AI-based methods and systems for predicting sediment deposition and optimizing dredging

Abstract

This invention relates to the field of intelligent dredging technology, specifically disclosing an AI-based method and system for predicting and optimizing dredging sediment deposition. The method includes: collecting sediment deposition correlation data and raw operational data of multiple brands of dredging equipment from a target water area via a multi-protocol conversion gateway deployed at the edge; the multiple brands of dredging equipment include at least two different brands; and using the multi-protocol conversion gateway to perform protocol parsing and format unification of the raw operational data based on a protocol dictionary, generating standardized data. This invention achieves seamless access and unified management of multiple brands of dredging equipment through multi-protocol adaptive conversion and data standardization; based on a two-level AI decision-making model, it integrates global optimization and local real-time response, significantly improving the overall efficiency and adaptability of dredging operations; and by combining digital twins and closed-loop correction, it ensures the accuracy of command execution and system stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent dredging technology, and in particular to a method and system for predicting and optimizing dredging sediment deposition based on AI calculations. Background Technology

[0002] In traditional dredging projects, the collaborative operation of dredging equipment from multiple brands faces serious data heterogeneity issues. Different manufacturers use proprietary industrial protocols and custom parameter naming rules, resulting in the inability to collect and parse equipment operation data uniformly, making it difficult for the system to achieve global data integration and real-time monitoring.

[0003] Furthermore, when adding equipment from unknown brands, the lack of an adaptive protocol learning mechanism necessitates manual configuration, which is not only inefficient but also prone to introducing errors, thereby affecting the accuracy of sediment deposition prediction and the timeliness of dredging decisions.

[0004] Existing dredging control systems largely rely on preset rules or single optimization algorithms, making it difficult to balance overall efficiency with real-time response to local anomalies. For example, in navigable waters, dredging operations often conflict with vessel traffic, and traditional methods cannot dynamically calculate avoidance strategies, leading to navigation delays or a surge in equipment energy consumption. Simultaneously, in strong hydrodynamic environments, dredging operations easily trigger sediment diffusion and secondary deposition. The lack of effective source control measures results in poor dredging effectiveness, frequent rework, and significantly increased operating costs and environmental risks. Summary of the Invention

[0005] This invention aims to at least partially address one of the technical problems in related technologies. Therefore, the objective of this invention is to propose an AI-based method and system for predicting and optimizing dredging sediment deposition, thereby improving the intelligence level of dredging projects.

[0006] To achieve the above objectives, a first aspect of the present invention proposes an AI-based method for predicting and optimizing dredging sediment deposition, comprising the following steps:

[0007] S1. Data access steps: Collect sediment deposition correlation data of the target water area and raw operating data of dredging equipment from multiple brands through a multi-protocol conversion gateway deployed on the edge side; the multi-brand dredging equipment includes at least two different brands of dredging equipment.

[0008] S2. Data standardization steps: The multi-protocol conversion gateway is used to perform protocol parsing and format unification on the original operating data based on the protocol dictionary library to generate standardized data; wherein, the protocol parsing includes converting private industrial protocols of different brands into a unified standard format based on the pre-stored protocol dictionary library, and the format unification includes mapping similar parameters of different devices to unified parameter names;

[0009] S3. Two-level AI decision-making steps: Input the standardized data and sediment deposition correlation data into the two-level AI scheduling model to generate dredging control commands; wherein, the two-level AI scheduling model includes a global strategy layer and a local response layer. The global strategy layer uses the non-dominated sorting genetic algorithm (NSGA-II) to generate the globally optimal dredging strategy based on a multi-dimensional objective function. The local response layer uses a deep Q-network (DQN) to generate a real-time adjustment strategy for local anomalies based on real-time high-frequency data.

[0010] S4. Virtual-Real Synchronization and Execution Steps: Convert the dredging control commands into exclusive commands for each brand of equipment and issue them for execution. At the same time, drive the equipment twin in the digital twin system to perform simulated actions.

[0011] S5. Closed-loop calibration step: Real-time monitoring of the actual motion parameters of the physical device and the simulated motion parameters of the device twin. When the deviation between the two exceeds the preset threshold, a calibration command is generated to perform closed-loop control on the physical device.

[0012] To achieve the above objectives, a second aspect of the present invention proposes an AI-based sediment deposition prediction and optimized dredging system, comprising:

[0013] The data acquisition and conversion module includes a multi-protocol conversion gateway and a protocol dictionary library, which is used to collect environmental and equipment data and perform protocol parsing and format unification to output standardized data;

[0014] The two-level AI scheduling module includes a global policy unit and a local response unit; the global policy unit is used to run a non-dominated sorting genetic algorithm to generate a global policy, and the local response unit is used to run a deep Q-network to make local anomaly decisions.

[0015] The digital twin monitoring module is used to construct a device twin and compare the deviation between the simulated actions and the actual actions of the physical device in real time.

[0016] The execution control module is used to convert global or local adjustment strategies into specific instructions for the corresponding brand of equipment and issue them.

[0017] A closed-loop feedback module is used to generate a correction instruction when the deviation exceeds a threshold, and to iteratively optimize the parameters of the two-level AI scheduling module based on scheduling performance indicators.

[0018] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described AI-based sediment deposition prediction and optimized dredging method.

[0019] The AI-based sediment deposition prediction and optimized dredging method and system of this invention achieves seamless access and unified management of multi-brand dredging equipment through multi-protocol adaptive conversion and data standardization; based on a two-level AI decision model, it integrates global optimization and local real-time response, significantly improving the overall efficiency and adaptability of dredging operations; combined with digital twins and closed-loop correction, it ensures the accuracy of command execution and system stability; at the same time, through dynamic avoidance and sediment diffusion suppression mechanisms, it effectively balances navigation safety and operational continuity, reduces the risk of secondary deposition, and thus comprehensively improves the intelligence level and overall benefits of dredging projects. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the process of the AI-based sediment deposition prediction and optimized dredging method provided by the present invention;

[0021] Figure 2 This is a schematic diagram of sediment deposition data acquisition and conversion in the AI-based sediment deposition prediction and optimized dredging method provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the optimization process curve of the non-dominated sorting genetic algorithm in the AI-based sediment deposition prediction and optimized dredging method provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the decision-making process for handling sudden anomalies in the AI-based sediment deposition prediction and optimized dredging method provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the motion trajectory of two devices and the conflict prediction time window in the AI-based sediment deposition prediction and optimized dredging method provided by the present invention.

[0025] Figure 6 This is a schematic diagram of the path optimization process of the dynamic avoidance strategy in a complex environment in the AI-based sediment deposition prediction and optimized dredging method provided by the present invention.

[0026] Figure 7 This is a schematic diagram illustrating the dynamic changes of the target performance index at different time steps in the two-level optimization process of global and local optimization at different time steps in the AI-based sediment deposition prediction and optimized dredging method provided by this invention.

[0027] Figure 8 This is a schematic diagram illustrating the implementation of the AI-based sediment deposition prediction and optimized dredging system provided by the present invention.

[0028] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] The following description, with reference to the accompanying drawings, describes an AI-based method, system, and electronic device for predicting and optimizing sediment deposition based on dredging.

[0031] Example 1:

[0032] Figure 1 This is a flowchart illustrating an AI-based method for predicting and optimizing dredging sediment deposition, according to an embodiment of the present invention. The method specifically includes the following steps:

[0033] S1: This method begins with the data access step. In this step, sediment deposition correlation data and raw operational data of multiple brands of dredging equipment are collected from the target water area through a multi-protocol conversion gateway deployed at the edge. Sediment deposition correlation data may include, but is not limited to, environmental parameters such as water depth and topography data, suspended sediment concentration, water flow velocity and direction, and sediment particle size distribution; multiple brands of dredging equipment refers to at least two different brands of dredging equipment, such as cutter suction dredgers and trailing suction hopper dredgers from different manufacturers.

[0034] Among them, the multi-protocol conversion gateway, as the hub of data aggregation, is responsible for establishing connections with various physical devices through multiple industrial communication interfaces, such as ModbusTCP, PROFIBUS, CAN bus, etc., and receiving raw operating data encapsulated by their private protocols in real time.

[0035] For example, the raw operating data may include hundreds of parameters such as the equipment's rake head depth, cutting depth, sludge concentration, hopper capacity utilization, motor speed, pump pressure, and fuel consumption. These parameters vary depending on the brand and model, and their data frame structure, encoding rules, and physical meanings are all significantly different.

[0036] S2: Upon data access, the system immediately executes a data standardization process. The core of this step lies in using a multi-protocol conversion gateway to parse and unify the format of the raw operational data, thereby generating standardized data that can be used for subsequent AI model analysis. The protocol parsing process relies on a pre-stored protocol dictionary, which is essentially a set of extensible mapping rules that stores the conversion relationships between the proprietary industrial protocols of known brands of equipment and unified standard formats. For example, for a specific brand of dredging equipment, the parameter identifier representing "dredge depth" in its proprietary protocol might be the byte sequence 0xAA0xBB, while the value uses big-endian 16-bit integer encoding. By querying the protocol dictionary, the gateway can accurately parse it into a floating-point "operating depth" value.

[0037] The unified format focuses on semantic integration, mapping heteronymous parameters for the same physical concept in different devices to unified parameter names, thereby eliminating terminological ambiguity. Specifically, the system uniformly maps parameters representing digging depth, such as rake depth and cutting depth, to working depth h; and uniformly maps parameters representing material loading, such as sludge concentration and hopper utilization rate, to material loading rate L.

[0038] The formula for calculating the material loading rate L is defined as follows:

[0039] ;

[0040] In the formula, The actual material quantity refers to the volume or mass of the silt mixture loaded by the dredging equipment at the current moment. Rated material quantity refers to the maximum amount of material that the equipment is allowed to safely load under design operating conditions.

[0041] Through this mapping, regardless of the device brand, its core state can be described using a unified variable system, which greatly simplifies the complexity of subsequent data processing and model building.

[0042] Optionally, before generating standardized data, the system also performs a rigorous outlier filtering step to further improve data quality. This step employs a two-stage filtering strategy:

[0043] The first level is the initial filtering based on the rated parameter range of the equipment. The system will pre-store the normal operating parameter range of various types of equipment. For example, the rated operating depth range of a certain type of cutter suction dredger is five to twenty meters. If the operating depth value in the received raw data is fifty meters, it will be judged as obviously abnormal and removed.

[0044] The second level is cross-validation filtering based on adjacent data of similar equipment. For example, multiple devices of the same model deployed in the same work area should have certain spatiotemporal correlation in their operating parameters. If a parameter value of a certain device deviates significantly from the data of neighboring devices in the same period, such as its unit energy consumption suddenly soaring to more than three times the average value, the system will mark the data as a suspicious value and perform secondary verification or filtering in combination with context information.

[0045] This combined filtering mechanism effectively resists noise data contamination caused by sensor failure, transmission interference, or sudden changes in operating conditions.

[0046] Optionally, a key feature of this embodiment is its adaptive protocol learning capability for newly connected dredging equipment of unknown brands. When the multi-protocol conversion gateway detects a new device in the network that is not registered in the protocol dictionary, it automatically triggers a protocol learning process, including the following:

[0047] The process begins with the system controlling the unbranded dredging equipment to execute a standard sequence of actions containing preset numerical gradients. This sequence is carefully designed to elicit data frames from the equipment that encompass the changing patterns of its key operating parameters. For example, it might instruct the equipment to operate at 10%, 20%, or even 100% of its rated power in stages, or control its rake arm to gradually swing from its minimum angle to its maximum angle. Simultaneously, the gateway continuously collects the raw data frames output by the equipment.

[0048] Subsequently, the system utilizes sequence alignment algorithms, such as dynamic time warping or sequence analysis algorithms based on hidden Markov models, to perform in-depth analysis of these raw data frames. By comparing the correspondence between the instruction sequence and the response data frame, the algorithm can identify parameter identifier bits, i.e., specific byte or bit patterns in the data frame used to distinguish different parameters, and parse out the numerical encoding rules, such as whether integers or floating-point numbers are used, whether big-endian or little-endian, and whether a checksum exists.

[0049] Finally, the system generates a new set of protocol parsing rules based on the identified rules and dynamically updates it to the protocol dictionary. At this point, the unknown brand device is successfully integrated into the system's unified management framework, achieving seamless plug-and-play access.

[0050] like Figure 2 It demonstrates the complete process from raw data collection to standardized data processing. Figure 2 The horizontal axis in the graph displays hourly data for the entire day, while the vertical axis represents the dynamic changes in sediment concentration.

[0051] Figure 2The blue curve represents the raw, unprocessed data. The frequent fluctuations in the curve indicate significant noise interference from external factors affecting the sampling equipment, including instrument accuracy errors and outlier deviations caused by changes in environmental variables. The red curve represents the result after standardizing the raw data using a data transformation algorithm. The standardized data... Figure 2 The data exhibits a smoother trend, and by eliminating outliers and normalizing adjustments, the consistency and comparability of the data have been significantly improved.

[0052] from Figure 2 The waveform changes can be further analyzed. The transformed data curve eliminates the random disturbances present in the original sampled data and reduces the non-uniformity of the peak region. Furthermore, Figure 2 The color changes in the medium-concentration data emphasize the convenience of comparative observation. The blue raw data intuitively presents the severity of noise interference, while the red reflects the effectiveness of data quality improvement.

[0053] S3: After the data standardization step is completed, the standardized data and sediment deposition correlation data are input together into the two-level AI scheduling model to generate dredging control instructions.

[0054] In this embodiment, a two-level AI scheduling model is introduced as a core decision-making component, but its specific implementation details, especially the algorithmic details of the global strategy layer and the local response layer, will be elaborated in subsequent embodiments. It should be clarified here that the input to this model is high-quality data that has undergone rigorous standardization and cleaning, and its output is abstract control commands for multi-brand equipment, including target operating depth, target material load rate, and travel speed.

[0055] S4: The system then enters the virtual-real synchronization and execution step. This step first converts the abstract dredging control commands generated by the two-level AI scheduling model into proprietary commands for each brand of equipment. This conversion process is the reverse of the data standardization step. It relies on the reverse mapping rules in the protocol dictionary to "translate" the unified format control commands into private protocol commands that the target equipment can recognize and execute.

[0056] For example, the system generates a unified instruction that is “set the working depth h to ten meters”. For equipment of brand A, this instruction is translated into sending a specific Modbus message to its PLC controller; for equipment of brand B, it may be translated into a write variable command based on the OPCUA protocol.

[0057] After the command is issued, the system synchronously drives the corresponding device twin in the digital twin system to perform simulated actions. The device twin is a high-fidelity dynamic model of the physical device in virtual space, reflecting the device's geometric state, kinematics, and dynamic behavior in real time. Through virtual-physical synchronization, operators can intuitively observe the expected actions that the device should perform on the digital twin interface, providing a visual reference benchmark for subsequent closed-loop calibration.

[0058] Optionally, to ensure a high degree of consistency between the physical device's actions and the simulated actions of the digital twin, this embodiment designs a sophisticated closed-loop calibration step. The core of this step is the real-time monitoring and comparison of the actual motion parameters of the physical device with the simulated motion parameters of the device twin. The system calculates the deviation rate between the two at preset time intervals, such as once per second or every five seconds. .

[0059] Among them, deviation rate The calculation formula is defined as follows:

[0060] ;

[0061] In the formula, The simulated motion parameters that represent the equipment twin, such as the simulated rake depth, cutter speed, or lateral movement speed; This represents the actual operating parameters of the physical device obtained through sensor measurements.

[0062] This formula quantifies the degree of deviation between the actual execution and the simulation expectations.

[0063] Optionally, the system presets an execution deviation threshold, such as 3%. When the calculated deviation rate δ is greater than or equal to 3%, the system determines that there is a significant execution deviation; subsequently, the system automatically activates the deviation cause analysis module. This module comprehensively examines equipment status data, environmental data, and historical operating data, and uses rule-based reasoning or lightweight machine learning models to quickly diagnose the possible root causes of the deviation. Possible causes include, but are not limited to, response lag caused by wear of mechanical transmission components, load surge caused by sudden changes in underwater geological conditions, sensor zero-point drift, or actuator jamming. Based on the analyzed deviation cause, the system generates correction instructions containing specific target parameter values. For example, if the diagnosed cause is water flow impact causing rake head positioning drift, the correction instruction might be "increase the rake head hydraulic cylinder pressure by 5% from the current level"; if the diagnosed cause is a decrease in mud pump efficiency, the correction instruction might be "increase the motor speed by 2% from the current level".

[0064] Optionally, the generated calibration command is also converted into a specific command for the corresponding brand of equipment by the command conversion module and immediately sent to the physical equipment for execution. After the command is issued, the system continuously repeats the above steps of calculating the deviation rate δ, forming a closed-loop control loop. This iterative process will continue until the deviation rate is reached. A stability below 3% indicates that the actual actions of the physical equipment have re-aligned with the simulation expectations of the digital twin. This real-time closed-loop correction mechanism based on digital twins effectively compensates for execution errors caused by factors such as equipment aging and environmental interference, ensuring that AI decisions can be accurately implemented in the physical world, thereby significantly improving the control accuracy and reliability of the entire intelligent dredging system.

[0065] The infrastructure built in this embodiment not only solves the compatibility and data consistency issues of multi-brand devices, but also provides clean and reliable data fuel and execution channels for upper-layer AI applications. Specifically: the adaptive protocol learning mechanism enables the system to expand to cope with future new devices, avoiding the risk of system obsolescence due to technological iteration; the unified parameter mapping model constructs a universal semantic understanding layer across devices and scenarios, laying the foundation for large-scale collaborative optimization; and the digital twin-driven closed-loop correction upgrades traditional open-loop control to an intelligent closed loop capable of self-sensing, self-diagnosing, and self-adjusting, significantly improving the robustness of system engineering applications.

[0066] Example 2:

[0067] This second embodiment provides a specific implementation scheme for the global strategy layer in the intelligent dredging method. It focuses on how to coordinate multiple dredging devices of different brands and performance levels to collaboratively complete dredging tasks across a large area of ​​water, achieving comprehensive optimization of multiple dimensions such as cost, energy consumption, navigation assurance, and equipment load balancing. It includes the following:

[0068] After the global strategy layer of this embodiment is activated, a dynamic multi-brand equipment capability matrix is ​​first constructed. This matrix is ​​the basic data model for global resource allocation and strategy optimization. It uses the unique equipment IDs of all online dredging equipment in the system as row vectors, with each equipment ID corresponding to a row in the matrix. A set of predefined real-time capability indicators is used as column vectors. These indicators collectively characterize the comprehensive working status and performance potential of each piece of equipment at the current moment. The real-time capability indicators include at least operating efficiency, unit energy consumption, sediment compatibility type, and maximum operating depth.

[0069] Among them: operating efficiency refers to the volume of mud and sand that the equipment can excavate and transport per unit time; unit energy consumption refers to the energy consumed by the equipment to process a unit volume of mud and sand; mud and sand compatibility type describes the type of mud and sand that the equipment is best suited to process, such as sand, silt or clay; and maximum operating depth reflects the limit of the effective excavation depth that the equipment can achieve under the current hydrogeological conditions.

[0070] This capability matrix is ​​a dynamically updated entity that evolves in real time with device status, environmental conditions, and task progress, providing accurate and timely input for subsequent optimization algorithms.

[0071] Optionally, after constructing the complete capability matrix, the global strategy layer invokes the Non-Dominated Sorting Genetic Algorithm (NSGA-II) as its core optimization engine. This algorithm is particularly suitable for handling complex problems with multiple conflicting optimization objectives. In this embodiment, the algorithm aims to find a Pareto optimal solution set for a dredging strategy, where each solution represents an optimal trade-off among multiple objectives such as cost, navigation, energy consumption, and load balancing. To quantify these objectives, the system defines a sophisticated multi-dimensional objective function. The objective function The mathematical expression can be defined as:

[0072] ;

[0073] In the formula, , , , These are the weights for cost, navigation reliability, total energy consumption, and equipment load deviation rate. These weights are configured and adjusted by the system administrator based on the specific project's priorities; for example, the navigation reliability weight can be increased in busy waterways. In energy-scarce regions, the total energy consumption weight can be increased. ;

[0074] It represents the actual unit dredging cost, which includes all related costs such as equipment depreciation, fuel consumption, and labor maintenance, and is converted into the unit dredging volume. The unit dredging cost represents a benchmark cost, a preset reference value for unit cost under ideal operating conditions, while the function term... It reflects the degree of cost savings compared to the benchmark cost; the larger the value, the better the cost control.

[0075] Represents the actual total energy consumption of all dredging equipment involved in the operation; Representing rated total energy consumption, it refers to the theoretical energy consumption estimate for completing the same dredging task under rated operating conditions; function term This represents the actual energy utilization efficiency; the smaller the value, the higher the energy efficiency.

[0076] This is the average load deviation rate of all devices, which measures the average deviation of the actual load rate of all devices in the system from the ideal average load rate. The load deviation rate threshold is an upper limit of the maximum allowable deviation; function term The larger the value, the more evenly the load is distributed across all devices, and the higher the overall health and stability of the system.

[0077] Air traffic reliability The impact of dredging operations on normal navigation traffic in waterways has been quantified, and its calculation formula can be defined as:

[0078] ;

[0079] in, The total time during which social vessels can pass through the work area safely and without delay during the designated navigation period. Without delay means that vessels do not need to slow down, wait, or change their planned routes due to dredging operations. The total navigation time is the total amount of time recorded, usually one calendar day or one tidal cycle.

[0080] Air traffic reliability A higher value indicates better coordination between dredging operations and shipping traffic, and a smaller impact on economic activities. The system accurately calculates this by accessing data from the Automatic Identification System (AIS) and combining it with predictions of ship tracks from a digital twin system. and .

[0081] Optionally, during runtime, the non-dominated sorting genetic algorithm uses the aforementioned multi-brand device capability matrix as the basis for generating the initial population. Each individual's chromosome encodes a possible device scheduling and task allocation scheme. The algorithm iteratively evolves the population through genetic operations such as selection, crossover, and mutation. In each generation, the algorithm calculates the multi-dimensional objective function of the scheme corresponding to each individual in the population. The algorithm then stratifies the solutions according to the non-dominated sorting rule and uses crowding calculation to maintain the diversity of solution distribution on the Pareto front. After a predetermined number of algebraic evolutions, the algorithm outputs a non-dominated solution set that approximates the true Pareto front.

[0082] Optionally, in the final decision-making stage, the system selects a final solution from this non-dominated solution set as the globally optimal dredging strategy. The selection rule can be based on pre-defined business rules, such as directly selecting the objective function. The optimal dredging strategy will be selected based on the highest value; alternatively, the system administrator can interactively choose from the Pareto front based on real-time preferences. The generated globally optimal dredging strategy will include a specific set of instructions, such as assigning the optimal operating area to each device, a recommended travel path, a target operating depth sequence, and a desired material load rate range. These strategy instructions will be passed to the execution steps described in Example 1, converted into device-specific instructions, and issued for execution, while simultaneously driving synchronous simulation using a digital twin.

[0083] like Figure 3 This demonstrates the dynamic changes of the Pareto front solution set across different generations in a multi-objective optimization problem, achieved through the evolutionary solution process of NSGA-II. The horizontal axis represents the first objective parameter in the optimization process, which can be understood as the optimization objective of equipment efficiency, while the vertical axis represents the second objective parameter, representing the optimization objective of energy consumption. The optimization process aims to achieve a trade-off between the two objectives, maximizing efficiency while minimizing energy consumption.

[0084] Figure 3 The scatter plots are colored to represent five different evolutionary generations, each with dozens of solutions. It's clear that with each generation, the solution set gradually shifts from the top of the plot to the lower right, becoming more concentrated and closer to the theoretically ideal Pareto front. The first generation (marked in blue) has a relatively scattered distribution of solutions, deviating from the target optimal region overall. From the second to the fourth generation, the solutions gradually converge to the lower right, indicating that NSGA-II can gradually generate better solution sets through evolutionary operations. The fifth generation (marked in green) achieves an optimal balance between efficiency and energy consumption in its Pareto solution distribution.

[0085] As the frontier solution set is updated, the genetic algorithm gradually improves the solution set through operations such as selection, crossover, and mutation, based on the multi-objective evaluation criteria in the method of this embodiment. Figure 3 The color changes of the scatter plots visually reflect the dynamic optimization process at different evolutionary stages, and the distribution trend of the objective function values ​​demonstrates a significant improvement in the performance of the optimization scheme. The numerical distribution of the solution set in the final generation is more concentrated and compact, reflecting the comprehensiveness and effectiveness of the algorithm in generating multi-objective equilibrium solutions.

[0086] Optionally, the global optimization process described in this embodiment is not completed in one go, but rather is a periodic, rolling process. The system will re-trigger the global optimization calculation at fixed time intervals, such as every hour, or when a significant change in the operating environment or equipment status is detected. This dynamic adjustment mechanism ensures that the dredging strategy can always adapt to changes in internal and external conditions and continuously maintain a globally optimal or suboptimal state.

[0087] Through the global strategy layer provided in this embodiment, the present invention achieves a leap from single-machine automation to multi-machine collaborative intelligence, and can seek to maximize overall benefits under complex constraints, which is unattainable by traditional control systems that rely on human experience or single-objective optimization.

[0088] Example 3:

[0089] Embodiment 3 of the present invention provides a specific implementation scheme for the local response layer in the intelligent dredging method. This scheme aims to solve the problems of transient and local abnormal working conditions and multi-device coordination conflicts that occur during the execution of the global strategy.

[0090] Unlike the global strategy layer in Implementation Example 2, which focuses on macro-level long-term planning, this implementation focuses on real-time decision-making at a micro-level short-term scale. By deploying a Deep Q-Network (DQN) and designing a dedicated reward function that includes a collaborative reward mechanism, the system possesses the ability to quickly perceive and respond to unexpected situations, thereby effectively ensuring operational safety, improving equipment collaboration efficiency, and suppressing the risk of secondary deposition. This includes the following:

[0091] The core decision engine of the local response layer is a specially trained deep Q-network. This network receives a high-dimensional state vector as input, which consists of real-time high-frequency data, mainly including the fine operating parameters of each dredging device at the current moment, local environmental sensing data in the working area, and short-term expected targets from the global strategy layer.

[0092] For example, operating parameters may include cutter head rotation speed, mud pump pressure, traverse speed, and real-time position coordinates; environmental sensor data may include instantaneous measurements of suspended sediment concentration, water flow velocity, and direction within a small area around the work site. The output of the deep Q-network is a set of Q-value evaluations for local equipment actions, with each Q-value corresponding to a possible discrete or continuous action, such as fine-tuning the cutter head rotation speed, changing the traverse speed, or briefly raising or lowering the rake head.

[0093] Optionally, to accurately guide the learning direction of the deep Q-network and ensure its decisions align with overall operational efficiency, this embodiment designs a reward function (Reward) that incorporates multiple constraints and objectives. This reward function is not solely goal-oriented but comprehensively balances four key aspects: processing efficiency, energy consumption control, risk avoidance, and collaborative effectiveness. Its mathematical expression can be defined as:

[0094] ;

[0095] In the formula, the coefficients 0.3, 0.2, -0.2 and 0.3 before the formula are the weights of each sub-objective, reflecting the different emphases of the system on processing speed, energy consumption, risk and collaborative efficiency.

[0096] Optionally, the first term of the reward function Reward is dedicated to improving the efficiency of exception handling; This represents the actual processing time from the moment the system identifies a local anomaly until the anomaly is effectively eliminated or controlled within an acceptable range. This represents the preset processing time threshold for this type of anomaly. If the time threshold is exceeded, it is considered that the processing is not timely and may trigger a chain reaction. The value increases as the actual processing time decreases, encouraging agents to take rapid and effective intervention measures.

[0097] Optionally, the second term of the reward function Reward focuses on the energy efficiency of local actions; This represents the actual energy consumption of the equipment or group of equipment involved in implementing the local adjustment strategy during the response to anomalies. This represents the rated energy consumption of the equipment or group of equipment when performing work of equal intensity under rated operating conditions. The value is positive when the actual energy consumption is lower than the rated energy consumption, and negative otherwise, thus guiding the deep Q network to choose energy-saving operation methods as much as possible while solving problems.

[0098] Optionally, the third term of the reward function Reward is designed to proactively mitigate the risk of secondary sedimentation caused by dredging operations; The secondary deposition risk quantification index is defined by the following formula: In the formula, The increase in the volume of newly generated silt within a certain monitoring range around the work site within a preset number of days after the completion of the current local dredging operation, such as 24 hours or 48 hours. This refers to the original volume of sediment within the same area before this localized operation was performed. The higher the value, the more significant the resuspension and redeposition effects of sediment in the surrounding environment caused by the current operation, i.e., the greater the risk of secondary deposition. This item has a negative weight in the reward function, meaning that any action that may lead to a high risk of secondary deposition will be penalized, prompting the agent to learn to adopt environmentally friendly operating modes.

[0099] Optionally, the fourth term of the reward function Reward focuses on optimizing the efficiency of collaborative work among multiple devices; It is a comprehensive measure of the actual collaborative efficiency of multiple devices, reflecting the degree of cooperation and coordination among multiple devices when jointly completing a task in a sub-area under a local adjustment strategy, such as the smoothness of task handover, the complementarity of work area coverage, and the consistency of overall progress. Multi-device target collaboration efficiency is the expected value of collaboration performance set based on global strategy and ideal operating conditions; The closer the value is to one, the closer the actual collaborative effect is to the ideal goal. This measure incentivizes the deep Q-network with positive weights to make decisions that improve the overall operational efficiency of the team.

[0100] Optionally, another key function of the local response layer is to identify coordination anomalies in real time based on standardized data. The system continuously analyzes the operating trajectories, speed profiles, and load status of multiple devices. When the identification results indicate that the operating areas of multiple devices overlap, i.e., there is a risk of spatial interference between the physical operating ranges of two or more devices, the system determines it as a spatial conflict anomaly. When the identification results indicate that the operating rhythm is mismatched, such as one device slowing down for some reason, causing subsequent devices to wait, or a bottleneck occurs in the material conveying process between devices, the system determines it as a timing conflict anomaly. Optionally, once the above coordination anomalies are identified, the system immediately triggers the local response layer to generate and execute a conflict resolution scheme, including the following:

[0101] For overlapping work areas, solutions may include dynamically redefining the fine work boundaries of each piece of equipment, adjusting the travel paths of the equipment to stagger the intersections, or instructing one of the pieces of equipment to briefly raise its rake head to make way.

[0102] For mismatches in work rhythm, solutions may include dynamically fine-tuning the working speed of relevant equipment to resynchronize, temporarily changing the working mode of a certain equipment to alleviate the bottleneck, or putting equipment waiting for instructions into a low-power standby state.

[0103] All these solutions are generated in real time by a deep Q-network based on the current state and the reward function.

[0104] like Figure 4 It demonstrates the process by which the equipment dynamically adjusts and restores its operating status through a decision-making response mechanism after a sudden anomaly occurs. Figure 4 The horizontal axis represents the running time, ranging from 0 to 20 units of time, while the vertical axis represents the actual running status value, such as the dynamic changes in speed, load, or other key performance indicators.

[0105] Figure 4 The green curve represents the equipment's operating status under normal conditions. The status value exhibits regular periodic fluctuations over time, with its average value remaining stable. This can be understood as the equipment's operating baseline under normal conditions. This curve reflects the equipment's efficient operating capability in a stable production environment. The red dashed curve represents the equipment's operating status during sudden anomalies. Within 8 to 12 time units, a sudden anomaly causes severe disturbances, resulting in a significant spike in the status value under abnormal interference, indicating that the equipment's response exceeds the normal fluctuation range. This anomaly can be mapped to common abnormal operating conditions during real-time operation, such as external interference, overload operation, and equipment component failure.

[0106] Figure 4 The blue curve represents the equipment's decision-making and response process under sudden abnormal conditions. It can be clearly seen that from the onset of the anomaly, the equipment gradually recovers to a normal state through the decision-making mechanism. This recovery process combines rapid response with a smooth transition, consistent with the dynamic recovery mechanism. The dynamic response portion of the panic phase—the rapid decline after anomaly detection and the subsequent gradual return to a stable state—fully demonstrates the effectiveness of the decision-making algorithm design, achieving both rapid suppression of spikes and avoiding secondary problems such as repeated oscillations.

[0107] from Figure 4 Data analysis reveals that within 2 to 3 time units after the anomaly phase ends, the blue curve smoothly reverts to near-normal (overlapping with the green curve). This indicates that the decision-making response mechanism can quickly intervene after detecting anomalies, effectively quelling the anomalies through dynamic strategy adjustments, and ensuring the system returns to stability within a short period.

[0108] In this embodiment, the decision-making process of the local response layer is a continuous perception-decision-action loop. The deep Q-network calculates the action with the highest Q-value based on real-time state variables, and this action is output as a real-time adjustment strategy. This strategy operates in parallel with the operation commands issued by the global strategy layer. The fine-tuning of the local response layer is a local parameter adjustment performed without violating the globally optimal dredging strategy. The generated adjustment commands are converted into dedicated commands and issued to the physical device, and synchronous simulation is performed in the digital twin system.

[0109] Through this embodiment, the system achieves agile response to dynamic risks during operation and precise control of multi-device collaboration, making up for the shortcomings of global strategies in terms of real-time performance, and together forming a complete intelligent decision-making system that takes into account both long-term planning and instantaneous response.

[0110] Example 4:

[0111] This fourth embodiment provides a dynamic avoidance scheme for handling conflicts between high-frequency navigation and equipment energy consumption in intelligent dredging methods. This scheme specifically addresses the spatial and temporal conflicts that may occur between dredging operations and other vessel navigation in densely navigable waters. It achieves this by real-time access to data from the Automatic Identification System (AIS), precise calculation of conflict time windows, quantitative assessment of the energy costs of different avoidance strategies, and the introduction of intelligent decision-making logic. This results in an optimal balance between navigation safety and operational energy consumption. The scheme includes the following:

[0112] The activation of this embodiment relies on real-time perception of the waterway traffic situation. The system seamlessly connects to the real-time data stream of the Automatic Identification System (AIS) via a data interface. This data includes key information such as the precise location, speed, course, type, and size of vessels. Based on this information, and combined with a predefined safe distance model for the operating area, the system dynamically calculates the estimated time for each vessel to enter the safe operating range of the dredging equipment. The safe distance model considers the operating radius of the dredging equipment, the maneuvering characteristics of the vessel, water flow conditions, and the safety margin required by regulations. Through a trajectory prediction algorithm, the system can identify potential conflict risks in advance and calculate the complete time interval from the current moment until the conflict occurs and is resolved; this interval is defined as the conflict time window. Conflict time window Essentially, it is a time span that represents the necessary duration for dredging equipment to clear its operating area for specific social vessels.

[0113] like Figure 5 The movement trajectories of the two devices during operation and the conflict prediction time window are shown. Figure 5 The horizontal axis represents time, covering an operating cycle from 0 to 10 hours; the vertical axis represents the position of the two devices, in meters. Figure 5 The solid blue and red curves represent the time-position trajectories of device 1 and device 2, respectively. Based on the speed and initial position settings, the trajectories of the two devices gradually converge after a period of time and briefly overlap within a certain range, indicating that there is a potential conflict zone between the two devices.

[0114] Figure 5 The area marked with semi-transparent yellow indicates the conflict time window, predicting the time range during which the distance between devices will be less than the safe distance (5 meters). As can be clearly seen in the diagram, the conflict time window is approximately between 3 and 5 hours. During this period, the distance between devices is less than the safe range, potentially leading to a collision risk. The yellow markings emphasize this high-risk area, providing a visual warning for the equipment system to avoid potential hazards.

[0115] This embodiment employs a dynamic prediction model based on position, speed, and relative distance, which can assess the risk of conflict between devices in real time during the operation plan and identify potential collision windows in advance. Figure 5 The intersection and conflict time windows of the two curves clearly reflect the core logic and capability of the decision-making algorithm optimization: by accurately calculating relative positions and limiting safe distances, it can predict in real time the time range in which target devices will physically intersect. For specific application scenarios, such as collaborative operation of multiple devices or automated scheduling systems, this method can avoid positional overlap between devices by adjusting trajectories or speeds, significantly improving the reliability and safety of system operation.

[0116] Optionally, during the acquisition of the conflict time window The system then immediately initiates an energy consumption assessment process to quantify the energy costs of two typical avoidance strategies:

[0117] The first strategy is non-stop, load-reducing avoidance. This strategy requires the dredging equipment to not completely stop operating during the avoidance period, but rather to keep the mud pumps idling while moving the equipment to a critical safe position. This position ensures a safe distance from other vessels and facilitates rapid resumption of operations after the conflict ends. Energy consumption for non-stop, load-reducing avoidance. This represents the total energy consumed in executing this strategy, and its calculation formula can be defined as:

[0118] ;

[0119] In the formula, The power required to maintain the dredging equipment at idle speed is mainly used to overcome the internal frictional resistance of the dredging pump and pipeline, maintain a minimum fluid circulation to prevent pipeline blockage, but its value is far lower than the rated operating power. The propulsion power required to yaw or partially move the dredging equipment from its current position to a critical safe position depends on the distance the equipment travels, water flow resistance, and the inertia of the equipment; the integral calculation represents the total time window of the collision. The cumulative energy consumption of maintaining power and mobile power.

[0120] The second strategy evaluated was complete shutdown avoidance. This strategy involves the system instructing the dredging equipment to completely cease sludge removal and other major operational functions within the conflict time window, putting the equipment into a low-power standby state. Complete shutdown avoidance consumes less energy. This characterizes the total energy consumed in executing this strategy, and its calculation formula is explicitly defined as:

[0121] ;

[0122] In the formula, This refers to the startup impact energy consumption required to restart the equipment from a completely static state and restore it to its rated operating conditions. This energy consumption is characterized by high peak value and short duration, and is mainly used to overcome static friction of the equipment, accelerate rotating parts, and re-establish normal mud transport conditions within the pipeline. This refers to the basic standby power consumed by the equipment during shutdown, used to maintain the normal operation of the control system, basic sensing units, and communication modules; The restart time of the equipment refers to the time elapsed from receiving the start command to the equipment stabilizing at its rated operating condition; in the formula, This indicates the pure standby time after deducting the restart time.

[0123] Optionally, after calculating the energy consumption of the two avoidance strategies, the system enters the core decision-making stage. The decision-making logic in this stage is based on a dual-condition judgment, aiming to select the avoidance scheme with lower overall energy consumption and operational feasibility.

[0124] Specifically, the judgment logic is expressed as follows: if the non-stop load reduction and avoidance energy consumption are simultaneously satisfied... Less than the energy consumption coupling coefficient Complete shutdown to avoid energy consumption The product of, and the conflict time window Less than the preset idling time threshold Then the two-level AI scheduling model will eventually output a non-stop load reduction and avoidance command.

[0125] Among them, energy consumption coupling coefficient This is an important adjustment parameter, with a value ranged from 0.6 to 0.9. It reflects the system's conversion factor for restart impact energy consumption and empirical corrections for the energy efficiency characteristics of different devices; idling time threshold. It is a preset time value based on equipment characteristics and engineering experience, used to determine the acceptable range of non-stop idling time, and to avoid excessive idling leading to uneconomical operation or equipment damage.

[0126] Optionally, when the decision logic determines that a non-stop, load-reducing avoidance strategy is adopted, the specific control instructions generated by the system include two core components. First, the instructions control the dredging equipment to maintain the mud pump in an idling state, thereby avoiding the risk of blockage and high restart energy consumption that may occur after the mud pump stops completely. Second, the instructions control the propulsion system of the dredging equipment to move semi-steadily to a pre-calculated critical safety position. This critical safety position is a dynamically calculated point that ensures an absolute safe distance from other vessels, while taking into account the impact of water flow on equipment drift. Furthermore, this position should allow the equipment to return to its original working point or connect to the next working position with minimal energy consumption and time cost after the conflict time window ends.

[0127] Optionally, in order to reflect the subtle impact of avoidance decisions on air traffic support rate in the global optimization objective, the system will, after making a non-stop load reduction avoidance decision, apply the following to the multi-dimensional objective function at the global strategy layer: Introducing a penalty factor Used to adjust the weight of air traffic support rate This operation is a feedback mechanism that acknowledges that in non-stop avoidance mode, equipment may not have completely left the work area, potentially causing a slight impact on air traffic efficiency, although this impact is far less than complete lane closure. This is achieved through dynamic fine-tuning of weights. The system ensures the consistency and accuracy of long-term global optimization goals, making the calculation of the air traffic support rate more in line with the complexity of actual operational scenarios.

[0128] Optionally, the energy consumption coupling coefficient The value of is not fixed, but rather a parameter that can be dynamically optimized using a machine learning model. The system continuously records actual energy consumption data, conflict time window data, and equipment status data from historical avoidance events, and uses this data to... The value is periodically calibrated.

[0129] For example, during periods when equipment is well-maintained and restarts efficiently, The value may approach the lower limit of the range, 0.6, indicating that the system is more inclined to consider a shutdown strategy; while during periods of equipment aging or low ambient temperature leading to increased restart energy consumption, The value may approach the upper limit of 0.9, making the system more inclined to choose non-stop avoidance. This adaptive parameter adjustment mechanism further improves the refinement and intelligence of avoidance decision-making.

[0130] Optionally, conflict time window The accuracy of the calculation directly affects the effectiveness of avoidance decisions. The system employs an advanced ship trajectory prediction algorithm, which not only relies on real-time data provided by the Automatic Identification System (AIS) but also integrates historical trajectory patterns, waterway traffic rules, meteorological and sea state data, and ship maneuvering models. Through Kalman filtering or deep learning sequence prediction models, the system can predict the future position of ships with high confidence, thereby calculating a more reliable conflict time window. For situations with high prediction uncertainty, the system will adopt a conservative strategy, appropriately widening the safety distance or triggering avoidance decisions in advance to ensure absolute safety.

[0131] like Figure 6 This diagram illustrates the path optimization process of a dynamic obstacle avoidance strategy in a complex environment. The horizontal axis represents the actual position of the device in the X-direction, and the vertical axis represents its actual position in the Y-direction. Green dots represent the starting position of the device, blue dots represent the final target position the device needs to reach, and large red circles represent the positions of static obstacles in the operating environment, such as potential devices, obstacles, or other insurmountable areas.

[0132] The solid black line shows the dynamically optimized movement path of the device when avoiding obstacles. From the starting point to the target point, the system designs an optimal path that meets the safety distance requirements based on the distribution of obstacles. This path is numerically represented by path points of 4 meters, 6 meters, and 8 meters in the X direction and the corresponding 4 meters, 6 meters, and 8 meters in the Y direction. As can be seen from the graph, the device makes significant path adjustments when encountering obstacles, such as detouring around the left and top of groups of obstacles, but still smoothly reaches the target point, demonstrating the flexibility and accuracy of the path planning algorithm.

[0133] Figure 6 As demonstrated, in dynamic obstacle avoidance tasks, the method of this embodiment can generate optimal path planning in real time based on target point requirements and obstacle distribution, while ensuring that the moving equipment remains within a dynamically safe position range. This dynamic planning not only effectively avoids collisions between the moving equipment and static obstacles, but also further optimizes the overall path's travel distance and time cost, exhibiting high technical reliability.

[0134] The dynamic avoidance method described in this embodiment is a highly automated closed-loop process. From AIS data access to conflict prediction, energy consumption calculation and decision generation, and finally command issuance and execution, the entire process requires no human intervention and is completed within seconds or minutes. This real-time response capability is crucial for handling the rapidly changing traffic situation in frequently navigable waterways. It effectively resolves the contradiction in traditional dredging operations where either excessive avoidance leads to a severe decrease in operational efficiency and a surge in energy consumption, or insufficient avoidance leads to navigation safety risks.

[0135] Example 5:

[0136] Embodiment 5 of this invention provides a scheme for suppressing sediment diffusion at its source in intelligent dredging methods for strong hydrodynamic environments. This scheme specifically addresses the technical challenges of suspended sediment diffusion, secondary deposition, and widespread siltation during dredging operations under strong hydrodynamic conditions such as high-speed water flow and complex flow directions. By real-time sensing of the hydrodynamic environment, quantitative calculation of disturbance diffusion risks, and dynamic triggering of refined thin-layer uplift and cutting modes, this embodiment can effectively suppress sediment diffusion at its source, improve the environmental friendliness and long-term effectiveness of dredging operations, and compensate for the shortcomings of traditional methods in adaptability to dynamic environments. The scheme includes the following:

[0137] The activation of this embodiment relies on precise perception of the hydrodynamic environment of the work area. During dredging operations, the system utilizes an acoustic Dozener current profiler deployed on the work vessel or underwater platform to collect real-time water velocity vector data of the work area. This instrument, by emitting acoustic waves and analyzing their echo frequency shift, can non-invasively measure the velocity and direction of flow in different water layers, generating a three-dimensional water flow field model. The collected water velocity vectors include not only the magnitude of the velocity but also directional information, providing a foundation for subsequent analysis.

[0138] Meanwhile, the system calculates the critical initiation shear stress based on bottom sediment characteristic data obtained through pre-survey or real-time sensing. Critical starting shear stress This is a key physical parameter that characterizes the minimum bed shear stress required for sediment particles to begin moving from a resting state under specific flow conditions. Its calculation typically relies on characteristics such as sediment grain size distribution, density, and shape factor, and can be performed using classical sediment transport mechanics formulas such as Shields curves or empirical models trained on local data. A higher value for this parameter indicates that the sediment is less easily initiated by water flow, and vice versa.

[0139] Optionally, after obtaining real-time hydrodynamic data and critical initiation shear stress, the system enters the core risk assessment phase, which involves calculating the disturbance diffusion flux index generated by the current dredging operation in real time. This index is a dimensionless or quantitative indicator with specific physical meaning, used to comprehensively characterize the intensity and potential impact range of suspended sediment diffusion caused by the combined effects of dredging operations and water flow. Its calculation formula can be defined as:

[0140] ;

[0141] In the formula, The concentration of suspended sediment around the work site is monitored in real time by optical or acoustic turbidity sensors. It is usually expressed in milligrams per liter or kilograms per cubic meter. It directly reflects the instantaneous abundance of suspended particulate matter generated by dredging disturbance. The modulus of the current water flow velocity, i.e. the magnitude of the flow velocity, is expressed in meters per second. It reflects the strength of the water flow's energy in transporting sediment. The angle between the water flow direction and the normal direction of the cutting surface of the dredging equipment affects the direct scouring and transport efficiency of the water flow on the suspended sediment. When the water flow direction is perpendicular to the cutting surface, The value is the largest, and the diffusion effect is the most significant. This is the critical initiation shear stress calculated above, which serves as the denominator and normalizes the calculation. A lower critical initiation shear stress means that the sediment itself is more prone to initiation, and may lead to a higher risk of diffusion even under weaker disturbances.

[0142] Therefore, the diffusion flux index The higher the value, the greater the risk of widespread sediment diffusion and subsequent backfilling under the current operating conditions.

[0143] Optionally, the system presets a diffusion warning threshold. As a benchmark for risk assessment, the diffusion warning threshold λ is a configurable parameter whose value is set based on aquatic environmental sensitivity requirements, historical observation data, numerical simulation results, or regulatory standards.

[0144] For example, in ecological protection areas or near drinking water intakes, thresholds may be set lower to allow for more stringent preventative measures.

[0145] The system will calculate the diffusion flux index in real time. With diffusion warning threshold Perform continuous comparisons. When the system determines the current calculated value... Greater than the preset threshold At that time, that is Regardless of whether the instantaneous increase in backfill volume detected in a local area exceeds other independent thresholds, the system will forcibly determine that there is a significant risk of backfill in a wide area.

[0146] This decision-making logic reflects the precautionary principle, prioritizing the suppression of the spread process that may have a wide-ranging and delayed impact on downstream or border control areas, rather than focusing solely on the immediate dredging effect at the work site.

[0147] Optionally, once the system determines that there is a risk of widespread siltation and forcibly triggers the response mechanism, a specialized operating process called the thin-layer siltation and diversion mode will be immediately initiated. The core idea of ​​this mode is to reduce the amount of sediment suspended and the diffusion rate at the source by reducing the intensity of a single cut and optimizing the interaction between cutting dynamics and water flow.

[0148] The implementation of the thin-layer uplift cutting mode involves several key operational adjustments: First, the system instructs the dredging equipment to significantly reduce the single cutting depth *h* from the original global or local strategy setting to within 50% to 70% of the original setting. For example, if the originally planned cutting depth was one meter, the adjusted cutting depth will be between 0.7 meters and 0.15 meters. This depth reduction directly reduces the volume of sediment stirred up in a single operation, thus reducing the amount of sediment released into the water body instantaneously.

[0149] Optionally, a second key adjustment to the thin-layer undulating cutting mode is the dynamic optimization of the reamer speed. The system adjusts the reamer speed n from the reference speed... Adjust to a value that corresponds to the real-time water flow velocity modulus. The dynamic speed values ​​that show a negative correlation The calculation formula can be expressed as:

[0150] ;

[0151] In the formula, The reference cutter speed is set under calm water flow or standard operating conditions; k is the speed suppression coefficient, which is a dimensionless empirical coefficient between zero and one, used to control the degree of speed reduction as the flow rate increases; a higher k value means that the speed is reduced more under strong water flow. The system records the historical maximum flow rate, which is used to normalize the current flow rate.

[0152] The physical meaning of this formula is that when the water flow velocity... When the flow rate increases, the water flow itself has a stronger sediment-carrying capacity. In this case, actively reducing the cutter speed can reduce excessive disturbance to the bed surface and avoid creating excessive suspended sediment beyond the water flow's carrying capacity, thereby inhibiting the formation and diffusion of sediment clouds. Conversely, at lower flow rates, a relatively higher speed is permissible to maintain a certain cutting efficiency.

[0153] Optionally, to compensate for potential losses in per-pass output due to reduced cutting depth and speed adjustments, the system strategically increases the number of cuts within the same area. This means the equipment will repeatedly traverse the same work strip, removing only a thin layer of sediment each time. By using this time-for-space strategy, while maintaining the total removal volume, the high-intensity concentrated disturbance is decomposed into a series of low-intensity dispersed disturbances, allowing the water more time to settle naturally and dilute the limited amount of sediment generated by each cut, thus significantly reducing the overall diffusion flux. The system continuously monitors the diffusion flux index after implementing the thin-layer sag cutting mode. This cyclical monitoring and adjustment process will continue until real-time monitoring detects [the issue]. The value decreased and stabilized at the diffusion warning threshold. The following are the conditions that are met. conditions.

[0154] Optionally, the rotational speed suppression coefficient k is not fixed but can be optimized through machine learning. The system accumulates historical operation data, including data at different flow rates. The diffusion flux index actually monitored when using different k values The system can dynamically adjust the k-value through regression analysis or reinforcement learning algorithms to achieve optimal dust suppression under specific hydrogeological conditions while minimizing the impact on operational efficiency. This adaptive parameter optimization ensures the refinement of the suppression strategy and its adaptability to different scenarios.

[0155] Optionally, the critical starting shear stress The calculations also need to consider its spatiotemporal variability. Within large work areas, the characteristics of bottom sediments may exhibit a heterogeneous distribution. Therefore, the system can combine high-frequency bottom sediment sampling data or bottom sediment classification maps based on acoustic inversion to construct spatially distributed... When calculating the risk of diffusion, the system will call up local data for that location based on the precise current location of the dredging equipment. This allows for a more accurate assessment of geological risks, making the assessment more closely reflect the actual conditions. This spatial refinement is particularly important in areas with dramatic variations in sediment types.

[0156] Optionally, the triggering and exiting mechanism of the thin-layer sag cutting mode needs to have a smooth transition to avoid mechanical shock to the equipment or drastic fluctuations in the work plan. When the system determines that it needs to enter the thin-layer sag cutting mode, the generation and issuance of control commands will follow a gradual ramp function, so that the adjustment of the cutting depth h and the reamer speed n will smoothly transition to the target values ​​within tens of seconds. Similarly, when the risk of diffusion is eliminated and the system exits the mode, the equipment parameters will gradually return to normal operating values. This smooth transition ensures the stable operation and lifespan of the equipment.

[0157] Optionally, the sediment diffusion source suppression method described in this embodiment has a functional synergy with the local response layer in Embodiment 3. The local response layer in Embodiment 3 may be based on the risk of secondary deposition. This triggers local adjustments, while this embodiment is based on a more forward-looking diffusion flux index. By intervening, the two can form a defense-in-depth system:

[0158] when Upon warning, this embodiment first initiates source suppression; if local accumulation still exceeds the limit, the mechanism of Embodiment 3 is activated for supplementary adjustments. This synergy ensures full-process control from diffusion potential to actual accumulation results.

[0159] Optionally, the effectiveness of this embodiment can be verified and optimized by deploying an additional monitoring network. Several suspended sediment concentration monitoring stations are deployed downstream of the work area to form a diffusion impact monitoring network. The system will compare the actual concentration distribution monitored by the network with... The predicted diffusion patterns are compared to continuously verify and refine the calculation model and threshold settings. This feedback loop based on actual results enables the system to continuously improve its prediction accuracy and the effectiveness of its suppression strategies.

[0160] The sediment diffusion source suppression method described in this embodiment transforms the traditional passive mode, which focuses on monitoring the results after dredging, into an active intervention mode based on real-time hydrodynamic sensing and diffusion prediction. By quantifying the diffusion flux index and coupling it with dynamic operational parameters, the system can take suppression measures before sediment spreads on a large scale, fundamentally reducing the risk of dredging operations impacting the water environment. It is particularly suitable for key water areas such as estuaries and waterways that are sensitive to water turbidity and have complex hydrodynamic conditions.

[0161] Example 6:

[0162] like Figure 8 As shown in the above method embodiments, this invention proposes an AI-based sediment deposition prediction and optimized dredging system, comprising:

[0163] The data acquisition and conversion module includes a multi-protocol conversion gateway and a protocol dictionary library, which is used to collect environmental and equipment data and perform protocol parsing and format unification to output standardized data;

[0164] The two-level AI scheduling module includes a global policy unit and a local response unit; the global policy unit is used to run a non-dominated sorting genetic algorithm to generate a global policy, and the local response unit is used to run a deep Q-network to make local anomaly decisions.

[0165] The digital twin monitoring module is used to construct a device twin and compare the deviation between the simulated actions and the actual actions of the physical device in real time.

[0166] The execution control module is used to convert global or local adjustment strategies into specific instructions for the corresponding brand of equipment and issue them.

[0167] A closed-loop feedback module is used to generate a correction instruction when the deviation exceeds a threshold, and to iteratively optimize the parameters of the two-level AI scheduling module based on scheduling performance indicators.

[0168] like Figure 7 This figure illustrates the dynamic changes in the target performance index at different time steps during the global and local two-level optimization process. The horizontal axis represents the time span of the optimization process, ranging from 0 to 20, corresponding to the time series of the global and local optimization algorithms in the simulation. The vertical axis represents the effect parameters related to the target performance during the optimization process, such as equipment operating efficiency, energy consumption, and resource utilization. The solid blue line represents the performance output of the global optimization method, while the dashed red line represents the performance result after superimposing local dynamic adjustments on top of the global optimization.

[0169] The blue global optimization curve illustrates a trend of overall steady decline with periodic fluctuations. The global optimization output initially has a high value (the initial performance target is 10), and gradually causes the performance index to change smoothly towards optimization. The decline process, which is negatively correlated with time, indicates that the global optimization strategy can effectively reduce energy consumption or improve efficiency over a large scale. Meanwhile, the small fluctuations on the curve indicate that some unstable factors in the operating equipment still exist in the global optimization output, leaving room for further local optimization adjustments.

[0170] The red dashed line represents the dynamic adjustment result of local optimization based on global optimization. It is clear from the figure that the local optimization strategy can mitigate the fluctuations present in global optimization and further refine the performance output. For example, within the time interval of 8 to 12, local optimization significantly suppresses the drastic fluctuations of the blue global trend, successfully compressing the red curve to a lower performance index range during this stage, thus demonstrating the precise compensation function of local optimization for global deviations.

[0171] This simulation illustrates the characteristics and advantages of the two-level optimization algorithm. Global optimization seeks overall performance improvement through macro-level control within a large decision-making scope, while local optimization further refines and compensates for positional error data based on a real-time adjustment mechanism on a global basis. Figure 7 The comparison of the curves shows that the two-level optimization strategy can work together to achieve rapid convergence of parameters over a wide range through global optimization, and then achieve fine compensation and dynamic adjustment through local optimization, thereby achieving a good balance between optimization speed and performance accuracy.

[0172] In summary, the system's workflow begins with the data acquisition and conversion module, which collects real-time sediment deposition correlation data and raw operating data of dredging equipment from multiple brands in the target water area through a multi-protocol conversion gateway, and performs protocol parsing and format unification to generate standardized data.

[0173] Subsequently, the standardized data is input into the two-level AI scheduling module, where the global strategy unit generates the globally optimal dredging strategy based on the non-dominated sorting genetic algorithm, while the local response unit processes the real-time high-frequency data based on the deep Q network to generate local anomaly adjustment strategies.

[0174] The execution control module converts the generated dredging control commands into specific commands for each brand of equipment and issues them for execution. Simultaneously, the digital twin monitoring module constructs a twin of the equipment and compares the actual motion parameters of the physical equipment with the simulated motion parameters of the twin in real time.

[0175] Finally, the closed-loop feedback module monitors the deviation. When the deviation exceeds the preset threshold, it generates a correction command to perform closed-loop control on the physical device and iteratively optimizes the parameters of the two-level AI scheduling module based on the scheduling effect index, forming a continuously optimized closed-loop system.

[0176] Example 7:

[0177] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0178] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0179] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0180] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0181] The memory 103 stores a computer program corresponding to the AI-based sediment deposition prediction and optimized dredging method of the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0182] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0183] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting and optimizing dredging sediment deposition based on AI computation, characterized in that, Includes the following steps: S1. Collect sediment deposition correlation data and raw operating data of multi-brand dredging equipment in the target water area through multi-protocol conversion gateways deployed on the edge side; S2. Using the multi-protocol conversion gateway based on the protocol dictionary, perform protocol parsing and format unification on the original running data to generate standardized data; S3. Input the standardized data and sediment deposition correlation data into the two-level AI scheduling model to generate dredging control instructions; wherein, the two-level AI scheduling model includes a global strategy layer and a local response layer, the global strategy layer uses a non-dominated sorting genetic algorithm to generate a globally optimal dredging strategy based on a multi-dimensional objective function, and the local response layer uses a deep Q-network to generate a real-time adjustment strategy for local anomalies based on real-time high-frequency data. The global strategy layer constructs a multi-brand device capability matrix with device ID as the row vector and real-time capability indicators as the column vector. The multi-dimensional objective function of the non-dominated sorting genetic algorithm Defined as: ； in, These are respectively the cost weight, the air traffic support rate weight, the total energy consumption weight, and the equipment load deviation rate weight; This refers to the actual unit dredging cost; The dredging cost is the benchmark unit. To ensure air traffic safety; This represents the actual total energy consumption. This is the rated total energy consumption; This represents the average load deviation rate of each device. The load deviation rate threshold; The deep Q-network of the local response layer employs a reward function that includes a collaborative reward mechanism. The reward function Defined as: ； in, Time spent on handling local anomalies; To set a processing time threshold; This represents the actual energy consumption of local equipment. Rated energy consumption for local equipment; Risk of secondary deposition; To improve the actual collaborative efficiency of multiple devices; To improve the efficiency of multi-device target collaboration; S4. Convert the dredging control command into a dedicated command for each brand of equipment and issue it for execution. At the same time, drive the equipment twin in the digital twin system to perform simulated actions. S5. Real-time monitoring of the actual motion parameters of the physical equipment and the simulated motion parameters of the equipment twin. When the deviation between the two exceeds a preset threshold, a correction command is generated to perform closed-loop control of the physical equipment.

2. The method according to claim 1, characterized in that, In step S2, when dredging equipment of an unknown brand is connected, the multi-protocol conversion gateway executes a protocol learning process, which includes: Control the dredging equipment of the unknown brand to execute a standard action sequence containing a preset numerical gradient; Collect the raw data frames output by the device when executing the standard action sequence; The original data frame is analyzed using a sequence alignment algorithm to identify parameter identifier bits and numerical encoding rules, generate new protocol parsing rules, and update the protocol dictionary.

3. The method according to claim 1, characterized in that, In step S2, generating standardized data includes mapping similar parameters from different devices to unified parameter names, specifically including the following: Map the rake head depth and cutting depth to the working depth. ; The sludge concentration and hopper utilization rate are uniformly mapped to the material loading rate. The material loading rate The calculation formula is: ； in, This refers to the actual quantity of materials. This is the rated material quantity; Before generating standardized data, an outlier filtering step is also included: initial filtering based on the rated parameter range of the equipment, and cross-validation filtering based on adjacent data of similar equipment.

4. The method according to claim 1, characterized in that, The air traffic support rate The calculation formula is: ； in, The duration of no delays during the operating hours; Total flight duration; The real-time capability indicators include operational efficiency, unit energy consumption, sediment compatibility type, and maximum operational depth.

5. The method according to claim 1, characterized in that, The risk of secondary deposition The calculation formula is: ； in, This is the increase in surrounding siltation within a predetermined number of days after dredging; This is the original amount of silt. Step S3 further includes: identifying collaborative anomalies based on the standardized data; if the identification result is that the operating areas of multiple devices overlap or the operating rhythm is mismatched, then the local response layer is triggered to generate a conflict resolution scheme.

6. The method according to claim 5, characterized in that, Step S5, which involves generating correction commands for closed-loop control of the physical device, specifically includes: Calculate the actual operating parameters of the physical equipment at preset time intervals. Simulation motion parameters of the equipment twin deviation rate : ； like If an execution deviation is detected, the system will automatically analyze the cause of the deviation and generate a correction instruction containing the target parameter value based on the cause of the deviation. The correction command is converted into a dedicated command by the command conversion module and then sent to the physical device. The above calculation steps are repeated until... .

7. The method according to claim 4, characterized in that, The method also includes a dynamic avoidance step to handle conflicts between high-frequency navigation and equipment power consumption: By accessing real-time data from the Automatic Identification System (AIS), conflict time windows can be calculated to determine the safe operating distance for other vessels. ; The system is based on the conflict time window Real-time calculation and comparison of energy consumption for complete shutdown to avoid collisions Energy consumption by non-stop load reduction and avoidance ; The energy consumption of non-stop load reduction and avoidance The calculation formula is: ； in, To maintain the power required for the dredging equipment to keep the mud pumps idling; The propulsion power required for the equipment to yaw to a critical safe position; The complete shutdown to avoid energy consumption The calculation formula is: ； in, The starting impact energy consumption required for the equipment to recover from a standstill to its rated operating condition; This is the base standby power during shutdown. This refers to the time taken during the restart process; The decision-making logic is as follows: like and The two-level AI scheduling model then outputs a non-stop load reduction and avoidance command, controlling the dredging equipment to maintain the operation of the mud pump and move it to a critical safe position, while simultaneously fulfilling the objective function. Introducing a penalty factor Adjust the weighting of air traffic support rate ; in, The energy consumption coupling coefficient has a value range of [value range missing]. ; This is the preset idling time threshold.

8. The method according to claim 6, characterized in that, The method also includes a step to suppress sediment diffusion sources in strong hydrodynamic environments: During dredging operations, an acoustic Doppler current profiler was used to collect the water velocity vector in the working area in real time, and the critical initiation shear stress was calculated based on the characteristics of the bottom sediments. ; Real-time calculation of the disturbance diffusion flux index generated by the current dredging operation The calculation formula is as follows: ； in, The concentration of suspended sediment is monitored in real time around the work site; This represents the current water flow velocity modulus. The angle between the direction of water flow and the dredging cutting surface; The suppression decision logic is as follows: The system sets a diffusion warning threshold. ;when At that time, it is determined that there is a risk of widespread siltation, regardless of the increase in siltation in the surrounding area. Whether the limit is exceeded, the system will forcibly trigger the thin-layer sag-cutting mode. The thin-layer sag cutting mode includes: increasing the single cutting depth Reduce to the original setting value ; Increase the auger speed Adjust to match the water flow rate The dynamic speed values ​​are negatively correlated and satisfy the following conditions: ； in, This is the speed suppression coefficient; This is the highest flow rate in history; The reference cutter speed is set under calm water flow or standard operating conditions; The production loss caused by the reduced depth of cut is compensated by increasing the number of cutting passes until the detected level is reached. .

9. An AI-based sediment deposition prediction and optimized dredging system, used to execute the AI-based sediment deposition prediction and optimized dredging method of claim 1, characterized in that, include: The data acquisition and conversion module includes a multi-protocol conversion gateway and a protocol dictionary library, which is used to collect environmental and equipment data and perform protocol parsing and format unification to output standardized data; The two-level AI scheduling module includes a global policy unit and a local response unit; the global policy unit is used to run a non-dominated sorting genetic algorithm to generate a global policy, and the local response unit is used to run a deep Q-network to make local anomaly decisions. The digital twin monitoring module is used to construct a device twin and compare the deviation between the simulated actions and the actual actions of the physical device in real time. The execution control module is used to convert global or local adjustment strategies into specific instructions for the corresponding brand of equipment and issue them. A closed-loop feedback module is used to generate a correction instruction when the deviation exceeds a threshold, and to iteratively optimize the parameters of the two-level AI scheduling module based on scheduling performance indicators.

Images