For downstream water projects j in Inbound traffic at any given time.

[0027] Furthermore, the hydrological boundary conditions are calculated using the following formula:

[0028]

[0029] In the formula, To optimize the flood process between water project i and control section 2 within the scheduling interval; To control the flow rate at time t at section 2; , , This represents the flood evolution coefficient; To optimize the outflow of water project i at time t within the scheduling interval.

[0030] Furthermore, the objective function for maximizing the remaining flood control capacity is as follows:

[0031]

[0032] In the formula, The sum of the remaining flood control capacity of the reservoir group within the optimized scheduling interval; n is the number of reservoirs participating in the scheduling within the optimized scheduling interval; This represents the upper limit of the flood control capacity of reservoir i; Let be the actual storage capacity of reservoir i at the end of the scheduling period and in time period T.

[0033] The hydraulic characteristic constraints include water level constraints, and the calculation formula is as follows:

[0034]

[0035] In the formula, , Water projects The minimum and maximum water levels that are allowed to be reached during the scheduling process; For water engineering At the end of the scheduling period, the The actual water level during the period; The preset target water level at the end of the water project period;

[0036] Flow constraints include the discharge capacity constraints of water projects, the safe discharge capacity constraints of downstream river channels, and the safe flow constraints of control sections. The calculation formulas are as follows:

[0037]

[0038]

[0039]

[0040] In the formula, Let be the actual discharge flow of the i-th water project at time t. For water engineering Maximum discharge capacity For the downstream of the water project The flow rate at time t of each control section; For water engineering The safe discharge capacity of the downstream river channel; The safe flood discharge flow rate for the j-th control section downstream;

[0041] The water balance constraint is calculated using the following formula:

[0042]

[0043] in, For water engineering The water storage volume at time t; Water projects Inflow rate during time period t; Water projects Outflow rate during time period t; To calculate the time step;

[0044] Hydraulic constraints are calculated using the following formula:

[0045]

[0046]

[0047] In the formula, For water engineering The water storage at time t, For water engineering The water surface elevation at time t; For water engineering Water level-reservoir capacity relationship function; For water engineering The water level-discharge capacity relationship function.

[0048] The formula for calculating the outbound flow rate variation constraint is as follows:

[0049]

[0050] In the formula, Water projects Outflow rate during time period t; For water engineering The maximum allowable flow rate variation within adjacent time periods.

[0051] Furthermore, the main control node synchronously distributes the watershed hydrological boundary conditions and basic water engineering parameters to the two computing nodes, and unifies the scheduling decision-making period and total computing time.

[0052] Furthermore, the watershed hydrological boundary conditions include precipitation, runoff, and inter-regional floods; the basic parameters of the water engineering include engineering thresholds, water level-storage capacity curves, and water level-discharge capacity curves; and the scheduling decision period is 4 hours.

[0053] Furthermore, the abnormal situation is that if a node experiences a computational lag or failure, the redundancy computing mechanism is immediately activated to redistribute the node's tasks to a backup computing node, ensuring that the parallel computing process is continuous and uninterrupted.

[0054] Furthermore, the XGBoost model construction includes:

[0055] A sample dataset is constructed by combining the collected historical flood monitoring data of the basin with the results of inverse flood calculation and parallel scheduling solution to determine the sample features and sample labels, forming a sample library required for model training and testing. The sample features include upstream section flow, water level and scheduled discharge flow, and the sample labels are the flow and water level at the corresponding time at the downstream section.

[0056] Set the key hyperparameters of the XGBoost model, divide the sample dataset into training and validation sets according to a preset ratio, train the XGBoost model, verify the model performance through core evaluation metrics, and complete the model calibration.

[0057] The downstream discharge flow at the control section, the downstream discharge flow of the optimized scheduling section, and the upstream water level obtained through parallel computing are used as inputs to the XGBoost model, which are then substituted into the model to predict the downstream flow and water level at the corresponding time.

[0058] Furthermore, the key hyperparameters of the XGBoost model include the learning rate, maximum tree depth, L1 regularization parameter, and L2 regularization parameter; an optimization strategy of random sampling of samples and features is adopted, combined with an early stopping mechanism to dynamically optimize the number of iterations; the preset ratio of the sample dataset is 8:2, that is, the training set accounts for 80% of the sample dataset, and the validation set accounts for 20% of the sample dataset; the preset core evaluation metrics include the coefficient of determination R², root mean square error RMSE, and mean absolute error MAE.

[0059] The beneficial effects of this invention are as follows: This invention addresses the problems of low efficiency and strong dependence on single nodes in traditional serial solutions. By splitting the conventional and optimization heterogeneous scheduling tasks into two categories through a heterogeneous parallel architecture, it achieves synchronous operation of multiple nodes, effectively avoiding the drawback of the solution cycle extending linearly with the number of projects. It also distributes the computing load, improves the fault tolerance of the system, significantly shortens the overall solution time, and accurately meets the real-time decision-making needs of emergency scenarios such as flood control.

[0060] To address the bottlenecks of large iterations and long computation time in traditional river channel calculations, the XGBoost algorithm is adopted to replace the traditional solution method, which significantly reduces the number of iterations and core computation time. While breaking through the bottleneck of solution speed, it ensures computational accuracy comparable to traditional methods, effectively avoids result distortion, and provides reliable support for accurate solutions to joint scheduling schemes.

[0061] Based on the structural dimensionality reduction approach and heterogeneous parallel architecture, it accurately matches the computing power requirements of water engineering clusters, effectively solving pain points such as difficulty in task decomposition, computing power mismatch, and data synchronization difficulties in complex scenarios; it flexibly adapts to the development trend of expanding scale and increasing types of water engineering clusters, and improves the adaptability to complex watershed heterogeneous scheduling scenarios.

[0062] The system optimizes scheduling by maximizing the remaining flood control capacity, and combines efficient and accurate solution capabilities to coordinate the scheduling behavior of various water projects, optimize the allocation of water resources in the basin, and effectively reduce flood risks. It is highly compatible with existing scheduling systems, does not require major reconstruction of the existing architecture, has low deployment costs and is easy to implement, and provides stable and reliable technical support for scientific management of water resources in the basin and practical engineering scenarios such as flood control and disaster reduction.

[0063] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0064] Figure 1 This is a flowchart of the intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method of the present invention.

[0065] Figure 2 This is a schematic diagram of the intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling of the present invention. Detailed Implementation

[0066] Example 1:

[0067] This embodiment presents a method for jointly scheduling and solving a group of watershed water conservancy projects using intelligent acceleration and heterogeneous parallel approaches. Figure 1 As shown, it includes the following steps:

[0068] Step S1, Data Acquisition and Preprocessing:

[0069] The collection and organization of basic watershed data, including engineering and hydrological data, ensures data consistency and reliability, providing fundamental support for subsequent analyses of the impact of upstream water project discharges on downstream water project inflows and water levels, XGBoost model training and calibration, and constraint and boundary setting. Specifically:

[0070] 1. Engineering Data: Comprehensive collection of core parameters for various water projects within the basin, including reservoirs, sluice gates, flood storage and detention areas, and river channels. Reservoir parameters include reservoir capacity curves, spillway dimensions, gate control parameters, flood control levels, characteristic reservoir capacity, discharge capacity, and maximum operating water levels; sluice gate parameters include gate type, gate opening dimensions, and opening / closing control parameters; flood storage and detention area parameters include area-volume curves and activation thresholds; and river channel parameters include cross-sectional dimensions, roughness, flood evolution parameters, safe discharge capacity, and flood evolution time lag.

[0071] 2. Hydrological Data: Collect long-term hydrological monitoring data of the watershed, including reservoir inflow / outflow processes, inflow / outflow at various river sections, water level, precipitation data, etc. Simultaneously complete the construction of historical flood sample datasets, covering all the basic hydrological data required for model training, constraint and boundary setting, and provide data support for subsequent analysis and modeling.

[0072] Data source: All basic data are from watershed hydrological monitoring stations and water project monitoring sensors to ensure the authenticity, timeliness and continuity of the data.

[0073] Preprocessing methods: To ensure data quality, the 3σ criterion was used to remove outliers. For missing data in the hydrological time series, linear interpolation was performed using measured data at adjacent times. The linear relationship between flow and water level at two points was used to fit the missing values ​​to complete the data and ensure that the data meets the requirements of subsequent analysis and model training.

[0074] Step S2, hydrological control zone delineation:

[0075] Based on the natural water system network, spatial distribution of water conservancy projects, and hydraulic coupling relationships within the basin, and using pre-set control sections 1 and 2 as core boundaries, the hydrological control zones of the basin are divided, laying the foundation for subsequent differentiated scheduling and parallel computing. Specific implementation details are as follows:

[0076] Figure 2The study demonstrates the topological relationships of the reservoir group within the study area, the coupling logic of the river channel calculation module and the scheduling module, and the functions and relationships of each module and element are as follows:

[0077] The upstream parallel reservoir group (dashed box area): consists of Reservoir 1 and Reservoir 2 connected in parallel. The discharge from both reservoirs flows into Reservoir 3 / control section 1, forming the first-level control unit in the upstream of the basin and serving as the water input source for the entire dispatching system.

[0078] Downstream series reservoirs / control sections: Reservoir 3 simultaneously serves as the confluence node of the upstream reservoir group and control section 1, receiving the outflow from reservoirs 1 and 2. It also serves as the starting point for downstream river channel calculations and is the core hub for upstream and downstream scheduling.

[0079] End-point reservoir / control section (dashed box area): Reservoir 4 and control section 2 constitute the second-level regulation unit in the downstream of the basin. The discharge from reservoir 4 and the inflow from control section 1 are used as inputs to the XGBoost river channel calculation module. After calculation, the flood process of control section 2 is output, which is a key control node for flood control safety in the basin.

[0080] 1. Core control section functions: Control section 1 is defined as the conventional scheduling control boundary, and its upstream area is classified as the conventional scheduling interval, including all reservoirs, sluices and supporting flood storage and detention areas within this interval. The scheduling is carried out using the established conventional scheduling rules of the project. Control section 2 is defined as the optimized scheduling control boundary, and its upstream area (including the area from the downstream of control section 1 to the upstream of control section 2) is classified as the optimized scheduling interval, including the water conservancy projects within this interval. The optimized scheduling model with the maximum remaining flood control capacity as the core objective is used to solve the problem.

[0081] 2. Boundary Delineation: Collect inflow / outflow, water level, engineering parameters, and watershed hydrological monitoring data for all water projects within the basin. Determine the strength of hydraulic connections between water projects using the hydraulic influence weighting formula to ensure close hydraulic connections within each interval. The control section is the sole coupling node between intervals, avoiding direct hydraulic coupling across intervals. This achieves structural dimensionality reduction for the watershed's water project group. The hydraulic influence weighting calculation formula is as follows:

[0082]

[0083] In the formula, The weight of the hydraulic influence of upstream water project i on downstream water project j; The outflow rate (m³ / s) of the upstream water project after a time lag of T. (For flood evolution time lag); K is the river flow propagation coefficient (0) ); For downstream water projects j in Inflow rate at any given time (m³ / s).

[0084] 3. Inter-section water engineering unit sorting: sort out the list of reservoirs, sluices and flood storage areas in the conventional scheduling interval (upstream of control section 1) and the optimized scheduling interval (downstream of control section 1 to upstream of control section 2), clarify the input and output parameters, engineering constraint thresholds and hydraulic connection relationships of water engineering in each interval, and form two independent scheduling units that only realize data interaction through control sections.

[0085] Step S3, Construction of the differentiated scheduling model:

[0086] Scheduling models were constructed for both conventional and optimized scheduling intervals. The conventional scheduling model followed the established engineering rules, while the optimized scheduling model reconstructed the objective function and constraints, with the core objective being to maximize the remaining flood control capacity, while retaining fundamental constraints such as engineering safety and hydraulic characteristics. The specific implementation is as follows:

[0087] 2.1 Conventional Scheduling Model (Upstream Section of Control Section 1):

[0088] The existing conventional scheduling rules of the watershed water conservancy project group are directly adopted. The flood control scheduling plan and the correspondence between water level and discharge flow are the core of the project design. The scheduling trigger thresholds (such as the flood limit water level of the reservoir and the activation threshold of the flood storage and detention area), gate opening and closing rules, and discharge flow allocation principles are clearly defined. There is no need to build an additional mathematical model. Only the conventional scheduling parameters of the project need to be entered into the calculation system as the basic model for parallel computing.

[0089] 2.2 Optimized scheduling model (downstream of control section 1 - upstream of control section 2):

[0090] With the core objective of maximizing the remaining flood control capacity of the water conservancy project group at the end of the scheduling period while ensuring project safety, and incorporating the back-calculation results of inter-regional floods, basic constraints such as project safety, hydraulic balance, and flow variation are set to complete the construction of an optimized scheduling model, specifically including:

[0091] 2.2.1 Inverse calculation of inter-regional floods:

[0092] By combining the discharge flow at control section 1 and the measured hydrological data at control section 2, the interval flood inversion for the optimized scheduling interval is completed using the Muskinggan method, providing hydrological boundary conditions for the optimized scheduling model. The formula is as follows:

[0093]

[0094] In the formula, To optimize the flood process between water project i and control section 2 within the scheduling interval, m 3 / s; To control the flow rate at time t at section 2, m 3 / s; , , This represents the flood evolution coefficient; To optimize the outflow from water project i at time t within the scheduling interval, m 3 / s.

[0095] 2.2.2 Objective Function Construction (Maximizing Remaining Flood Control Capacity):

[0096] The scheduling objective is to maximize the sum of the remaining flood control capacity of all reservoirs within the optimized scheduling interval at the end of the scheduling period, thereby ensuring the regulation capacity of subsequent flood control scheduling in the basin. The objective function is as follows:

[0097]

[0098] In the formula, The sum of remaining flood control capacity of the reservoir group within the optimized scheduling interval (m 3 ); n represents the number of reservoirs participating in the scheduling within the optimized scheduling interval; The upper limit of the flood control capacity of reservoir i (m³, determined by the difference in capacity between the flood control limit water level and the highest flood control water level); Let be the actual storage capacity (m³) of reservoir i at the end of the scheduling period (time period T).

[0099] 2.2.3 Setting Constraints:

[0100] Under the core objective of maximizing remaining flood control capacity, multi-dimensional constraints are set to ensure that the scheduling process conforms to engineering safety and actual hydraulic characteristics, as follows:

[0101] 1) Water level constraints: The water level of the water project is controlled within the allowable range to ensure the safety of the project structure. The water level at the end of the period conforms to the requirements of the scheduling plan. The calculation formula is as follows:

[0102]

[0103] In the formula, , Water projects The minimum and maximum water levels allowed to be reached during the scheduling process, in meters (m); For water engineering At the end of the scheduling period (the first The actual water level (in meters) during the specified time period. The preset target water level at the end of the water project period is in meters.

[0104] 2) Flow constraints: These include constraints on the discharge capacity of water projects, the safe discharge capacity of downstream river channels, and the safe flow constraints at control sections. The calculation formulas are as follows:

[0105]

[0106]

[0107]

[0108] In the formula, Let m be the actual discharge flow rate of the i-th water project at time t. 3 / s, For water engineering Maximum discharge capacity, m 3 / s, For the downstream of the water project The flow rate at time t at each control section, m 3 / s; For water engineering The safe discharge capacity of the downstream river channel, m 3 / s; For the safe flood discharge of the j-th control section downstream, m 3 / s.

[0109] 3) Water balance constraint: Ensure water conservation during the scheduling process. The calculation formula is as follows:

[0110]

[0111] 4) Hydraulic constraints: The functional relationships between water level and reservoir capacity, and between water level and discharge capacity, must satisfy the water level-reservoir capacity curve and the water level-discharge capacity curve, respectively. The calculation formulas are as follows:

[0112]

[0113]

[0114] In the formula, For water engineering Water level-reservoir capacity relationship function; For water engineering The water level-discharge capacity relationship function.

[0115] 5) Outflow rate variation constraint: To avoid frequent gate opening and closing, the outflow rate variation between adjacent time periods must be controlled within the allowable range. The calculation formula is as follows:

[0116]

[0117] In the formula, For water engineering The maximum allowable flow rate variation within adjacent time periods, m 3 / s.

[0118] Step S4, Regular / Optimized Scheduling of Heterogeneous Parallel Computation:

[0119] By establishing two independent scheduling intervals, a heterogeneous parallel computing framework is constructed. The conventional scheduling model (upstream of control section 1) and the optimized scheduling model (downstream of control section 1 to upstream of control section 2) are assigned to two independent computing nodes to achieve synchronous parallel solution. This avoids the sequential dependency of traditional serial computing and improves the solution efficiency of the scheduling model. The specific implementation is as follows:

[0120] Parallel computing framework construction: A master-slave parallel computing architecture is adopted, with one master control node and two slave computing nodes (node ​​1: conventional scheduling computing, node 2: optimized scheduling computing). The master control node is responsible for initializing computing parameters, issuing computing tasks, monitoring the running status of nodes in real time, and receiving the computing results of nodes. The slave computing nodes independently execute the solution computing of the corresponding scheduling model.

[0121] The main control node receives the results of the division of the watershed hydrological control zone, various constraint parameters and scheduling target parameters, splits the tasks according to the differentiated scheduling requirements, sends the routine scheduling solution tasks to computing node 1, and sends the optimized scheduling solution tasks to computing node 2, and clarifies the calculation scope and data interaction frequency of the two computing nodes.

[0122] Parameter synchronization: The main control node synchronously distributes the watershed hydrological boundary conditions (precipitation, runoff, and inter-regional floods) and basic water engineering parameters (engineering thresholds, water level-storage capacity curves, and water level-discharge capacity curves) to the two slave calculation nodes, unifying the scheduling decision period (4 hours) and the total calculation time to ensure that the calculation time base of the two nodes is consistent.

[0123] Heterogeneous parallel solution execution:

[0124] Computation Node 1: Based on conventional scheduling rules, it substitutes the hydrological and engineering parameters of the upstream section of control section 1, calculates the scheduling results such as water level, discharge flow, and flood storage and detention area activation status of each water project in the section in time intervals, and uploads the discharge flow process of control section 1 to the main control node in real time.

[0125] Computation Node 2: Particle Swarm Optimization (PSO), Genetic Algorithm (GA), or hybrid intelligent optimization algorithm is used to solve the optimal scheduling model. With the goal of maximizing the remaining flood control capacity, the optimal scheduling scheme of each water project in the interval (time-by-time water level, discharge flow, and storage capacity changes) is obtained through iterative optimization. During the solution process, the discharge flow process of control section 1 issued by the main control node is used as the upstream boundary input to ensure hydraulic coupling with the conventional scheduling interval.

[0126] Node fault tolerance: When the main control node starts up, it performs running status detection and communication link testing on itself and the two computing nodes. If an abnormality occurs, it pauses the startup and troubleshoots the fault.

[0127] The master control node monitors the running status of the two slave computing nodes in real time. If a node experiences a computing bottleneck or failure, the redundancy computing mechanism is immediately activated to redistribute the node's tasks to the backup computing node, ensuring that the parallel computing process is continuous and uninterrupted.

[0128] Parallel computing results summary: After the two slave computing nodes complete their respective scheduling model solutions, they upload the scheduling results (water level, flow rate, and storage capacity of each water project at different time intervals, and flow process of control sections 1 and 2) to the main control node, and the main control node completes the preliminary result collection and organization.

[0129] Step S5, XGBoost algorithm for river flood evolution calculation:

[0130] After the main control node aggregates the parallel computing results, it uses the XGBoost algorithm to replace the traditional Muskingan method to perform flood evolution calculations across the entire basin, achieving rapid and high-precision prediction of river cross-section water level and flow. Simultaneously, it couples the hydraulic results of the conventional and optimized scheduling intervals. Specific implementation details are as follows:

[0131] Sample dataset construction: The collected historical flood monitoring data of the basin, including inflow / outflow, water level, channel roughness, cross-sectional size, flood evolution time lag, etc., of each cross-section of the river, combined with the inverse calculation results of the interval flood and the parallel scheduling solution results, are used to construct the training and testing samples of the XGBoost model. Among them, the sample features include upstream cross-section flow / water level and scheduled discharge flow, and the sample label is the flow / water level of the corresponding time of the downstream cross-section.

[0132] The discharge flow at the control section and the discharge flow of the water project in the optimized scheduling section obtained by parallel computing are used as model inputs. Combined with the water level data of the upstream section, the model predicts the flow and water level at the corresponding time of the downstream section, thus completing the prediction of river evolution.

[0133] The model employs conventional hyperparameters for reasonable configuration: setting appropriate maximum tree depth and learning rate, introducing L1 and L2 regularization and row and column random sampling strategies to suppress overfitting and improve the model's generalization ability; and dynamically optimizing the number of iterations in conjunction with an early stopping mechanism.

[0134] The sample library constructed according to the 8:2 ratio is divided into a training set and a validation set. The training set is input into the XGBoost machine learning model to complete the training, and the validation set is input into the trained model for validation.

[0135] Choose three core metrics to evaluate the performance of model training and validation:

[0136] (1) Coefficient of determination

[0137] The model's fit to the data is measured using a value in the range [0,1]. Values ​​closer to 1 indicate a better fit. The calculation formula is as follows:

[0138]

[0139] In the formula: The mean of the true values; The leakage flow rate corresponding to the sample; is the leakage flow rate output by the model; n is the number of samples.

[0140] (2) Root Mean Square Error (RMSE)

[0141] Measure the average deviation between predicted and actual values, with units consistent with the output. The calculation formula is:

[0142]

[0143] (3) Mean Absolute Error (MAE)

[0144] Measured as the absolute deviation between the predicted and actual values, in units of 1. The calculation formula is:

[0145] .

[0146] The discharge flow at control section 1 and the discharge flow of water projects in the optimized scheduling section obtained by parallel computing are used as upstream input features of the XGBoost model and substituted into the trained and calibrated XGBoost channel evolution model.

[0147] 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 solving the joint scheduling of intelligent acceleration-heterogeneous parallel watershed water engineering groups, characterized in that, Includes the following steps: Collect and organize basic watershed data, including engineering and hydrological data, to ensure data consistency in time sequence and perform preprocessing. Based on the natural water system network, spatial distribution of water conservancy projects, and hydraulic coupling relationships of the basin, and combined with the pre-set control section boundaries, the division of the basin's hydrological control zone is completed; Determine the strength of hydraulic connections between water projects, ensure close hydraulic connections between water projects in each section, and use the control section as the only coupling node in each section to avoid direct hydraulic coupling across sections. Scheduling models are constructed for both conventional and optimized scheduling intervals. The conventional scheduling model follows the established engineering rules, while the optimized scheduling model reconstructs the objective function and constraints. The Muskingu method was used to back-calculate flood data for the optimized scheduling interval, providing hydrological boundary conditions for the optimized scheduling model. The optimization objective is to maximize the remaining flood control capacity, while respecting engineering safety and hydraulic constraints. A heterogeneous parallel computing framework is built, which allocates the conventional scheduling model and the optimized scheduling model to two independent computing nodes to achieve synchronous parallel solution, avoids the sequential dependency of traditional serial computing, and improves the solution efficiency of the scheduling model. Define the functions of the main control node, computing node 1, and computing node 2. The main control node is responsible for parameter initialization, task distribution, status monitoring, and result summarization. Computing node 1 is responsible for solving the conventional scheduling model, and computing node 2 is responsible for solving the optimized scheduling model. The main control node receives the results of the division of the watershed hydrological control zone, constraint parameters, and scheduling target parameters. It splits the tasks according to the differentiated scheduling requirements, distributes the routine scheduling solution tasks to computing node 1, distributes the optimized scheduling solution tasks to computing node 2, and clarifies the calculation scope and data interaction frequency of the two computing nodes. The two computing nodes receive task instructions and parameters, independently start the corresponding scheduling model to solve the problem, and report the running status back to the main control node. The master control node monitors the running status of the two computing nodes in real time and performs corresponding processing operations for abnormal situations to ensure computing synchronization; After the two computing nodes complete their calculations, they upload the results, and the main control node merges and outputs the calculation results. After the main control node aggregates the parallel computing results, it uses the XGBoost algorithm to perform flood evolution calculations across the entire basin, enabling the prediction of water level and flow at river sections and completing the coupling of hydraulic results between the conventional and optimized scheduling intervals.

2. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 1, characterized in that, The division of the watershed hydrological control zone is based on control section 1 as the conventional scheduling control boundary, with its upstream area classified as the conventional scheduling zone, including all reservoirs, sluices and supporting flood storage and detention areas within the zone; and control section 2 as the optimized scheduling control boundary, with its upstream area including the region from downstream of control section 1 to upstream of control section 2, classified as the optimized scheduling zone, including the water conservancy projects within the zone.

3. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 1, characterized in that, The strength of the hydraulic correlation between water projects is determined by the hydraulic influence weight, and the calculation formula is as follows: In the formula, The weight of the hydraulic influence of upstream water project i on downstream water project j; The outflow from the upstream water project is the time interval T after the upstream water project lags behind. The flood evolution time lag; K is the river flow propagation coefficient, 0 ; For downstream water projects j in Inbound traffic at any given time.

4. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 2, characterized in that, The hydrological boundary conditions are calculated using the following formula: In the formula, To optimize the flood process between water project i and control section 2 within the scheduling interval; To control the flow rate at time t at section 2; , , This represents the flood evolution coefficient; To optimize the outflow of water project i at time t within the scheduling interval.

5. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 1, characterized in that, The objective function for maximizing the remaining flood control capacity is as follows: In the formula, The sum of the remaining flood control capacity of the reservoir group within the optimized scheduling interval; n is the number of reservoirs participating in the scheduling within the optimized scheduling interval; This represents the upper limit of the flood control capacity of reservoir i; Let be the actual storage capacity of reservoir i at the end of the scheduling period and in time period T. The hydraulic characteristic constraints include water level constraints, and the calculation formula is as follows: In the formula, , Water projects The minimum and maximum water levels that are allowed to be reached during the scheduling process; For water engineering At the end of the scheduling period, the The actual water level during the period; The preset target water level at the end of the water project period; Flow constraints include the discharge capacity constraints of water projects, the safe discharge capacity constraints of downstream river channels, and the safe flow constraints of control sections. The calculation formulas are as follows: In the formula, Let be the actual discharge flow of the i-th water project at time t. For water engineering Maximum discharge capacity For the downstream of the water project The flow rate at time t of each control section; For water engineering The safe discharge capacity of the downstream river channel; The safe flood discharge flow rate for the j-th control section downstream; The water balance constraint is calculated using the following formula: in, For water engineering The water storage volume at time t; Water projects Inflow rate during time period t; Water projects Outflow rate during time period t; To calculate the time step; Hydraulic constraints are calculated using the following formula: In the formula, For water engineering The water storage at time t, For water engineering The water surface elevation at time t; For water engineering Water level-reservoir capacity relationship function; For water engineering The water level-discharge capacity relationship function; The formula for calculating the outbound flow rate variation constraint is as follows: In the formula, Water projects Outflow rate during time period t; For water engineering The maximum allowable flow rate variation within adjacent time periods.

6. The intelligent acceleration-heterogeneous parallel watershed water conservancy project group joint scheduling solution method according to claim 1, characterized in that, The main control node synchronously distributes the watershed hydrological boundary conditions and basic parameters of water engineering to the two calculation nodes, and coordinates the decision-making period and the total calculation time.

7. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 6, characterized in that, The watershed hydrological boundary conditions include precipitation, runoff, and inter-regional floods; the basic parameters of the water project include engineering thresholds, water level-storage capacity curves, and water level-discharge capacity curves; the scheduling decision period is 4 hours.

8. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 1, characterized in that, The abnormal situation is that if a node experiences a computational lag or failure, the redundancy mechanism is immediately activated to redistribute the node's tasks to a backup computing node, ensuring that the parallel computing process is continuous and uninterrupted.

9. The intelligent acceleration-heterogeneous parallel watershed water engineering group joint scheduling solution method according to claim 1, characterized in that, The XGBoost model construction includes: A sample dataset is constructed by combining the collected historical flood monitoring data of the basin with the results of inverse flood calculation and parallel scheduling solution to determine the sample features and sample labels, forming a sample library required for model training and testing. The sample features include upstream section flow, water level and scheduled discharge flow, and the sample labels are the flow and water level at the corresponding time at the downstream section. Set the key hyperparameters of the XGBoost model, divide the sample dataset into training and validation sets according to a preset ratio, train the XGBoost model, verify the model performance through core evaluation metrics, and complete the model calibration. The discharge flow at the control section, the discharge flow of the water project in the optimized scheduling section, and the water level at the upstream section obtained by parallel computing are used as inputs to the XGBoost model, and then substituted into the model to predict the flow and water level at the corresponding time downstream.

10. The intelligent acceleration-heterogeneous parallel watershed water conservancy project group joint scheduling solution method according to claim 9, characterized in that, The key hyperparameters of the XGBoost model include the learning rate, maximum tree depth, L1 regularization parameter, and L2 regularization parameter. An optimization strategy of random sampling of samples and features is adopted, combined with an early stopping mechanism to dynamically optimize the number of iterations. The preset ratio of the sample dataset is 8:2, that is, the training set accounts for 80% of the sample dataset, and the validation set accounts for 20%. The preset core evaluation metrics include the coefficient of determination R², root mean square error (RMSE), and mean absolute error (MAE).

Images