A ship multi-level dynamic speed tuning system and method thereof

Abstract

The application discloses a ship multilevel dynamic sailing speed optimization system, adopts a four-layer coupling architecture of a space planning layer Layer 0, a macroscopic strategy layer Layer 1, a rolling optimization layer Layer 2 and a supervision closed loop layer S4, Layer 0 generates a sailing waypoint and a sailing section, Layer 1 takes the minimum comprehensive cost per nautical mile as a target (Mode A), or switches to a Lagrange KKT pure fuel minimum mode (Mode B) when arriving at a port later than a contract ETA, obtains a segmented target sailing speed, Layer 2 takes the STW as an action and the cumulative mileage as a state, utilizes Bellman transfer to perform dynamic optimization, and adopts terminal state tolerance search and cross-window error carryover, and S4 performs monitoring and rescheduling according to the lateral deviation XTE and the ETA deviation, and the application improves the sailing speed consistency, the anti-interference and the energy saving effect.

Description

Technical Field

[0001] This invention belongs to the field of ship operation optimization, navigation control and ship energy efficiency management technology, and specifically relates to a multi-level dynamic speed optimization system and method for ocean-going vessels, which is used to coordinate the optimization of speed, fuel consumption, voyage progress and arrival time. Background Technology

[0002] Most existing ship speed optimization schemes employ single-layer or one-time solutions for the entire voyage. This means that before departure, based on predicted sea conditions, ship experience curves, or fuel consumption models, a constant or coarsely segmented speed recommendation is given for the entire voyage. While these schemes can reduce fuel consumption to some extent, they still suffer from the following technical drawbacks under the combined influence of dynamic sea conditions, contract arrival time (ETA) constraints, and execution disturbances: (1) Static solution results fail quickly: power and speed over water (STW) are approximately related. The nonlinear relationship. Using the forecast mean as input for optimization, a systematic bias of about 1% to 3% will occur during the execution phase due to Jensen's inequality.

[0003] (2) Inconsistent physical properties of decision variables: Existing technologies often use ground speed (SOG) as a decision variable. However, the real determinant of ship fuel consumption and main engine safety is surface speed (STW), which is significantly affected by ocean current disturbances. Using both variables together can lead to significant deviations in optimization results in ocean current regions.

[0004] (3) Inability to absorb short-term disturbances: Existing solutions cannot effectively absorb rapid short-term changes in wind, waves and currents at the hourly level, resulting in a large amount of hourly energy-saving potential being wasted.

[0005] (4) Lack of unified architecture and closed-loop mechanism: Functions such as minimizing overall cost, satisfying ETA constraints, ensuring the stability of flight segment execution and real-time monitoring feedback are usually scattered across different modules or even different systems. The lack of a unified hierarchical organization and closed-loop optimization mechanism makes it difficult to stably and reliably implement the algorithm results. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention proposes a multi-level dynamic speed optimization system and method for ships. The purpose of this invention is not to improve a single optimization algorithm, but to establish a hierarchical dynamic technical architecture for ship speed optimization scenarios.

[0007] This architecture decouples responsibilities through four layers: (1) Decoupling of spatial planning and time optimization; (2) Decoupling of macroeconomic optimization and micro-physical execution; (3) Automatic switching between the two objectives of minimizing overall cost and minimizing pure fuel consumption under the contract ETA constraint; (4) Dynamic optimization with hourly scrolling windows as the granularity; (5) A unified closed loop of lateral deviation monitoring, ETA progress monitoring and rescheduling triggering.

[0008] The system provided by this invention adopts a four-layer sequential coupling architecture of "spatial planning - macro strategy - micro execution - supervision closed loop". Each layer has different state variables, control variables and output results, and the output of the previous layer is used as the input of the next layer.

[0009] 1. Layer 0 - Spatial Planning Layer: This layer only handles spatial geometry planning and does not introduce time and cost variables. It generates waypoint and leg sequences based on the origin and destination port coordinates using the following formula.

[0010] Haversine's formula for great circle distance: ; ; ; in These are the latitude and longitude of the starting point (unit: radians), respectively. These are the latitude and longitude of the endpoint (unit: radians), respectively. , ; R The radius of the Earth is taken as 3440.065 nautical miles; D This is the great circle distance (unit: nautical miles).

[0011] Initial azimuth formula: ; in The initial great circle azimuth (in radians) is the angle from the starting point to the ending point. It is the arctangent function in the four quadrants.

[0012] Spherical trigonometric interpolation formula (used to generate intermediate waypoints): ; ; ; ; , ; in These are interpolation parameters. The coordinates of the generated intermediate waypoints (in radians).

[0013] 2. Layer 1 - Macroeconomic Strategy Layer: This layer is responsible for generating macroscopic target speeds or target progress at the segment scale. This layer does not directly handle hourly disturbances, but instead forms macroscopic targets for the execution layer to track based on commercial constraints, fuel costs, and ETA conditions, and has the function of automatic switching between dual modes.

[0014] Mode A (Minimum Overall Cost): Based on the overall cost per nautical mile The minimum is the objective function.

[0015] ; in v Speed ​​over ground (SOG), unit: knots; This is the price per unit of fuel oil, in US dollars per ton; Daily fuel consumption (tons / day). For ground speed v and environmental conditions w The main engine power is (unit: kilowatt), and SFOC is the main engine specific fuel consumption (unit: grams / kilowatt-hour). F Charter day premium, in US dollars per day; =CargoValue·r / 365 is the daily cost rate for capital tied up in goods (unit: USD / ton / day), CargoValue is the CIF value of the goods (unit: USD / ton), and r is the annualized cost of capital for the goods (unit: % / year). Storage fees (unit: USD / ton / day); This refers to the volume of goods in transit (unit: tons). The volume of cargo in Hong Kong (unit: tons).

[0016] When solving, first use a coarse sweep step size. =0.25 section generates velocity mesh: ; Find We obtain the smallest initial value and then use the golden ratio method to find the exact solution. (Golden ratio) By continuously narrowing the interval until Section, take The optimal ground speed.

[0017] Mode B (Minimum fuel, automatic switching): When the estimated arrival time calculated in Mode A (final ETA) is later than the contractually agreed time. When this happens, the system automatically switches to this mode. This mode aims to minimize total fuel consumption, with ETA (Electronic Toll Collection) as a hard constraint. ; Introducing Lagrange multipliers ≥0, construct the Lagrange function: ; According to the first-order optimality condition of Karush-Kuhn-Tucker (KKT), For a given The optimal speed for each leg of the journey can be obtained by using the ship's fuel consumption model. and flight time model Substituting, we get: ; Summarized as follows: ; in Indicates the first i The fuel consumption (in tons) for each leg of a journey at a constant ground speed is expressed as follows: ; In the formula This refers to the distance of the voyage (nautical miles). For ground speed (knots). To the average environmental conditions of the flight segment The main engine power is (kW), and SFOC is the main engine specific fuel consumption (g / kWh). Because... The above equation can be solved discretely using a power proxy interface and a fuel consumption model. Specifically, for each flight segment, within the speed range... The inner part can be found using the bisection method or the golden section method to satisfy the above equation. The partial derivatives The numerical difference method is used for approximate calculation: ; in This is a small value, for example, 0.01 knots. This numerical difference can be calculated quickly in each iteration, thus efficiently solving for the optimal speed for each segment.

[0018] If all flight segments have been solved, then calculate the total time. Then, through the analysis of... Perform a binary search until... Finally, the optimal speed sequence that satisfies the ETA constraints is obtained. .

[0019] It should be noted that Layer 1 is used to solve for the segmented target's ground speed. In this scenario, the average ocean current component within the navigation segment is used for time calculations, assuming that the ocean current remains constant within the segment. In actual navigation, spatial variations in ocean currents will cause differences in the surface speed (STW) and the target speed (STW) relative to the ground. The dynamic deviation between Layer 1 and Layer 2 will be corrected window-by-window by the rolling dynamic programming and window error carry-over mechanism. Therefore, the macro-level objectives of Layer 1 only provide an economically optimal schedule reference and do not require precise tracking.

[0020] 3. Layer 2 - Micro-rolling optimization layer: Based on the segmented target ground speed given in Layer 1, this layer uses a 3-hour rolling window to make local corrections based on real-time environmental data and outputs hourly-level surface speed (STW) commands.

[0021] Separation of action and state variables: Clearly distinguish between action variables and state update variables. The action variable is the speed over water (STW), which determines fuel consumption; the state update variable is the speed over land (SOG), used for mileage accumulation.

[0022] ; in This represents the axial component of the ocean current in the direction of the route (unit: knots), with positive for downstream.

[0023] Bellman transfer equation: within the rolling window, to discretize the STW mesh For the action space, use discretized cumulative mileage For state variables, optimal decisions are made through dynamic programming.

[0024] ; ; , in Indicates the first t Hourly cumulative mileage Minimum cumulative fuel consumption per hour (unit: tons); The time step is set to 1 hour; For the first t Hourly meteorological grid data (including information on wind, waves, currents, etc.); The step size for discretizing the mileage is set to 0.1 nautical miles.

[0025] Final state tolerance search and window error carry-over: Final state tolerance search: Due to the discretization of mileage, the probability of accurately hitting the target mileage is extremely low. Therefore, tolerance search is used to select the optimal final state.

[0026] ; ; in The target mileage (in nautical miles) to be completed in the current window. For tolerance, the default value is 0.2 nautical miles.

[0027] Cross-window error carryover: The residual between the actual mileage and the target mileage at the end of the current window. The process is carried over to the next window to achieve stable convergence for long-haul operations.

[0028] ; To prevent the error from accumulating and diverging, the residual is truncated: ; in The duration of the scrolling window is set to 3 hours. Maximum permissible speed (unit: knots); This is three times the tolerance, representing the upper limit of the allowable mileage error (default 0.6 nautical miles). The distance navigable within 0.1 times the window length at maximum speed is used as the absolute upper limit for error carryover (default 0.1 × 3 × 19 = 5.7 nautical miles). The smaller of these two values ​​is chosen as the cutoff boundary to suppress abnormally large residuals within a single window while preventing the infinite accumulation of errors across multiple windows. The coefficient 0.1 is selected based on the following engineering experience: within a single rolling window (3 hours), short-term disturbances in speed relative to the target value typically do not cause mileage deviations exceeding 0.1 times the window length multiplied by the maximum speed. This value is much greater than three times the tolerance (0.6 nautical miles), and as an absolute upper limit, it prevents residual runaway caused by extreme outliers; while... As a soft boundary for daily residuals, it allows small errors caused by dynamic programming (DP) discretization to be smoothly carried over across multiple windows.

[0029] 4. S4 - Supervision Closed-Loop Layer: This layer is independent of the three layers mentioned above. It continuously determines whether the current execution result is still within the allowed technical range and triggers recalculation when necessary.

[0030] Lateral Deviation (XTE) Monitoring: Calculates lateral deviation based on real-time vessel position and planned route.

[0031] ; ; ; in The coordinates of the starting point of the current flight segment. For real-time ship position coordinates, The azimuth of the planned flight segment. The bearing is the angle from the ship's position to the start of the voyage. The distance from the ship's position to the starting point of the voyage (unit: nautical miles).R This is the Earth's radius (nautical miles).

[0032] Dynamic thresholds and triggering mechanisms: The introduction of dynamic thresholds allows for greater tolerance in longer flight segments.

[0033] ; in The base threshold is 2.0 nautical miles by default. scale This is the scaling factor, with a default value of 0.05. This represents the length of the current segment (unit: nautical miles). When... continuous N Next (default) N =3) When the threshold is exceeded, the system suspends Layer 2 and triggers Layer 1 to re-execute the macro strategy solution with the current ship position as the starting point and the remaining ETA as the constraint.

[0034] Compared with the prior art, the present invention has the following beneficial effects: (1) By using a multi-level dynamic speed optimization architecture, geometric route generation, economic target allocation, execution layer dynamic correction and supervision closed-loop feedback are incorporated into a unified technical solution, which overcomes the problem that traditional solutions cannot simultaneously take into account planning accuracy, execution stability and real-time optimization capabilities.

[0035] (2) By decoupling Layer 1 and Layer 2, macro-business objectives and micro-physical execution are processed at different time scales, thus avoiding the expansion of model complexity.

[0036] (3) By introducing The variable relationships were established, which enabled a strict distinction between action variables and state update variables, allowing fuel decisions to be based on water speed, and improving the physical consistency and interpretability of the optimization process.

[0037] (4) By using the rolling window Bellman transition, final state tolerance search and cross-window error carry-over mechanism, the system can continuously absorb short-term environmental disturbances, avoid systematic deviations caused by Jensen's inequality in static optimization, and ensure the stability and convergence of long-haul dynamic optimization.

[0038] (5) By using an independent supervision closed-loop layer to uniformly monitor XTE and schedule deviation, this invention is upgraded from an "offline optimization tool" to an "online executable technical system".

[0039] (6) The responsibilities of each layer are clearly defined and the interfaces are clear, making it easy to deploy in the ship's industrial control computer or energy efficiency management system, and it has good engineering feasibility and scalability.

[0040] Other features and advantages of the invention will be set forth in the following description or may be learned by practicing the invention. Attached Figure Description

[0041] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0042] Figure 1 This is a general schematic diagram of the layered architecture of the present invention.

[0043] Figure 2 It is a Layer 1 dual-mode automatic switching process.

[0044] Figure 3 It is a Layer 2 rolling dynamic tuning process.

[0045] Figure 4 This is a Layer 2 two-dimensional state space and Bellman transition diagram; the horizontal axis represents the time step t=0,1,2,T (T is the number of discrete points within the window, usually T=3 corresponds to 3 hours); the vertical axis represents the discretized cumulative mileage. Different nodes represent different mileage states; solid arrows indicate an action (STW). k The state transition caused by ); the fuel consumption cost at each node accumulates through the Bellman transition equation; the final state (t=T) in the tolerance band I The node with the lowest cumulative fuel consumption is selected.

[0046] Figure 5 This is a structural diagram of the software application system of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to embodiments.

[0048] 1. Implementation of Layer 0: like Figure 1 As shown, this invention employs a four-layer sequential coupling architecture. This subsection describes the specific implementation of Layer 0.

[0049] Taking a transoceanic voyage from Singapore (1.3521°N, 103.8198°E) to Rotterdam (51.9225°N, 4.4792°E) as an example, the system first calls the Layer 0 module. This module takes the coordinates of the origin and destination ports as input and calculates the total great circle distance using the aforementioned Haversine formula. The initial azimuth angle was calculated, resulting in a total distance of approximately 8258 nautical miles. Subsequently, based on the preset number of waypoints or the maximum segment length, a series of intermediate waypoints were generated by uniformly interpolating between the origin and destination using spherical trigonometric interpolation formulas. Finally, Layer 0 outputs a sequence containing multiple segments. The sequence, each segment has a clearly defined start and end coordinates and distance. and planned azimuth .

[0050] 2. Implementation of Layer 1: like Figure 2 As shown, the automatic switching process between Mode A and Mode B is as follows: The Layer 1 module receives the segment sequence from Layer 0. The system initially defaults to Mode A (minimum overall cost mode). For each segment... The system reads preset economic parameters (fuel price per unit). = $650 / ton, time charter fee F = $15,000 / day, cargo value CargoValue = $1,500 / ton, cost of capital r = 5.0% / year, storage fee rate =0.5 USD / ton / day, etc.), ship hydrodynamic model (i.e. power proxy interface), and average environmental prediction data (wind, waves, etc.) for this segment.

[0051] Solution in Mode A: System in the velocity range Within a range of 8 to 19 knots, a coarse sweep is performed with a step size of 0.25 knots to calculate the comprehensive cost per nautical mile at each discrete speed. Find the After reaching the minimum speed range, the search phase begins using the golden ratio. By continuously narrowing the range until its length is less than 0.01 knots, the optimal ground speed for that flight segment is finally obtained. After solving for all flight segments, the estimated arrival times (ETAs) are aggregated to obtain the final ETA.

[0052] Mode switching determination: The system compares the final ETA with the arrival time stipulated in the contract. If final ETA> The system automatically switches to Mode B.

[0053] Mode B solution: System construction of Lagrangian function And for Lagrange multipliers In the interval Perform a binary search within the interval. For each trial value... The system solution satisfies Calculate the speed of each segment and the total sailing time. Through approximately 60 rounds of iterative search, until... The convergence process ultimately yields a set of optimal ground speed sequences that satisfy the ETA hard constraints. .

[0054] 3. Implementation of Layer 2: like Figure 3 and Figure 4 As shown, this layer uses a scrolling window and a two-dimensional state-space Bellman transition.

[0055] The Layer 2 module receives the output from Layer 1. As a macro-level objective, the system enters a cyclical optimization process with a 3-hour rolling window.

[0056] Window initialization: For a duration of With a 3-hour rolling window, the system first obtains the current ship's position within the segment (denoted as segment number). k (Segment) and remaining distance and the distance of subsequent segments. Then, follow these steps to determine the target mileage and dwell time for each segment within the window: (1) Let the cumulative time =0, cumulative target mileage =0, segment index i = k ; (2) Calculate the current flight segment i Above, at target ground speed Time required to complete the remaining distance ; (3) If ,but ,renew and move to the next leg of the journey ( i = i +1, and = Repeat step (2); (4) If If so, then only part of the time within the current flight segment belongs to the window, at which point... The target mileage contributed is The process ends after accumulation; (5) The final result This is the total target mileage that should be completed in this window. This refers to the dwell time of the window on each flight segment.

[0057] The above process ensures the consistency and uniqueness of the target mileage within the window with the macroscopic target speed. It should be noted that in actual navigation, due to factors such as ocean currents and waves, the ship's actual ground speed may deviate from the target ground speed. This causes the actual mileage within the window to differ from the target mileage. This produces a deviation.

[0058] This deviation will be gradually absorbed and corrected by the rolling dynamic programming of this layer (see the Bellman transfer equation later) and the cross-window error carry-over mechanism, and will not affect the effectiveness of window initialization.

[0059] Dynamic programming solution: The system discretizes the 3 hours within the window into 3 steps of 1 hour each. =1 hour). The state space is in mileage, in increments of 0.1 nautical miles ( Discretization is performed at time steps of 0.1 nautical miles. The action space is STW (Stop-Wave), ranging from 8 to 19 knots, with a step size of 0.1 knots. For each time step... t The system enumerates all possible current mileage states. d and all feasible STW actions. For each action, the system calls the power proxy interface, based on the current STW and the meteorological grid data for the current hourly location ( ) Obtain the host power Calculate the fuel consumption for this step. Meanwhile, based on the current axial component of the ocean current... Calculate the speed to ground The mileage is then updated. Finally, the two-dimensional DP table is updated using the Bellman transition equation. .

[0060] Results Output and Carry-over: When the window ends, the system returns to the target mileage. A tolerance zone nearby Within 0.2 nautical miles, search for the state with the lowest cumulative fuel consumption. The STW sequence corresponding to this state is the hourly instruction for this window. Simultaneously, the actual mileage reached is calculated. and residual After being cut off, it will be carried over to the next 3-hour window as a new starting point.

[0061] 4. Implementation of S4: The S4 monitoring layer continues to operate while the ship navigates according to the STW instructions output by Layer 2.

[0062] XTE Calculation: The system obtains real-time ship positions at intervals of 1 minute or less. Based on the starting coordinates of the current planned flight segment. End point coordinates and azimuth The lateral deviation XTE is calculated using the formula.

[0063] Exceedance Detection: The system calculates a dynamic threshold based on the length of the current segment. For example, for a 100-nautical-mile segment, the threshold is 2.0 + 0.05 × 100 = 7.0 nautical miles. If the XTE exceeds 7.0 nautical miles three times consecutively (e.g., for three consecutive minutes), the system determines that a major deviation has occurred.

[0064] Triggering rescheduling: Once the triggering condition is met, the S4 layer immediately suspends the current task in Layer 2 and triggers the rescheduling process, which specifically includes: (1) Reset starting point: based on the current real-time ship position As a new starting point, the sequence of remaining waypoints from this position to the destination in the original planned route remains unchanged (i.e., the spatial planning of Layer 0 is not re-executed, only the waypoints already traversed are truncated). (2) Update constraint: The remaining ETA under the contract (the original contract ETA minus the time already used) shall be used as the new arrival time constraint; (3) Re-execute Layer 1: Call the Layer 1 module, and based on the new starting point and the remaining segment sequence, re-execute Mode A (or Mode B) to obtain the new segment target-to-ground speed sequence. ; (4) Reset Layer 2: Clear all states of the current scrolling window (including the DP table and residual cumulative values), and start a new scrolling window. For the macro-level objective, we will re-enter the 3-hour rolling window optimization cycle from scratch; (5) Alarm prompts: Issue alarms on the user interface to prompt operators that route replanning has been performed.

[0065] The above process ensures the global consistency of the system after rescheduling and a smooth restart of rolling optimization.

[0066] 5. Baseline reference line: To fairly evaluate the energy-saving effect of the hierarchical dynamic speed optimization system described in this invention, a baseline reference line is established. The speed of this baseline is strictly derived from the following formula and cannot be manually entered: ; in Total route distance (in nautical miles) output for Layer 0; The arrival time (in hours) is the time agreed upon in the contract. If the contract ETA is not defined, the final ETA output from Layer 1 will be used. This represents the axial component of the average ocean current for the entire voyage (unit: knots), with positive values ​​for downstream currents.

[0067] After this reverse calculation, the Baseline can be compared fairly with Layer 1 (direct execution of macro-level segmented objectives) and Layer 2 (micro-level rolling optimization in this invention) under the same ETA constraint in terms of energy consumption.

[0068] 6. Power proxy interface: In this invention, the daily fuel consumption function for Layer 1 is: ; And Layer 2 single-step fuel consumption calculation: ; The main power used in All are provided by a unified power proxy interface.

[0069] In engineering implementation, this interface consists of a pre-trained regression model (such as XGBoost, Random Forest, or Neural Network). The model's features include, but are not limited to:

[0070] During training, the host power is used as the target value. After training, the output is the predicted power value of a certain STW under different sea conditions.

[0071] The training data comes from the ship's historical voyage log, including main engine speed, shaft power, fuel consumption rate, and corresponding meteorological and sea state data. The model can be trained offline and then deployed to the ship's industrial control computer, or it can be updated online periodically using new data collected during navigation. This regression model is a black-box module, and its specific structure (such as the number of hidden layer nodes, the number of trees, etc.) can be adjusted according to engineering experience and does not constitute a limiting feature of this invention.

[0072] The power proxy interface, as a replaceable black-box module, does not constitute a differentiating innovation of this invention. The core of this invention lies in its layered architecture and the decoupling and interaction methods between its layers. The specific implementation of the power proxy interface can employ any known ship power prediction method in the art.

[0073] 7. Software application system structure: like Figure 5 As shown, the software system of the present invention is divided into a presentation layer, a business layer, a service layer, and a data contract layer.

[0074] To enable the aforementioned multi-level dynamic speed optimization architecture to be implemented in the shipboard software, this invention further divides the software application system into the following layers: (1) Presentation layer: Provides user interaction interface for displaying route information, real-time speed instructions, fuel consumption statistics, ETA prediction and monitoring alarm information; at the same time, it receives user input of origin and destination port coordinates, contract ETA, economic parameters (fuel price per unit, time charter day price, etc.) and ship parameters.

[0075] (2) Business layer: Contains core algorithm modules, organized according to the layered architecture described in this invention: Layer 0 module: Implements spatial planning and outputs waypoints and segment sequences; Layer 1 module: Implements macroeconomic strategies, including automatic switching between Mode A and Mode B; Layer 2 module: Enables micro-roll tuning, including Bellman transfer, final state tolerance search, and cross-window error carry-over; S4 module: Implements a closed-loop supervision mechanism, calculates lateral deviations, and determines triggering conditions.

[0076] (3) Service layer: Provides data services and external interfaces, including: Meteorological data services: Obtain real-time or forecast grid data on wind, waves, currents, etc. Power proxy interface: Calls the regression model to calculate host power; Ship status service: Collects real-time data such as ship position, speed, main engine speed, and fuel consumption.

[0077] (4) Data Contract Layer: Defines the data exchange format between each layer and with external systems, including flight segment data structure, meteorological grid data structure, optimization result data structure, etc.

[0078] The aforementioned software layered structure enables the method described in this invention to be deployed in a modular manner on shipboard industrial control computers, energy efficiency management systems, or navigation assistance systems. The layers communicate with each other through clear data contracts, facilitating maintenance, upgrades, and portability.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, modifications, equivalent substitutions, and improvements can be made to the technical solutions of each embodiment, or equivalent substitutions can be made to some of the technical features, without departing from the spirit and principles of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-level dynamic speed optimization system for ships, characterized in that, It includes a sequentially coupled Layer 0 spatial planning layer, Layer 1 macroeconomic strategy layer, Layer 2 micro-rolling optimization layer, and S4 monitoring closed-loop layer, wherein: Layer 0, the spatial planning layer, is configured to generate waypoints and segment sequences from the origin and destination port coordinates using the Haversine great circle distance formula and spherical trigonometric interpolation formula, and output a set of segments. ; Layer 1, the macroeconomic strategy layer, is configured to calculate the comprehensive cost per nautical mile for each voyage segment. Minimize the objective function to find the optimal ground speed. And aggregate to obtain the final ETA of the estimated arrival time; when the final ETA > When the time comes, it automatically switches to the pure fuel minimum mode, which aims to minimize total fuel consumption, uses the contract ETA as a hard constraint, and is based on Lagrange multipliers. And the first-order optimality condition of KKT; Layer 2, a micro-rolling optimization layer, is configured to perform rotations over a fixed period of time. Within the scrolling window, the action space is defined by a discrete STW grid representing the water speed, and the cumulative mileage is defined by a discrete... As state variables, dynamic programming is used to solve the problem through the Bellman transition equation, and final state tolerance search and cross-window error carry-over are performed at the end of the window, outputting hourly STW instructions; The S4 monitoring closed-loop layer is configured to calculate the lateral deviation XTE based on real-time ship position, when If the dynamic threshold is exceeded N times consecutively, Layer 2 is suspended and Layer 1 is triggered to re-execute the solution with the current ship position as the starting point and the remaining ETA as the constraint.

2. The system according to claim 1, characterized in that, In Layer 2, the action variable is the water speed (STW), and the cumulative mileage update is based on the ground speed (SOG) = STW + ,in This represents the axial component of the ocean current in the direction of the route, thereby eliminating the physical consistency deviation within the ocean current domain when SOG is used as the action variable.

3. The system according to claim 1, characterized in that, The Layer 2 is set at the end of each window. Perform a final state tolerance search and assign mileage residuals. Carry forward to the next window, and process the residual as follows: Truncation is performed to prevent the cumulative divergence of multiple windows.

4. The system according to claim 1, characterized in that, The automatic switching to the pure fuel minimum mode specifically involves switching the solver to a Lagrange function-based mode. The pure fuel minimum mode, and for Perform a binary search until... convergence.

5. The system according to claim 1, characterized in that, Set a baseline for reference, the speed of which is determined by the formula. By reverse deduction, we can obtain that, Prioritize retrieving the contract ETA; otherwise, retrieve the final ETA output from Layer 1. This represents the average axial component of ocean currents for the entire voyage, used for energy consumption comparison under the same ETA constraint.

6. The system according to claim 1, characterized in that, It also includes the software application system architecture, which is divided into a presentation layer, a business layer, a service layer, and a data contract layer. The business layer is internally organized into core algorithm modules according to Layer 0, Layer 1, Layer 2, and S4.

7. A method for multi-level dynamic speed optimization of ships based on the system described in claim 1, characterized in that, The following steps are executed in sequence: S1: Generate waypoints and segment sequences from the coordinates of the origin and destination ports using the Haversine formula and spherical trigonometric interpolation; S2: Comprehensive cost per nautical mile for each leg of the journey. To achieve the target, the optimal ground speed is solved using coarse sweep and the golden section method, and the results are aggregated to obtain the final ETA. S3: Determine if final ETA> If the condition is true, switch to the pure fuel minimum mode based on the first-order condition of Lagrange KKT, and apply the Lagrange multipliers. In the interval Perform a binary search within the range until... convergence; S4: Enter a fixed duration A rolling loop, constructing the target mileage within each window. The system executes Bellman transition update dynamic programming (DP) tables using a discrete STW grid as the action space, performs final state tolerance search and window error carry-over, and outputs hourly STW instructions. S5: According to the formula Real-time monitoring of lateral deviation; after N consecutive exceedances of the threshold, Layer 2 is suspended and S2-S4 reassignment is triggered based on the current ship position and remaining ETA.

8. The method according to claim 7, characterized in that, In step S4, the duration of the scrolling window... =3 hours, time step =1 hour, distance discretization step size =0.1 nautical miles, tolerance =0.2 nautical miles.

Images