Tidal estuary gate opening and closing method based on multi-modal perception and dynamic optimization decision

Abstract

The invention discloses a tidal estuary gate opening and closing method based on multi-modal perception and dynamic optimization decision, which comprises the following steps: collecting multi-source environment and forecast data, and judging whether to trigger scheduling or not based on forecast uncertainty; a pre-constructed physical information agent model library is called, safety evaluation is carried out on the multiple candidate opening and closing scenes, and the agent model library is obtained based on offline three-dimensional fluid-solid coupling simulation training; in the situation that structural safety constraints are met, according to the current flood control, energy saving, navigation and salt tide suppression requirements, an optimal opening and closing scheme is generated through dynamic weight multi-objective optimization; the optimal opening and closing scheme is executed, the water level difference deviation is monitored in real time in the execution process, and if the water level difference deviation exceeds the limit, a safe backspacing mechanism is started; and continuously updating the proxy model library based on the operation feedback data. According to the method, safe, efficient and adaptive cooperative regulation and control of the high-risk coastal estuary gate under multi-disaster coupling are realized, and the structural safety, decision robustness and comprehensive operation benefits are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and information technology in water conservancy projects, and in particular to a method and design method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making. Background Technology

[0002] Against the backdrop of global climate change, extreme weather events are becoming more frequent, and coastal estuaries face the risk of multiple disasters occurring simultaneously, including storm surges, regional floods, spring tides, and heavy rainfall—a significant increase in the risk of "triple or quadruple" disasters. Traditional tidal estuary gate scheduling relies mainly on manual experience and fixed rules, resulting in problems such as delayed response, lack of global optimization, and difficulty in simultaneously addressing multiple objectives such as flood control, navigation, and energy conservation.

[0003] In existing technologies, some studies attempt to introduce hydrodynamic models for scheduling simulation, but these generally embed high-cost 3D CFD simulations directly into the online decision-making process, resulting in high computational latency and inability to meet real-time requirements. Other solutions use purely data-driven models but ignore physical constraints, resulting in poor generalization ability and a tendency to make dangerous decisions under unseen conditions. Furthermore, existing methods often ignore the uncertainties of meteorological and hydrological forecasts, treating predicted values ​​as deterministic inputs, which can easily lead to misjudgments. They also lack real-time monitoring and emergency response mechanisms for sudden deviations during execution, such as equipment failures and sudden changes in water levels.

[0004] Therefore, there is an urgent need for an intelligent opening and closing method that combines physical interpretability, real-time response capability, uncertainty robustness, and multi-objective adaptive balancing capability to improve the overall scheduling efficiency and operational safety of tidal estuary gates in complex scenarios.

[0005] Chinese patent document CN119863016A discloses a multi-objective intelligent hydraulic control method for sluice gate groups in tidal river networks. This document focuses on the joint scheduling problem of sluice gate groups within tidal river networks, with the main tasks being to improve the water flow of the river network, enhance its self-purification capacity, and meet multiple objectives such as flood control, landscape, and water environment. However, its shortcomings are: First, the document only considers whether the water level is within the landscape / flood control range, failing to consider the fatal impact of the water level difference on the two sides of the gate on structural stability. For estuary gates directly facing the ocean, an excessive water level difference can directly lead to gate overturning. Second, it treats tidal forecasts as fixed values, which can easily lead to misscheduling due to forecast deviations. Third, it does not include saltwater intrusion suppression and navigation assurance, making it unsuitable for coastal estuary scenarios, and its fixed weights cannot dynamically adapt to changes such as typhoons and peak shipping seasons. Summary of the Invention

[0006] The technical problem this invention aims to solve is to provide a method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making. This method addresses the complex operating conditions of coastal tidal estuary gates facing multiple coupled disasters such as storm surges, upstream floods, spring tides, and heavy rainfall. It balances multiple objectives, including gate structural safety, flood control and drainage, ship navigation, saltwater intrusion suppression, and energy conservation. It overcomes the shortcomings of existing scheduling methods, such as reliance on deterministic forecasts, lack of physical constraints, inability to respond in real time, and neglect of safety monitoring during execution. This invention provides an intelligent opening and closing method with uncertainty perception capabilities, physical interpretability, dynamic multi-objective optimization, and a closed-loop safety backoff mechanism, enabling safe, efficient, and adaptive collaborative control of tidal estuary gates in high-risk environments.

[0007] To achieve the above objectives, this application provides a method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making, characterized by the following steps: S1. Collect multi-source environmental and forecast data, and determine whether to trigger scheduling based on forecast uncertainty; S2. Call the pre-built physical information proxy model library to perform security assessments on multiple candidate opening and closing scenarios. The proxy model library is obtained based on offline three-dimensional fluid-structure interaction simulation training. S3. In feasible scenarios that meet structural safety constraints, the optimal opening and closing scheme is generated through dynamic weighted multi-objective optimization based on current requirements for flood control, energy conservation, navigation and saltwater intrusion suppression. S4. Execute the optimal opening and closing scheme, and monitor the water level difference deviation in real time during the execution process. If the deviation exceeds the limit, activate the safety backoff mechanism. S5. Continuously update the agent model library based on the operation feedback data.

[0008] In S1, the step of determining whether to trigger scheduling based on forecast uncertainty includes: acquiring meteorological and hydrological ensemble forecast data and calculating whether the highest tide level of any future tide exceeds the activation threshold. probability If the probability is not lower than the preset confidence level, then the scheduling process is activated; ; In the formula: N is the number of members in the ensemble forecast; The tide level predicted by the kth member; For indicator functions; if Then the scheduling process will be activated. To pre-set the reliability threshold.

[0009] In S2, the physical information proxy model library is constructed in the following way: a three-dimensional fluid-structure interaction mechanism model is established using the Navier-Stokes equations and the GMO module, and offline high-fidelity simulations are performed on various opening and closing scenarios; a physical information neural network or Gaussian process regression model is trained based on the simulation data, and the input features include hydraulic parameters.

[0010] The hydraulic parameters include the Froude number, Reynolds number, and relative starting water level.

[0011] In S3, the structural safety constraints include: the maximum forward water level difference does not exceed a first threshold, and the maximum reverse water level difference does not exceed a second threshold.

[0012] In S3, if there are candidate opening and closing scenarios that satisfy the structural safety constraints, the optimal opening and closing scheme is generated from them based on dynamic weight multi-objective optimization; if the structural safety constraints are not satisfied, the opening and closing of the gate is postponed.

[0013] In S3, the weight coefficients in the dynamic weighted multi-objective optimization are automatically adjusted according to the real-time operating conditions, and the sum of the flood control risk weight, energy consumption intensity weight, navigation impact weight and saltwater intrusion suppression weight is 1.

[0014] The weight coefficients are dynamically output by the reinforcement learning agent based on feedback from historical scheduling results.

[0015] In S4, the safety rollback mechanism includes: issuing an alarm signal and initiating a reassessment process.

[0016] In S5, the continuous update includes short-term parameter compensation and long-term model fine-tuning: when the prediction deviation exceeds the limit continuously, the offset correction is applied to the starting water level setting; when the accumulated new samples reach the threshold, transfer learning is triggered to retrain the surrogate model.

[0017] Compared with the prior art, the above-conceptual technical solution conceived in this application has the following beneficial effects: This application fundamentally ensures the structural safety of the sluice gates under extreme conditions such as storm surges by introducing ensemble forecast uncertainty calculations and strict water level difference safety constraints, and by setting up emergency gate closure plans. It relies on a physical information proxy model built using offline simulation to balance prediction speed and physical interpretability, avoiding the non-physical interpretation risks of purely data-driven models. It incorporates saltwater intrusion suppression and navigation assurance into four-dimensional dynamic optimization objectives, and achieves multi-objective adaptive balance through automatic weight adjustment using reinforcement learning. Real-time deviation monitoring and safety rollback mechanisms during execution, along with a feedback-based model continuous update strategy, ensure the system possesses closed-loop security and long-term evolutionary capabilities. The overall solution significantly improves scheduling response speed, decision robustness, and comprehensive operational efficiency while ensuring safety, providing a feasible and reliable intelligent solution for high-risk coastal estuaries. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0019] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0020] To more clearly illustrate the purpose, technical solution, and beneficial effects of this application, a further detailed description of this application is provided below in conjunction with illustrations and specific embodiments. It should be specifically noted that the specific embodiments described below are only for illustrating the technical content of this application and do not constitute a limitation on the scope of protection of this application.

[0021] Regarding the explanation of terminology: In this application, "and / or" is used to describe the relationship between related objects, covering three possible situations: taking "A and / or B" as an example, it can indicate the situation where only A exists, A and B exist simultaneously, or only B exists; the symbol " / " indicates the "or" relationship between related objects, such as "A / B" which refers to A or B.

[0022] Regarding the description of the embodiments: The terms "exemplary" and "for example" appearing in this application are only used to illustrate the technical solutions through specific examples. It should be particularly emphasized that any implementation method or design scheme marked as "exemplary" or "for example" should not be construed as having an advantage over other solutions. Such expressions are only used to present the technical concepts more intuitively.

[0023] Example 1: See Figure 1 This embodiment provides a method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making, characterized by the following steps: S1. Collect multi-source environmental and forecast data, and determine whether to trigger scheduling based on forecast uncertainty.

[0024] The system connects to the ECMWF ensemble forecast system to obtain the tidal level forecast results of 51 members for the next 72 hours.

[0025] Specifically, determining whether to trigger scheduling based on forecast uncertainty includes: acquiring ensemble meteorological and hydrological forecast data and calculating whether the highest tide level for any future tide exceeds the activation threshold. probability If the probability is not lower than the preset confidence level, then the scheduling process is activated; ; In the formula: N is the number of members in the ensemble forecast; The tide level predicted by the kth member; For indicator functions; if Then the scheduling process will be activated. To pre-set the reliability threshold.

[0026] Confidence threshold The threshold can be dynamically adjusted according to the season: lowered to 75% during typhoon season and raised to 85% during dry season.

[0027] By adopting the above scheme, we can avoid using a single deterministic forecast as the basis for decision-making and significantly reduce the false trigger rate; the dynamic confidence mechanism improves the system's adaptability to different climate backgrounds.

[0028] S2. Call the pre-built physical information proxy model library to perform security assessments on multiple candidate opening and closing scenarios. The proxy model library is obtained based on offline three-dimensional fluid-structure interaction simulation training.

[0029] In S2, the physical information proxy model library is constructed in the following way: a three-dimensional fluid-structure interaction mechanism model is established using the Navier-Stokes equations and the GMO module to perform offline high-fidelity simulations of various opening and closing scenarios; a physical information neural network or Gaussian process regression model is trained based on the simulation data, with hydraulic parameters included as input features.

[0030] Specifically, the construction method of the physical information agent model library is as follows: Step 2.1: Establish a three-dimensional simulation model of the fluid-structure interaction mechanism; A three-dimensional numerical model of the tidal estuary lock area is constructed using computational fluid dynamics (CFD) software such as Flow-3D, OpenFOAM, or ANSYS Fluent. The governing equations use the Navier-Stokes equations to describe the water flow: ; ; In the formula: This refers to the fluid volume fraction. The velocity of the water flow; It is the viscous stress tensor; The momentum source term is introduced by the gate motion; It is the velocity tensor product; The pressure gradient drives the fluid to flow from high pressure to low pressure; It is a gravitational volume force; Mass source terms caused by molecular diffusion or numerical diffusion; This is a virtual mass source introduced by the GMO module, used to simulate the disturbance of the flow field by the gate motion.

[0031] The fluid-structure interaction during the gate rotation and sinking process was simulated using the GMO (General Moving Object) module or the immersion boundary method (IBM); the mesh resolution was controlled at 0.3–0.5 m and the time step was ≤0.1 s to ensure the capture of transient water level differences and structural stress. Step 2.2: Design a multi-dimensional set of start-stop scenarios and perform offline simulation; Based on the actual needs of the project, the system design covers a combination of opening and closing scenarios for typical and extreme working conditions, including: opening water level (1.0–3.0 m, step size 0.1 m), rotation time (30–60 min), sinking time (15–30 min), upstream flow rate (50–800 m³ / s), and tidal type (real-time tide, once-in-a-millennium storm surge, and typhoon composite scenario). A total of several independent simulation cases are generated, for example, 1024 cases, with each simulation outputting five key response variables: Maximum positive water level difference The absolute value of the maximum reverse water level difference Total energy consumption for opening and closing (E), and available window duration for navigation. and the cumulative time of salinity exceeding the standard. This forms a structured training dataset; Step 2.3: Construct and train the physical information agent model; Based on the above simulation data, a Physical Information Neural Network (PINN) model was trained; The input feature vector contains the original parameters and hydraulic parameters: ; In the formula: The starting water level is the water level in front of the gate when the opening and closing action begins. Rotation time is the time required for the gate to go from fully open to fully closed or vice versa. The sinking time is the time it takes for the gate to sink before it locks at the bottom. The upstream water flow rate; Froude number Reynolds number ; This is relative to the water level at the gate; This represents the daily tidal range, reflecting the intensity of the tidal background. The output label vector is a multi-target response: ; In the formula: This represents the maximum positive water level difference. This represents the absolute value of the maximum reverse water level difference. Total energy consumption for start-up and shutdown; The duration of the available window for air traffic control; This represents the cumulative time during which salinity exceeds the standard. The loss function is designed as follows: ; In the formula: To force the model output to satisfy the physical conservation law in order to substitute the model's predicted field into the residual norm generated by the Navier-Stokes equations; Weighting coefficients for data fidelity items; The mean square error between the predicted and actual values; This refers to the weighting coefficient for the physical consistency term; Step 2.4: Construct and train the physical information agent model; The trained PINN model undergoes INT8 quantization and operator fusion. It is then packaged into a Docker container and deployed on edge computing nodes, such as the Huawei Atlas 500 Pro, supporting real-time reception of activation / deactivation parameters and return of security assessment results via the gRPC interface. Simultaneously, the original CFD simulation interface is retained for backtesting verification in extreme events.

[0032] The opening and closing scenarios are generated by parameter space sampling, covering feasible combinations of variables such as opening and closing water level, rotation time, sinking time, upstream flow rate, and tidal pattern. The physical information proxy model library constructed through the above steps combines high accuracy, strong generalization, fast response, and physical consistency, providing reliable support for subsequent online safety assessment and optimization decisions, and significantly outperforming single methods such as pure data-driven or pure mechanism simulation.

[0033] S3. In feasible scenarios that meet structural safety constraints, the optimal opening and closing scheme is generated through dynamic weighted multi-objective optimization based on current requirements for flood control, energy conservation, navigation, and saltwater intrusion suppression.

[0034] Structural safety constraints include: the maximum forward water level difference does not exceed the first threshold, and the maximum reverse water level difference does not exceed the second threshold.

[0035] If there are candidate opening and closing scenarios that satisfy the structural safety constraints, the optimal opening and closing scheme is generated from them based on dynamic weight multi-objective optimization; if the structural safety constraints are not satisfied, the opening and closing of the gate is postponed.

[0036] In the dynamic weighted multi-objective optimization, the weight coefficients are automatically adjusted according to real-time operating conditions, and the sum of the flood control risk weight, energy consumption intensity weight, navigation impact weight, and saltwater intrusion suppression weight is 1.

[0037] The weight coefficients are dynamically output by the reinforcement learning agent based on feedback from historical scheduling results.

[0038] Specifically, structural safety constraints include: maximum positive water level difference (before the gate > after the gate): Maximum reverse water level difference (behind the gate > in front of the gate): ; If there are candidate opening and closing scenarios that satisfy structural safety constraints, the optimal opening and closing scheme is generated from them based on dynamic weighted multi-objective optimization; if the structural safety constraints are not satisfied, the opening and closing of the gate is temporarily suspended and the emergency command center is notified.

[0039] The objective function is: ; In the formula: the four weights satisfy .

[0040] The weighting rules are as follows: Typhoon warning period: α ≥ 0.6; Peak shipping season (08:00–18:00): γ ≥ 0.5; Dry season (October to March of the following year): δ ≥ 0.4; Other periods: β ≥ 0.6.

[0041] The weight coefficients are dynamically output by the reinforcement learning agent based on historical scheduling feedback, and their reward function is defined as: ; In the formula: ; The number of flooding incidents; This represents actual energy consumption. Total waiting time for the vessel; This represents the cumulative duration of salinity exceeding the standard.

[0042] Traditional scheduling methods typically focus on a single objective, such as flood control or water exchange, or use fixed weights to handle multiple objectives, making them ill-suited to the complex and ever-changing estuarine environment. This application integrates four core requirements—flood control (controlling the risk of waterlogging), energy conservation (reducing motor energy consumption), navigation (ensuring ship passage), and saltwater intrusion suppression (preventing saltwater backflow from affecting water supply)—into a unified optimization framework. A dynamic weighting mechanism adjusts the priority of each objective in real time; for example, it automatically increases the flood control weight during typhoon warning periods, enhances navigation guarantees during peak shipping seasons, and strengthens saltwater intrusion control during the dry season. This adaptively generates a globally optimal opening and closing strategy under different operating conditions, significantly improving the overall operational efficiency of the estuarine gate.

[0043] The optimization search space is clearly limited to feasible scenarios that meet structural safety constraints, namely, the maximum forward water level difference does not exceed 0.5 m and the maximum reverse water level difference does not exceed 1.5 m. This fundamentally eliminates the risk of sacrificing the structural safety of the gate in pursuit of a certain performance, such as rapid drainage, and achieves an organic unity between the inviolability of the safety baseline and flexible optimization of multiple objectives.

[0044] By introducing reinforcement learning agents, the system can automatically adjust weight coefficients based on historical scheduling results, such as actual flooding frequency, energy consumption deviation, and ship delay duration, enabling the optimization strategy to continuously approach the optimal solution in real-world scenarios. This closed-loop mechanism gives the scheduling system continuous learning and environmental adaptability, far superior to traditional methods that rely on manual rules or static models.

[0045] S4. Execute the optimal opening and closing scheme and monitor the water level difference deviation in real time during the execution process. If the deviation exceeds the limit, activate the safety backoff mechanism.

[0046] The security rollback mechanism includes: issuing an alarm signal and initiating a reassessment process.

[0047] Specifically, the measured water level difference is collected every 10 seconds during the opening and closing process. Compared with the predicted value If the condition is met twice consecutively: If this occurs, a safety rollback is triggered: an alarm signal is issued and a reassessment process is initiated.

[0048] This application collects the measured water level difference and the model prediction value at a high frequency during the opening and closing of the gate. Once the deviation is found to exceed the limit continuously, the safety backoff mechanism is immediately triggered: pausing the current action, locking the current position of the gate, issuing an alarm and re-evaluating. This effectively responds to serious deviations between prediction and reality caused by sudden upstream flood discharge, sudden changes in storm surge intensity, or sensor failure, and prevents the gate from continuing to operate under unknown dangerous conditions, fundamentally preventing structural overload, equipment damage, or even safety accidents.

[0049] S5. Continuously update the agent model library based on the operation feedback data.

[0050] In S5, the continuous update includes short-term parameter compensation and long-term model fine-tuning: when the prediction deviation exceeds the limit continuously, the offset correction is applied to the starting water level setting; when the accumulated new samples reach the threshold, transfer learning is triggered to retrain the surrogate model.

[0051] The short-term compensation involves applying an offset to the starting water level setting if the prediction deviations are in the same direction for three consecutive times; the long-term update involves triggering transfer learning to fine-tune the model and pushing updates via OTA when the cumulative number of new samples is ≥50 or the average absolute error is >0.12 m.

[0052] The tidal estuary environment is not static. River channels may change their cross-sectional shape due to dredging or siltation, gate equipment may wear out or become sluggish with age, and climate change can lead to trend changes in the frequency and intensity of storm surges. If a surrogate model is fixed after training, its prediction accuracy will gradually degrade over time. By introducing a two-layer update mechanism of short-term parameter compensation and long-term model fine-tuning, such as dynamically correcting the setpoint for the gate water level when the deviation continuously exceeds the limit, or triggering transfer learning retraining after accumulating a threshold of new samples, the model can continuously absorb real-world data, automatically calibrate parameters, and correct deviations, thereby maintaining high-precision prediction capabilities throughout operating cycles of several months or even years.

[0053] This application employs lightweight update strategies such as transfer learning to fine-tune only the output layer or key parameters while preserving the original model's physical knowledge framework. This avoids the computational overhead of training from scratch and ensures that the model can quickly adapt to new operating conditions.

[0054] S5 feeds real-world data generated during operation, such as actual water level differences, energy consumption, and navigation performance, back into the model training phase, enabling the system not only to make decisions but also to learn from experience. This closed-loop learning mechanism makes scheduling strategies more accurate and better with use, truly achieving a leap from static intelligence to continuously evolving intelligence.

[0055] This invention constructs a safe, intelligent, adaptive, and evolvable intelligent opening and closing system for tidal estuary gates through the aforementioned five steps. Compared with existing technologies, this invention not only solves the challenges of safety, efficiency, and ecological coordination in coastal tidal estuary gates under the coupling of multiple disasters, but also constructs a new intelligent control paradigm that integrates uncertainty perception, physical information modeling, dynamic multi-objective optimization, execution safety closed loop, and continuous model evolution. It possesses outstanding technological advancement, engineering practical value, and broad prospects for promotion.

[0056] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the invention. Modifications and variations made by those skilled in the art in accordance with the spirit of the invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for opening and closing tidal estuary gates based on multimodal sensing and dynamic optimization decision-making, characterized in that, Includes the following steps: S1. Collect multi-source environmental and forecast data, and determine whether to trigger scheduling based on forecast uncertainty; S2. Call the pre-built physical information proxy model library to perform security assessments on multiple candidate opening and closing scenarios. The proxy model library is obtained based on offline three-dimensional fluid-structure interaction simulation training. S3. In feasible scenarios that meet structural safety constraints, the optimal opening and closing scheme is generated through dynamic weighted multi-objective optimization based on current requirements for flood control, energy conservation, navigation and saltwater intrusion suppression. S4. Execute the optimal opening and closing scheme, and monitor the water level difference deviation in real time during the execution process. If the deviation exceeds the limit, activate the safety backoff mechanism. S5. Continuously update the agent model library based on the operation feedback data.

2. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S1, the step of determining whether to trigger scheduling based on forecast uncertainty includes: acquiring meteorological and hydrological ensemble forecast data and calculating whether the highest tide level of any future tide exceeds the activation threshold. probability If the probability is not lower than the preset confidence level, the scheduling process is activated. ； In the formula: N is the number of members in the ensemble forecast; The tide level predicted by the kth member; For indicator functions; if Then the scheduling process will be activated. To pre-set the reliability threshold.

3. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S2, the physical information proxy model library is constructed in the following way: a three-dimensional fluid-structure interaction mechanism model is established using the Navier-Stokes equations and the GMO module, and offline high-fidelity simulations are performed on various opening and closing scenarios; a physical information neural network or Gaussian process regression model is trained based on the simulation data, and the input features include hydraulic parameters.

4. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 3, characterized in that, The hydraulic parameters include the Froude number, Reynolds number, and relative starting water level.

5. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S3, the structural safety constraints include: the maximum forward water level difference does not exceed a first threshold, and the maximum reverse water level difference does not exceed a second threshold.

6. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S3, if there are candidate opening and closing scenarios that satisfy the structural safety constraints, the optimal opening and closing scheme is generated from them based on dynamic weight multi-objective optimization; if the structural safety constraints are not satisfied, the opening and closing of the gate is postponed.

7. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S3, the weight coefficients in the dynamic weighted multi-objective optimization are automatically adjusted according to the real-time operating conditions, and the sum of the flood control risk weight, energy consumption intensity weight, navigation impact weight and saltwater intrusion suppression weight is 1.

8. The tidal estuary gate opening and closing method based on multimodal perception and dynamic optimization decision-making according to claim 7, characterized in that, The weight coefficients are dynamically output by the reinforcement learning agent based on feedback from historical scheduling results.

9. The method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S4, the safety backoff mechanism includes: pausing the current opening and closing action, maintaining the current position of the gate, issuing an alarm signal, and initiating a reassessment process.

10. The method for opening and closing tidal estuary gates based on multimodal perception and dynamic optimization decision-making according to claim 1, characterized in that, In S5, the continuous update includes short-term parameter compensation and long-term model fine-tuning: when the prediction deviation exceeds the limit continuously, the offset correction is applied to the starting water level setting; when the accumulated new samples reach the threshold, transfer learning is triggered to retrain the surrogate model.

Images