Intelligent scheduling method and device of porous gate and server

Abstract

The application provides a kind of intelligent scheduling method, device and server of multi-hole gate, it is related to the technical field of water conservancy engineering automation and artificial intelligence optimization scheduling, comprising: using the historical operation data and simulation data of multi-hole gate, determine target input characteristics, and through light gradient boosting tree model, target input characteristics are carried out nonlinear regression training processing, determine the target flow prediction model of current time;Through multi-constraint search model, the gate opening combination is carried out hierarchical expansion processing, and the gate expansion scheduling scheme is obtained, and the target flow prediction model is used to carry out batch prediction processing and scoring processing on the gate expansion scheduling scheme, to obtain the gate candidate scheduling scheme set;Upper limit detection screening processing and error tolerance screening processing are carried out for the gate candidate scheduling scheme set, to determine the target scheduling scheme of multi-hole gate.The application can significantly improve the prediction accuracy of gate water outlet and the scheduling efficiency of multi-hole gate.

Description

Technical Field

[0001] This invention relates to the technical field of automation and artificial intelligence-based optimized scheduling of water conservancy projects, and in particular to an intelligent scheduling method, device and server for multi-hole gates. Background Technology

[0002] With the widespread application of large-scale water conservancy projects and multi-gate systems, gate scheduling plays a crucial role in flood control, power generation efficiency, and ecological regulation. Currently, related technologies propose gate scheduling based on physical model-based numerical simulation or intelligent optimization and machine learning methods. However, physical model-based numerical simulation methods involve massive computational loads, while intelligent optimization and machine learning methods suffer from low scheduling efficiency and insufficient accuracy when dealing with multi-gate combinations, resulting in poor practical feasibility of the scheduling schemes. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an intelligent scheduling method, device and server for multi-hole gates, which can significantly improve the prediction accuracy of gate water discharge and the scheduling efficiency of multi-hole gates.

[0004] In a first aspect, embodiments of the present invention provide an intelligent scheduling method for multi-orifice gates. The method includes: using historical operating data and simulation data of the multi-orifice gates to determine target input features, and using a lightweight gradient boosting tree model to perform nonlinear regression training on the target input features to determine the target flow prediction model at the current moment; using a multi-constraint search model to perform hierarchical expansion processing on the gate opening combinations to obtain gate expansion scheduling schemes, and using the target flow prediction model to perform batch prediction processing and scoring processing on the gate expansion scheduling schemes to obtain a set of gate candidate scheduling schemes; and performing upper limit detection screening processing and error tolerance screening processing on the set of gate candidate scheduling schemes to determine the target scheduling scheme for the multi-orifice gates.

[0005] In one implementation, the step of determining the target input features using historical operation data and simulation data of a multi-hole gate includes: constructing a sample database based on the historical operation data and simulation data; performing feature combination processing and nonlinear transformation processing on each input feature in the sample database to execute physical prior extension of each input feature, obtaining corresponding higher-order features, and combining the higher-order features to generate the target input features.

[0006] In one implementation, the step of determining the target traffic prediction model at the current moment by performing nonlinear regression training on the target input features using a lightweight gradient boosting tree model includes: using the lightweight gradient boosting tree model, based on five-fold cross-validation and a preset early stopping mechanism, filtering a preset set of candidate hyperparameters to determine the target hyperparameters, wherein the target hyperparameters are the parameter combinations with the smallest error during five-fold cross-validation; and performing nonlinear regression training on the target input features based on the target hyperparameters to obtain the target traffic prediction model.

[0007] In one implementation, after obtaining the target traffic prediction model, the method includes: using a prediction error adaptive correction model to correct the predicted traffic output by the target traffic prediction model by combining historical prediction errors, so as to perform real-time dynamic adjustment of the predicted traffic of the target traffic prediction model.

[0008] In one implementation, before the step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain the gate expansion scheduling scheme, the method includes: performing a retrieval process in a preset gate rule knowledge base for the predicted flow output by the target flow prediction model to determine the gate opening combination that can be executed under the current operating conditions and the minimum opening template vector.

[0009] In one implementation, the step of performing hierarchical expansion processing on gate opening combinations using a multi-constraint search model to obtain a gate expansion scheduling scheme includes: performing hierarchical expansion processing on gate opening combinations under the constraints of gate opening combinations and minimum opening template vectors using a multi-constraint search model to generate gate opening combinations corresponding to each layer of expansion, and determining each gate opening combination as a gate expansion scheduling scheme.

[0010] In one implementation, the step of performing upper limit detection screening and error tolerance screening on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate includes: setting an equal opening sequence in each candidate gate scheduling scheme in the set of candidate gate scheduling schemes, and performing flow prediction to obtain the maximum flow corresponding to each candidate gate scheduling scheme; when the maximum flow corresponding to the candidate gate scheduling scheme is less than the target flow, and the difference between the predicted flow and the target flow is not greater than a preset tolerance band, the candidate gate scheduling scheme is determined as the target scheduling scheme.

[0011] Secondly, embodiments of the present invention also provide an intelligent scheduling device for a multi-hole gate. The device includes: a model update module, which uses historical operating data and simulation data of the multi-hole gate to determine target input features, and uses a lightweight gradient boosting tree model to perform nonlinear regression training on the target input features to determine the target flow prediction model at the current moment; a constraint search module, which uses a multi-constraint search model to perform hierarchical expansion processing on the gate opening combination to obtain a gate expansion scheduling scheme, and uses the target flow prediction model to perform batch prediction processing and scoring processing on the gate expansion scheduling scheme to obtain a set of gate candidate scheduling schemes; and a scheme screening module, which performs upper limit detection screening processing and error tolerance screening processing on the set of gate candidate scheduling schemes to determine the target scheduling scheme for the multi-hole gate.

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

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

[0014] The embodiments of the present invention bring the following beneficial effects:

[0015] This invention provides an intelligent scheduling method, device, and server for multi-orifice gates. The method first uses historical and simulation data of the multi-orifice gates to determine target input features. Then, a lightweight gradient boosting tree model is used to perform nonlinear regression training on these features to determine the target flow prediction model for the current moment. Next, a multi-constraint search model is used to perform hierarchical expansion processing on the gate opening combinations to obtain an expanded gate scheduling scheme. The target flow prediction model is then used to perform batch prediction and scoring processing on the expanded gate scheduling schemes to obtain a set of candidate gate scheduling schemes. Finally, upper limit detection and error tolerance screening processing are performed on the candidate gate scheduling scheme set to determine the target scheduling scheme for the multi-orifice gate. This invention can significantly improve the prediction accuracy of gate discharge and the scheduling efficiency of multi-orifice gates.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an intelligent scheduling method for a multi-hole gate provided in an embodiment of the present invention;

[0020] Figure 2 A schematic diagram illustrating the specific process of an intelligent scheduling method for a multi-hole gate provided in an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the structure of an intelligent scheduling device for a multi-hole gate provided in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of the structure of a server provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] At present, the core task of gate group scheduling is to determine the reasonable combination of gate openings based on upstream and downstream water levels and target flow rates, so as to achieve safe, stable and economical joint operation. However, in actual engineering, this process is affected by multiple factors such as hydraulic coupling, structural constraints and regulations, and exhibits strong nonlinearity, high dimension and multiple constraints. Traditional empirical methods and single optimization algorithms are difficult to meet the real-time, accurate and executable scheduling requirements. Existing gate scheduling methods mainly include: (1) numerical simulation methods based on physical models, such as using computational fluid dynamics (CFD) to simulate the discharge and energy dissipation process. This method can accurately characterize the flow characteristics of gate jet and confluence zone, but the computational load is huge and sensitive to boundary conditions, making it difficult to support online real-time calculation and dynamic decision-making. (2) methods based on intelligent optimization and machine learning, such as genetic algorithms, particle swarm optimization, neural networks and reinforcement learning models. Although this method can improve the prediction accuracy to a certain extent, it generally has the following problems:

[0025] 1. The inability to explicitly express engineering procedures and strong constraints leads to unexecutable scheduling results. Existing intelligent methods often perform continuous optimization with the goal of global optimization, but lack a mechanism to express hard constraints on engineering procedures such as minimum opening, odd and even gate position linkage, partition priority, maintenance, and shielding. In the strongly constrained discrete optimization scenario of multi-hole gates, it is not possible for the optimization results to be feasible only in terms of algorithm, but unexecutable in terms of engineering, requiring manual re-verification or even recalculation.

[0026] 2. It is difficult to handle the exponential search space generated by multiple gate combinations, resulting in low search efficiency. For an 18-gate system, there are a large number of opening combinations. Traditional intelligent algorithms (such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Reinforcement Learning (RL)) usually rely on random perturbations or gradient exploration, which can easily get stuck in local optima. Furthermore, they have a large number of iterations and slow convergence, making it difficult to provide a usable solution within the second-level response time required by actual scheduling, resulting in insufficient real-time performance.

[0027] 3. Lack of interpretability and scheduling stability. Most neural network or reinforcement learning models output black box results, which cannot explain why a certain set of gates is opened. They are difficult to use directly as a scheduling basis in engineering operation. At the same time, their output is sensitive to input noise and does not have the stability expected in engineering such as minimal changes to symmetry and center priority.

[0028] 4. The prediction model does not incorporate prior engineering physics knowledge, resulting in insufficient accuracy under multi-hole coupling conditions. Most existing learning-based prediction models are based only on basic water level and opening input, making it difficult to accurately characterize the nonlinear coupling effect in the confluence zone of multi-hole gates. In particular, the error is large in complex conditions (such as transitional flow and submerged flow), which limits the reliability of subsequent optimization.

[0029] Therefore, there is an urgent need for an intelligent scheduling method that can balance flow prediction accuracy, constraint expression capability, and search efficiency while ensuring engineering procedures and structural safety. Compared with existing solutions, the intelligent scheduling method, device, and server for multi-hole gates provided by this invention have the following significant differences:

[0030] 1. A scheduling framework that integrates data-driven and procedural constraints (not available in existing methods): This invention explicitly encodes engineering procedures in the form of templates, sets, hard constraints, etc., reducing the search space from exponential combinations to an executable strategy space. It realizes the construction of feasible domains by first establishing procedures and then optimizing, ensuring that all results are 100% executable.

[0031] 2. This invention proposes a multi-constraint search (BeamSearch) instead of traditional GA or PSO, which improves efficiency by an order of magnitude. Through mechanisms such as hierarchical expansion, batch prediction, and neighborhood pruning, each round of optimization searches only within the feasible domain of the procedure. Compared with the strong randomness and slow convergence of GA / PSO, this invention can find the optimal opening scheme within 1 to 3 seconds, meeting the real-time scheduling requirements.

[0032] 3. By introducing engineering stability preferences such as minimal modification of symmetry center priority, this invention explicitly models the interpretability and stability of scheduling actions through a scoring function, avoiding frequent gate opening and closing, and improving the credibility of the engineering department. Traditional intelligent algorithms cannot guarantee these structured preferences.

[0033] 4. The flow prediction model has physical consistency and the ability to express different working conditions. Its accuracy is significantly higher than that of existing learning methods. Based on physical prior features (such as head energy, section opening, image difference, relative submergence ratio, etc.) and a segmented training mechanism for different working conditions (free / transition / submerged flow), this invention greatly improves the prediction accuracy under multi-pore coupling conditions (error not greater than 2%), which is better than ordinary neural networks or single regression models.

[0034] See Figure 1 The diagram shows a flowchart of an intelligent scheduling method for a multi-hole gate. The method mainly includes the following steps S102 to S106:

[0035] Step S102: Using historical operation data and simulation data of the multi-hole gate, determine the target input features, and use a lightweight gradient boosting tree model to perform nonlinear regression training on the target input features to determine the target flow prediction model at the current moment.

[0036] In one implementation, historical data modeling and sample construction can be performed first, followed by feature expansion, and finally, a traffic prediction model based on LightGBM can be established. The target traffic prediction model needs to be dynamically updated at different times. Specifically:

[0037] First, historical operational data of the project was collected and integrated, coupled with some simulation data, including the water level upstream of the dam, the water level downstream of the dam, the opening degree of each gate, and the corresponding measured discharge flow. Through data cleaning, anomaly removal, and time alignment, a high-quality sample database was established, providing a reliable training data foundation for machine learning models. Each record in the dataset represents a specific operating condition and its corresponding flow response.

[0038] To improve the model's ability to fit hydraulic laws, the system extends the original data by adding features such as water level difference, total opening, opening square term, and interaction term between opening and water level difference, in addition to the original water level and gate opening. This reflects the inherent coupling relationship between flow rate and head, cross-sectional area and energy loss. This process preserves the physical meaning and enhances the model's generalization performance.

[0039] After feature construction, the system uses the LightGBM algorithm (i.e., Lightweight Gradient Boosting Tree Model) for modeling. LightGBM is an efficient gradient boosting tree model suitable for nonlinear regression problems. It determines the optimal parameters through five-fold cross-validation and an early stopping mechanism, enabling the model to accurately predict the discharge flow of the gate group under different operating conditions. This model learns complex nonlinear relationships and can complete flow prediction in milliseconds. The five-fold cross-validation means that 4 / 5 of the data is used for training at each fold, and the remaining 1 / 5 of the data is used for validation. The RMSE is calculated on the validation set, and the metric for each iteration is recorded. If the RMSE does not decrease within a preset number of iterations, training is terminated, and the system rolls back to the historical best iteration point.

[0040] In summary, by combining machine learning models with engineering operation procedures, the method achieves an organic fusion of data-driven prediction and rule-constrained optimization. This method can not only automatically learn the nonlinear flow relationship between the upstream water level, downstream water level, and gate opening, but also ensure that the output results meet the requirements of the procedures. It solves the technical problem that traditional experience-based scheduling relies on manual judgment and is difficult to balance accuracy and feasibility.

[0041] Step S104: Through a multi-constraint search model, the gate opening combination is subjected to hierarchical expansion processing to obtain a gate expansion scheduling scheme. Then, the target flow prediction model is used to perform batch prediction and scoring processing on the gate expansion scheduling scheme to obtain a set of gate candidate scheduling schemes.

[0042] In one implementation, after model training, the inference phase begins. Based on real-time water levels upstream and downstream of the dam, and combinations of gate openings, the trained model predicts the discharge flow. To improve computational efficiency, the system employs both single-line and batch prediction modes, enabling rapid calculation of flow values ​​for hundreds or thousands of opening combinations. This provides an efficient objective evaluation function for subsequent optimization algorithms. Following this, operational procedures and minimum opening templates are generated: based on engineering operational procedures and zoning structures, the system automatically determines the permissible gate combinations for each water level and flow range. For example, different sets of permissible gate numbers correspond to different upstream water level ranges. For each gate combination, the system generates a minimum opening template to ensure all candidate opening schemes meet minimum opening / closing requirements and structural safety constraints. This template transforms manual procedures into computable constraints, forming the basis for optimization solutions.

[0043] In addition, during operation, if some gates are under maintenance, the system will automatically identify these gate numbers and remove them from the set of openable gates. If the key gate positions required by the regulations are blocked, other gate positions can be automatically supplemented according to the principle of mirror symmetry or same-area substitution, thereby ensuring the integrity and executability of the scheduling combination. This mechanism effectively avoids human input errors and unsafe operations.

[0044] Furthermore, a multi-constraint heuristic bundle search optimization is performed: Under the premise of satisfying the procedures and maintenance constraints, a heuristic bundle search algorithm is used to optimize the gate opening. This algorithm takes the error between the predicted flow and the target flow as the main objective, while comprehensively considering multiple additional constraints, including: symmetry constraint: ensuring the balance of gate opening on both banks; center priority constraint: prioritizing the opening and closing of the center gate of the structure; minimal modification constraint: maintaining similarity to the previous scheduling as much as possible; equal opening constraint: balancing the opening differences between the opening gates.

[0045] In summary, the multi-constraint search optimization algorithm simultaneously considers multiple objectives in the search space, including flow error, symmetry, center priority, minimal modification preference, and equal opening balance. It achieves efficient global search through dynamic bundle width and neighborhood quantization mechanisms. Compared to traditional genetic algorithms or exhaustive search, this method improves computational efficiency by an order of magnitude, and the results are stable and reliable. Furthermore, the introduction of a minimum opening template and an automatic gate shielding mechanism at the algorithm level makes the scheduling scheme physically feasible and safety-controllable. Even if some gates are under maintenance, the system can still automatically generate effective opening combinations based on structural symmetry or the principle of substitution within the same area, ensuring operational continuity and system robustness.

[0046] Step S106: Perform upper limit detection screening and error tolerance screening on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate.

[0047] In one implementation, during the search process described above, the system dynamically retains the top few optimal solutions among multiple candidate solutions, continuously expands and updates them in the neighborhood space, and gradually approaches the globally optimal opening combination, thereby determining the target scheduling scheme for the multi-hole gate. By introducing adaptive error tolerance and early stopping strategies, it can control with relative error under high flow conditions and with absolute error under low flow conditions, achieving intelligent early stopping and optimal utilization of computing resources, and significantly improving real-time scheduling performance.

[0048] The intelligent scheduling method for multi-gate systems provided in this invention can accurately predict the gate discharge process by introducing a lightweight machine learning model, realizing the nonlinear relationship modeling between upstream and downstream water levels, gate opening degree, and outflow. Simultaneously, it constructs parameterized engineering specifications and gate operation templates, embedding requirements such as minimum opening degree, maintenance shielding, structural symmetry, and safety margin into the optimization solution process as constraints. Based on this, an improved multi-constraint search optimization algorithm is used to quickly search and sort combinations of multi-gate opening degrees, ensuring that the scheduling result approximates the target discharge flow while meeting specification requirements. This significantly improves the intelligence and real-time response capability of gate group scheduling, realizing a shift from experience-based decision-making to data-driven and specification-constrained integration. Compared with traditional methods, this invention, while ensuring safety and compliance with specifications, controls the prediction error within 2% and shortens the computation time by an order of magnitude, demonstrating good engineering applicability and promotional value.

[0049] See Figure 2 The diagram shows a specific flowchart of an intelligent scheduling method for multi-hole gates. This embodiment of the invention also provides an implementation method for intelligent, refined, and safe control of multi-hole hydraulic gate groups under complex working conditions by introducing a scheduling framework that combines machine learning and multi-constraint search optimization. This method can transform the scheduling process of hydraulic gates from experience-driven to an automated decision-making process that combines data-driven, procedural constraints, and intelligent optimization. Under the premise of meeting engineering constraints such as safety, symmetry, and balance, it can achieve rapid, accurate, and interpretable optimal scheduling of gate groups, greatly improving scheduling efficiency and reliability. This provides key technical support for the digitalization and intelligentization of water conservancy project operation. See (1) to (4) below for details:

[0050] (1) Data Modeling and Feature Construction. Based on historical operation data and simulation data, a sample database is constructed. By performing feature combination processing and nonlinear transformation processing on each input feature in the sample database, physical prior extension of each input feature is performed to obtain the corresponding higher-order features. The higher-order features are then combined to generate the target input features. The historical operation data comes from historical operation records and includes the water level in front of the dam, the water level downstream of the dam, and 1– Gate opening and measured discharge flow:

[0051]

[0052] in, For the first Water level in front of the dam for each sample (m); For the first Water level (m) downstream of the dam for each sample; For the first The first sample Gate opening (m); For the first The measured flow rate (m³ / s) of each sample.

[0053] In one implementation, to enhance the model's ability to express hydrodynamic relationships, feature combination and nonlinear transformation are applied to the input variables. Key features include: water level difference. Total opening: Total square root term: Interaction items (reflecting the combined effect of the gate and the water head): .

[0054] Combined to form the input feature vector:

[0055]

[0056] in, The feature vector input to the model; For the first The first sample The opening degree of the sluice gate (m); For the first Water level difference (m) of each sample; For the first Total opening at any given time (m); For the first The square of the total opening at any given moment (㎡); For the first The product of the opening degree and the head difference of each sample (m²).

[0057] (2) Using the Lightweight Gradient Boosting Tree (LightGBM) model, based on five-fold cross-validation and a pre-defined early stopping mechanism, the pre-defined hyperparameter candidate set is screened to determine the target hyperparameters. Then, based on the target hyperparameters, nonlinear regression training is performed on the target input features to obtain the target traffic prediction model. The target hyperparameters are the parameter combinations with the smallest error during five-fold cross-validation. The objective function of the Lightweight Gradient Boosting Tree (LightGBM) model is:

[0058]

[0059] in, For a set of optional model functions, This is a regularization term used to prevent overfitting. This structure can handle nonlinear, high-dimensional features and enables fast training and prediction.

[0060] To further improve prediction accuracy, this invention introduces an enhanced regularization strategy:

[0061]

[0062] in, Let be the leaf weights of tree t. Output values ​​for leaf nodes. This is an adjustable regularization parameter. This innovation allows the model to maintain its ability to represent complex features while avoiding overfitting to extreme combinations of opening degrees.

[0063] In addition, the model training employs cross-validation and early stopping mechanisms:

[0064]

[0065] The optimal hyperparameters, including the number of leaf nodes, are selected through multi-fold cross-validation. Tree depth Learning rate This allows for a balance between training efficiency and prediction accuracy.

[0066] In one implementation, an adaptive correction model for prediction errors can be used to correct the predicted flow output by the target flow prediction model by incorporating historical prediction errors. This allows for real-time dynamic adjustment of the predicted flow of the target flow prediction model. In other words, during the inference phase, the model can adjust the predicted flow for any given gate opening vector. With water level status Perform millisecond-level predictions:

[0067]

[0068] To enhance engineering adaptability, this invention introduces an adaptive correction mechanism for prediction errors:

[0069]

[0070] in, This is due to historical prediction errors. These are adaptive correction coefficients. This mechanism can dynamically adjust the prediction output during long-term operation, ensuring that the error is stably controlled within ±2%, while providing a reliable objective evaluation for subsequent multi-constraint search optimization.

[0071] (3) For the predicted flow output by the target flow prediction model, the gate rule knowledge base is searched and processed to determine the gate opening combination and minimum opening template vector that can be executed under the current working condition. Then, through the multi-constraint search model, under the constraints of the gate opening combination and the minimum opening template vector, the gate opening combination is extended in layers to generate the gate opening combination corresponding to each layer of extension, and each gate opening combination is determined as the gate extension scheduling scheme.

[0072] Specifically, to ensure that the scheduling plan complies with the engineering operation procedures, the system retrieves the allowed gate opening combinations under the current operating conditions from the gate procedure knowledge base based on the water level in front of the dam, the target flow, and the operating status. The procedure knowledge base may include: gate priority and sequence strategy, zonal symmetry and linkage rules, safety, hydraulic, structural and other constraints, and maintenance and operation status information, etc.

[0073] Each procedure item is stored in a parameterized form as conditions, actions, and constraints, and can be adjusted according to different engineering or operational stages. In typical multi-gate projects, the scheduling system usually follows the following general operating strategy:

[0074] 1. Automatically determine the number and combination of gates to be opened based on the upstream water level and target discharge flow; 2. Maintain the balance and synchronization of gate actions during opening and closing to ensure structural safety and flow stability; 3. When the gate group is arranged by region, the main discharge area can be selected for opening and closing operations first, and gradually expanded to other areas according to operational needs to achieve zoned and phased control; 4. The opening and closing sequence within the same area can follow the principle of center priority and side-by-side progression to maintain uniform discharge; 5. When multiple gates are operating in conjunction, an alternating opening and closing strategy can be adopted based on structural symmetry or odd-even distribution characteristics to improve flow coordination and discharge balance.

[0075] Taking a typical multi-gate sluice gate project as an example, the operation process and constraint logic of the above scheme are explained: To ensure that the scheduling scheme complies with the engineering operation procedures, the system adjusts the scheduling based on the water level in front of the dam. Segmentation and target traffic Within the specified interval, retrieve the corresponding set of gate combinations, and then, based on the set... Generate the minimum opening template vector:

[0076]

[0077] It also includes three hard constraints: 1. If the set contains an even number of gates, then the minimum opening of all odd-numbered gates is... 2. If the assembly includes a gate in Zone 1, then the minimum opening degree for the gates in Zone 2 is... 3. If a maintenance gate exists, its corresponding opening is forcibly set to zero and the substitution mechanism is triggered, ultimately resulting in an N-dimensional vector:

[0078]

[0079] And quantized to a 0.5m grid:

[0080]

[0081] in, This refers to the number of gates; Number the gates; For the first Minimum opening value (m) of gate number 1; This represents the set of gates that are allowed to open under the current operating conditions; the constant 0.5 is the minimum opening depth (m) set in the engineering specifications, indicating that when a gate is to be opened, the minimum opening depth is 0.5m; round( The ) indicates a mathematical function for rounding, used to round real values ​​to the nearest integer or multiple.

[0082] Furthermore, unequal open-beam search optimization is performed, as detailed in (A) to (C) below:

[0083] (A) Search for the opening vector within the set allowed by the procedure. With minimizing the target flow deviation as the core objective, and taking into account constraints such as symmetry, center priority, opening balance, and historical similarity, a comprehensive scoring function is established (where each weight parameter can be set according to engineering operation and maintenance experience requirements):

[0084]

[0085] in, This refers to the number of gates; : Comprehensive scoring function; Previous scheduling plan; Peripheral penalty power (the closer to the center, the smaller the penalty); : A set of partition mirror pairs; Weight parameters; :variance; : Total outflow predicted by the model (m³ / s); Target discharge flow rate (m³ / s); L1 norm (sum of absolute differences) (m); : Last scheduling opening vector (m); : Number of symmetric mirror pairs; and : They all represent the opening degree of a single gate, but their meanings differ depending on their position. In the variance term, they represent the set of opening degree values ​​of all gates, used to evaluate whether they are equally open; in the symmetry term, they represent the two opening degrees of mirrored gate pairs on the left and right banks, used to penalize asymmetrical discharge. This represents the absolute error between the model's predicted flow and the target flow. It is the core optimization objective, ensuring that the scheduled flow meets control requirements. The unit is m³ / s, and the smaller the value, the better. L1 norm (i.e., the sum of the absolute values ​​of all gate opening changes) controls the extent of the new scheme's modification of the old scheme. If α is large, the algorithm tends to maintain the original scheme. Its engineering significance is to avoid mechanical shock and increased energy consumption caused by frequent gate opening and closing. This is used to punish excessive opening of the edge gates, thereby ensuring stable discharge and reducing imbalance between the left and right banks; Used to control the uniformity or non-uniformity of door opening, when Encourage unevenness (large differences, unequal opening), when Encourage equal opening (open all gates as similarly as possible), used for different scheduling strategies, such as uniform opening of discharge or unequal opening of discharge in different zones; Used to measure the difference in opening degree between mirrored doors on the left and right banks; It is the set of all mirror gate pairs, if The algorithm penalizes asymmetry (promoting symmetrical opening), thereby maintaining symmetrical discharge, reducing lateral displacement of water flow, and ensuring structural safety.

[0086] Physical constraints:

[0087]

[0088] in, The set of gates that are allowed to be opened (as determined by regulations / operating conditions). For the first Minimum opening template value (m) for gate number 1. , For a set of partition gates (e.g., left bank / right bank). For the first The opening degree of the gate, This represents the minimum opening degree that odd-numbered gates are typically opened to. This is the minimum opening degree that the gate in Zone 2 is typically opened to.

[0089] The comprehensive score, under the premise of meeting the physical hard constraints (only within the allowed gate group, no opening degree below the template, zone 1 triggering zone 2 ≥ 6m, even number triggering odd number ≥ 4m), minimizes the target flow deviation + minimal modification penalty + weighting items such as center / symmetric / (reverse) equal opening preference; each weight, threshold and zone / gate group rule are configurable parameters and can be flexibly adjusted according to different engineering conditions.

[0090] (B) Employing the BeamSearch approach: Simultaneously retaining several optimal candidates, expanding the search within the neighborhood, and gradually approaching the optimal solution.

[0091] 1. Initial seed generation: from the lowest template and the last dispatch Generate; Center priority: Prioritize increasing the center gate position of each partition; Odd-even alternation: Odd numbers open first, even numbers follow; Quantize and filter illegal solutions after generation.

[0092] 2. Neighborhood expansion: Generate a neighborhood for each candidate vector: single-gate ±0.5m adjustment, double-gate linkage ±0.5m, mirror pair linkage ±0.5m, large-step adjustment ±1.0m, and center three-linkage (center and left and right neighbors are adjusted together).

[0093] 3. Batch Prediction and Scoring: Use the batch interface to predict multiple candidates at once.

[0094]

[0095] Calculate the score Keep the first K records in ascending order. .

[0096] 4. Iteration and Early Stopping: Repeat neighborhood expansion and scoring until:

[0097]

[0098] Or the number of iterations has reached the limit. .

[0099] in, This is the absolute error tolerance value (m³ / s). This is the relative error tolerance coefficient. The predicted discharge flow rate is... The flow rate is released to the target.

[0100] (C) Output candidate set:

[0101]

[0102] Each entry contains: Opening vector Predicted flow ,error ,score ), regulations and procedures and minimum template The scores are then sorted from smallest to largest, and the top 5 are selected as the optimal scheduling scheme.

[0103] (4) Set an equal opening sequence in each gate candidate scheduling scheme in the gate candidate scheduling scheme set, and perform flow prediction to obtain the maximum flow corresponding to each gate candidate scheduling scheme; when the maximum flow corresponding to the gate candidate scheduling scheme is less than the target flow, and the difference between the predicted flow and the target flow is not greater than the preset tolerance band, the gate candidate scheduling scheme is determined as the target scheduling scheme.

[0104] In one implementation, to prevent the target traffic from exceeding the model's reachability, the system constructs an equally open sequence within each set of procedures:

[0105]

[0106] in, The specific values ​​can be customized according to different projects. The corresponding flow rate is predicted to obtain the maximum achievable flow rate. .like The target traffic limit has been exceeded.

[0107] Final result filtering:

[0108]

[0109] in, or .

[0110] If no satisfactory solution is found, the process reverts to the Top 5 solutions with the smallest error among all candidates.

[0111] Therefore, the above-mentioned solution of the present invention automatically models complex hydraulic relationships through machine learning, eliminating the need for traditional empirical formulas; it ensures the feasibility of the results through explicit quantification of procedural rules and physical constraints; it achieves smooth optimization of the scheduling scheme through bundle search combined with symmetry, center preference, and minimal modification strategies; and it has high interpretability and real-time performance, capable of outputting multiple schemes within seconds, providing decision support for scheduling personnel. Its complete calculation process can be summarized as follows:

[0112]

[0113] Ultimately, the optimal gate opening combination that meets multiple constraints and is closest to the target flow rate is obtained, realizing intelligent, efficient, and safe scheduling of hydraulic gate groups.

[0114] In practical applications, if the average annual flow at the dam site is about 1250 m³ / s and the design flood flow is about 8500 m³ / s, the normal water level range upstream of the dam is 36 to 46 m, and the water level fluctuation range downstream is 32 to 42 m, the spillway structure is equipped with 18 arc-shaped gates, each with a net width of 8 m and a height of 16 m, which are hydraulically controlled and can be operated independently or in groups.

[0115] Under the existing manual dispatching model, dispatchers mainly determine the gate combination and opening degree based on historical experience, forecasted water inflow, and operating curves. However, in the event of sudden floods or rapid adjustment, manual judgment often results in problems such as delayed response, uneven gate opening, and large deviation in discharge flow, leading to a decrease in dispatching safety margin.

[0116] To address the aforementioned challenges, the method proposed in this invention is integrated into the SCADA system of the dam, enabling intelligent prediction and multi-objective optimization control of gate opening. The system acquires the water level upstream of the dam through a real-time data acquisition module. Downstream water level Hole opening Generating flow rate of the unit and measured discharge flow The sampling period is 1 hour, and the data has high continuity and traceability.

[0117] The sample data covers the operational period from 2020 to 2024, with a cumulative collection of approximately 36,000 valid samples. To improve the model's stability and generalizability under extreme water level conditions, approximately 1,000 additional samples from high flood and low water level periods were added using a two-dimensional hydrodynamic simulation system. These samples, after noise smoothing and normalization, were then incorporated into the model training set. The final sample space is... , The uniform distribution within the range provides ample support for machine learning prediction and optimization.

[0118] When constructing and training the flow prediction model, the first step is feature design: to accurately describe the nonlinear discharge characteristics of the 18-gate sluice, the model input features not only include traditional physical variables but also introduce various higher-order interaction terms. The feature vector is defined as follows:

[0119] in, The difference in water levels between upstream and downstream; Total opening; Squared opening term, used to enhance nonlinear fitting; : Difference between adjacent holes, characterizing the turbulence effect caused by uneven opening; : The coupling term between hydrodynamic difference and flow area; Temporal characteristics, including seasonal, hourly components, and the rate of change of neighboring flow.

[0120] Next, model training was performed: the LightGBM algorithm based on gradient boosting trees was used as the traffic prediction model. The ratio of training set, validation set, and test set was 7:2:1. The main parameters are as follows: num_leaves (number of leaf nodes) is 64; max_depth (maximum tree depth) is 8; learning_rate (learning rate) is 0.05; feature_fraction (feature sampling rate) is 0.8; bagging_fraction (sample sampling rate) is 0.9; and lambda_l1 / lambda_l2 (regularization coefficients) are 0.1 / 0.2.

[0121] To prevent overfitting, 5-fold cross-validation and an early stopping mechanism (stopping_rounds=400) were used. After training, the model achieved the following metrics on the test set:

[0122]

[0123] The prediction error was kept within ±2%.

[0124] In addition, to adapt to changes in operating conditions across different seasons, a historical deviation correction term is introduced:

[0125]

[0126] in, and These are the average relative errors of the previous day and the previous week, respectively. This feature can dynamically correct model bias and maintain the time-varying consistency of prediction results.

[0127] Furthermore, multi-constraint search optimization scheduling is performed, including:

[0128] 1. Optimization Objectives and Constraints: Based on the predicted flow model, this invention constructs a multi-objective constrained optimization model for the gate group. Constraints include: Single-gate opening constraint: Constraint on the difference in opening between adjacent holes: Symmetry constraints: Total flow constraint: Adjustment rate constraint: The objective function is defined as:

[0129]

[0130] in, : Predicted flow rate error; Opening dispersion; Symmetry penalty term; Central preference term; typical weight values .

[0131] 2. Beam Search Implementation: Algorithm sets search width Search depth Using the predictive model as the evaluator, candidate gate opening combinations are generated in each layer of expansion. A comprehensive objective function is calculated for each combination. Keep the lowest scorer Each node serves as a search node for the next layer. By pruning and backtracking layer by layer, a heuristic search tree of finite width is formed, ultimately outputting the optimal gate opening scheme that satisfies multiple constraints. The algorithm takes an average of about 0.8 seconds per iteration and can run in real time within the 5-second update cycle of the SCADA system.

[0132] After the above model and optimization algorithm were deployed, field testing was completed during the main flood season in 2024. The test results showed that: 1. Flow prediction accuracy was significantly improved: compared with manual experience calculation, the prediction RMSE decreased by 47%, and the average error was controlled within ±1.9%; 2. Gate operation coordination was improved: the deviation of adjacent gate openings decreased from 0.72m in manual scheduling to 0.28m, and the discharge distribution was more uniform; 3. Response efficiency was improved: the automatic scheduling response time during sudden rises in water was shortened from 15 minutes to 3 minutes; 4. Safety and energy consumption were optimized: the energy loss of the overflow gate group was reduced by about 6%, and the downstream return water conditions of the unit were more stable. After six consecutive months of verification by the operation department, the system operated stably without any misadjustment or lag. The model prediction results were highly consistent with the measured flow, providing real-time decision support for scheduling.

[0133] The above embodiments fully verify the feasibility and superiority of the present invention in real engineering. By introducing a customized flow prediction model based on LightGBM and a multi-constraint search optimization algorithm, the system realizes adaptive, real-time, and coordinated control of the 18-gate group, which not only ensures flood control safety but also significantly improves operating efficiency and scheduling intelligence.

[0134] In summary, the present invention has the following beneficial effects:

[0135] 1. Achieves the unity of standardization and intelligence: This invention transforms the opening and closing sequence, flow segmentation, odd and even gate restrictions, and zoning symmetry requirements in traditional hydraulic engineering operation procedures into computable rule templates and constraints. This enables the scheduling calculation to automatically generate gate combination schemes while strictly adhering to the procedures, ensuring that each result output by the algorithm has engineering legality and execution feasibility.

[0136] 2. Breaking through the dependence on human experience, realizing the solidification and parameterization of scheduling knowledge: Through the quantitative abstraction of years of operational experience, the system solidifies the human start-up and closing experience (such as center priority, two-sided symmetry, even touch, odd start, etc.) into model constraints and scoring items, making the traditional scheduling logic that relies on human judgment reproducible, transferable and scalable, significantly reducing the dependence on human operation experience.

[0137] 3. A high-precision flow prediction model based on machine learning was constructed: The LightGBM model was used to establish a nonlinear mapping relationship between the upstream water level, downstream water level, gate opening and discharge flow. Compared with the traditional empirical formula model, the prediction accuracy is significantly improved (R² can reach more than 0.98, and the average relative error is less than 3%), providing an accurate and reliable hydraulic calculation basis for subsequent scheduling optimization.

[0138] 4. A multi-constraint search optimization algorithm is proposed to automatically solve for the optimal opening degree: This invention adopts a beam search mechanism to perform a heuristic search in a high-dimensional discrete space with multiple gates and constraints. This algorithm can simultaneously retain several optimal candidate solutions and, combined with multi-objective weighted strategies such as symmetry penalty, center preference, variable constraints, and uniformity adjustment, quickly converges to the optimal or near-optimal opening degree combination, significantly improving computational efficiency and result stability.

[0139] 5. Meets multi-objective scheduling requirements and achieves precise flow control: Automatically generates opening schemes with errors within the tolerable range under given target flow conditions. Actual tests show that the deviation between the predicted flow and the target flow of the generated scheme is usually less than 2%, effectively ensuring the accuracy of flow discharge while maintaining structural safety and hydraulic balance.

[0140] Regarding the intelligent scheduling method for multi-hole gates provided in the foregoing embodiments, this invention provides an intelligent scheduling device for multi-hole gates, see [link to relevant documentation]. Figure 3 The diagram shows a structural schematic of an intelligent scheduling device for a multi-hole gate, which includes the following parts:

[0141] The model update module 302 uses historical operation data and simulation data of the multi-hole gate to determine the target input features, and uses a lightweight gradient boosting tree model to perform nonlinear regression training on the target input features to determine the target flow prediction model at the current moment.

[0142] The constraint search module 304 performs hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme. It also uses a target flow prediction model to perform batch prediction and scoring processing on the gate expansion scheduling scheme to obtain a set of gate candidate scheduling schemes.

[0143] The scheme screening module 306 performs upper limit detection screening and error tolerance screening on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate.

[0144] The intelligent scheduling device for multi-hole gates provided in this application embodiment can significantly improve the prediction accuracy of gate water discharge and the scheduling efficiency of multi-hole gates.

[0145] In one embodiment, when determining the target input features using historical operation data and simulation data of the multi-hole gate, the model update module 302 is further configured to: construct a sample database based on the historical operation data and simulation data; perform feature combination processing and nonlinear transformation processing on each input feature in the sample database to perform physical prior extension of each input feature, obtain the corresponding higher-order features, and combine the higher-order features to generate the target input features.

[0146] In one implementation, when performing the step of nonlinear regression training on the target input features using a lightweight gradient boosting tree model to determine the target traffic prediction model at the current moment, the model update module 302 is further configured to: use the lightweight gradient boosting tree model, based on five-fold cross-validation and a preset early stopping mechanism, to screen a preset set of hyperparameter candidates to determine the target hyperparameters, wherein the target hyperparameters are the parameter combinations with the smallest error during five-fold cross-validation; and based on the target hyperparameters, perform nonlinear regression training on the target input features to obtain the target traffic prediction model.

[0147] In one embodiment, after obtaining the target traffic prediction model, the model update module 302 is further configured to: use the prediction error adaptive correction model to correct the predicted traffic output by the target traffic prediction model in combination with historical prediction errors, so as to perform real-time dynamic adjustment of the predicted traffic of the target traffic prediction model.

[0148] In one embodiment, before performing the step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme, the constraint search module 304 is further used to: perform retrieval processing in a preset gate rule knowledge base for the predicted flow output by the target flow prediction model, and determine the gate opening combination that can be executed under the current working condition and the minimum opening template vector.

[0149] In one embodiment, when performing the step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme, the constraint search module 304 is further configured to: perform hierarchical expansion processing on the gate opening combination under the constraints of the gate opening combination and the minimum opening template vector through the multi-constraint search model, generate gate opening combinations corresponding to each layer of expansion, and determine each gate opening combination as a gate expansion scheduling scheme.

[0150] In one embodiment, when performing upper limit detection screening and error tolerance screening on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate, the above-mentioned scheme screening module 306 is further configured to: set an equal opening sequence in each candidate gate scheduling scheme of the set of candidate gate scheduling schemes, and perform flow prediction to obtain the maximum flow corresponding to each candidate gate scheduling scheme; when the maximum flow corresponding to the candidate gate scheduling scheme is less than the target flow, and the difference between the predicted flow and the target flow is not greater than the preset tolerance band, the candidate gate scheduling scheme is determined as the target scheduling scheme.

[0151] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0152] This invention provides a server, specifically, the server includes a processor and a storage device; the storage device stores a computer program, which, when run by the processor, executes the method described in any of the above embodiments.

[0153] Figure 4 This is a schematic diagram of the structure of a server provided in an embodiment of the present invention. The server 100 includes: a processor 40, a memory 41, a bus 42 and a communication interface 43. The processor 40, the communication interface 43 and the memory 41 are connected through the bus 42. The processor 40 is used to execute executable modules, such as computer programs, stored in the memory 41.

[0154] The memory 41 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 43 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0155] Bus 42 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0156] The memory 41 is used to store programs. After receiving an execution instruction, the processor 40 executes the program. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0157] Processor 40 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 40 or by instructions in software form. Processor 40 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 41. The processor 40 reads the information in memory 41 and, in conjunction with its hardware, completes the steps of the above method.

[0158] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0159] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0160] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent scheduling method for multi-hole gates, characterized in that, The method includes: Using historical operation data and simulation data of multi-hole gates, target input features are determined, and nonlinear regression training is performed on the target input features through a lightweight gradient boosting tree model to determine the target flow prediction model at the current moment. By using a multi-constraint search model, the gate opening combination is subjected to hierarchical expansion processing to obtain a gate expansion scheduling scheme. Then, the target flow prediction model is used to perform batch prediction and scoring processing on the gate expansion scheduling scheme to obtain a set of gate candidate scheduling schemes. The upper limit detection screening process and error tolerance screening process are performed on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate; Before the step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain the gate expansion scheduling scheme, the method includes: performing a retrieval process in a preset gate rule knowledge base for the predicted flow output by the target flow prediction model to determine the gate opening combination that can be executed under the current working condition and the minimum opening template vector. The step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme includes: performing hierarchical expansion processing on the gate opening combination under the constraints of the gate opening combination and the minimum opening template vector through a multi-constraint search model, generating gate opening combinations corresponding to each layer of expansion, and determining each of the gate opening combinations as the gate expansion scheduling scheme.

2. The intelligent scheduling method for multi-hole gates according to claim 1, characterized in that, The step of determining the target input features using historical operating data and simulation data of the multi-hole gate includes: A sample database is constructed based on historical operational data and simulation data; By performing feature combination processing and nonlinear transformation processing on each input feature in the sample database, physical prior extension of each input feature is performed to obtain corresponding higher-order features, and the higher-order features are combined to generate the target input feature.

3. The intelligent scheduling method for multi-hole gates according to claim 1, characterized in that, The step of determining the target traffic prediction model at the current time by performing nonlinear regression training on the target input features using a lightweight gradient boosting tree model includes: Using a lightweight gradient boosting tree model, based on five-fold cross-validation and a preset early stopping mechanism, a preset set of candidate hyperparameters is screened to determine the target hyperparameter, wherein the target hyperparameter is the parameter combination with the smallest error during five-fold cross-validation. Based on the target hyperparameters, the target input features are subjected to nonlinear regression training to obtain the target traffic prediction model.

4. The intelligent scheduling method for multi-hole gates according to claim 3, characterized in that, After the step of obtaining the target traffic prediction model, the following is included: By using an adaptive correction model for prediction errors, the predicted flow output by the target flow prediction model is combined with historical prediction errors for correction, so as to perform real-time dynamic adjustment of the predicted flow of the target flow prediction model.

5. The intelligent scheduling method for multi-hole gates according to claim 1, characterized in that, The steps of performing upper limit detection screening and error tolerance screening on the candidate gate scheduling scheme set to determine the target scheduling scheme for the multi-gate include: An equal-opening sequence is set in each of the candidate gate scheduling schemes in the set of candidate gate scheduling schemes, and flow prediction is performed to obtain the maximum flow corresponding to each of the candidate gate scheduling schemes; When the maximum flow rate corresponding to the candidate gate scheduling scheme is less than the target flow rate, and the difference between the predicted flow rate and the target flow rate is not greater than the preset tolerance band, the candidate gate scheduling scheme is determined as the target scheduling scheme.

6. An intelligent scheduling device for a multi-hole gate, characterized in that, The device includes: The model update module uses historical operation data and simulation data of the multi-hole gate to determine the target input features, and uses a lightweight gradient boosting tree model to perform nonlinear regression training on the target input features to determine the target flow prediction model at the current moment. The constraint search module performs hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme. Then, it uses the target flow prediction model to perform batch prediction and scoring processing on the gate expansion scheduling scheme to obtain a set of gate candidate scheduling schemes. The scheme selection module performs upper limit detection screening and error tolerance screening on the set of candidate gate scheduling schemes to determine the target scheduling scheme for the multi-hole gate. Before the step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain the gate expansion scheduling scheme, the method includes: performing a retrieval process in a preset gate rule knowledge base for the predicted flow output by the target flow prediction model to determine the gate opening combination that can be executed under the current working condition and the minimum opening template vector. The step of performing hierarchical expansion processing on the gate opening combination through a multi-constraint search model to obtain a gate expansion scheduling scheme includes: performing hierarchical expansion processing on the gate opening combination under the constraints of the gate opening combination and the minimum opening template vector through a multi-constraint search model, generating gate opening combinations corresponding to each layer of expansion, and determining each of the gate opening combinations as the gate expansion scheduling scheme.

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

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

Images