Energy efficiency promotion method and system based on route-speed-pitch collaborative optimization

Abstract

The present application relates to the technical field of ship energy efficiency management, and discloses an energy efficiency improvement method and system based on route-speed-drag force collaborative optimization, which constructs route feature coding based on sailing environment prediction information and electronic chart information, combines with real-time state information and decision variables of the ship, and generates voyage cumulative energy consumption expectation by using a ship energy consumption prediction model of physical information fusion; taking the minimization of the expectation as the goal, coupling sailing time constraints, main engine working condition constraints and ship stability safety boundaries, a three-variable collaborative optimization model of route path, planned speed of each voyage section and planned drag force value is constructed and solved, and a collaborative optimization scheme is output; in the sailing, the model is updated online based on real-time data and dynamic re-planning is triggered. Through the three-variable collaborative optimization framework and physical information fusion prediction, global voyage energy consumption optimization is realized, and with the help of online learning and rolling optimization mechanism, the adaptability and robustness of the optimization scheme in the dynamic environment are improved.

Description

Technical Field

[0001] This invention relates to the field of ship energy efficiency management technology, and in particular to an energy efficiency improvement method and system based on route-speed-curve collaborative optimization. Background Technology

[0002] In the ocean shipping sector, fuel consumption is a major component of vessel operating costs, making energy efficiency a crucial factor in reducing operating costs and minimizing environmental impact. Currently, in actual operation, vessels typically navigate based on experience or pre-set route plans, primarily controlling speed by adjusting main engine power to achieve scheduled arrival times or meet specific voyage requirements. Some advanced energy efficiency management systems consider sea state information, independently optimizing speed or draft (affecting trim) to reduce fuel consumption. These methods improve vessel operating economics to some extent. However, vessel navigation is a dynamic process influenced by a combination of factors, including route geography, speed, vessel attitude, and external environmental factors such as wind, waves, and currents. Existing technologies often focus on static or sequential optimization of single or a few variables, failing to adequately coordinate the complex spatiotemporal coupling relationships of these multiple variables. This fragmented optimization approach often limits energy efficiency improvement strategies to local optima, failing to minimize energy consumption from a global voyage perspective. Meanwhile, due to the lack of effective modeling of uncertainties in the navigation environment and a dynamic feedback mechanism based on real-time data, the adaptability and robustness of the optimization schemes in actual implementation are limited, making it difficult to cope with the constantly changing realities during the voyage. Therefore, how to achieve global, coordinated, and adaptive optimization of ship navigation strategies in complex and dynamic navigation environments is a major challenge in improving the overall energy efficiency of ships. Summary of the Invention

[0003] In view of this, the purpose of this invention is to propose an energy efficiency improvement method and system based on route-speed-curve collaborative optimization. By constructing a prediction model that integrates physical information and a three-variable collaborative optimization framework, and introducing online adaptive and dynamic replanning mechanisms, this invention solves the problem that existing ship energy efficiency optimization methods are difficult to achieve global voyage energy consumption optimization and dynamic adaptability due to variable fragmentation and static decision-making.

[0004] To achieve the aforementioned technical objectives, in the first aspect, the technical solution adopted by the present invention is: an energy efficiency improvement method based on route-speed-curve coordinated optimization, comprising:

[0005] Obtain voyage mission information, real-time ship status information, and navigation environment forecast information for the target vessel;

[0006] Based on navigation environment forecast information and electronic nautical chart information, a route feature code containing spatiotemporal correlation features is constructed;

[0007] The route feature code, real-time ship status information, and preset speed and trim decision variables are input into the physical information fusion ship energy consumption prediction model to generate the segment energy consumption probability distribution and the expected cumulative energy consumption of the voyage corresponding to different decision schemes.

[0008] With minimizing the expected cumulative energy consumption of a voyage as the core optimization objective, and coupling sailing time constraints, main engine operating condition constraints and ship stability safety boundaries, a three-variable collaborative optimization model is constructed, consisting of the route path, the planned speed of each segment, and the planned trim value.

[0009] A hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is used to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme, which includes the recommended route sequence, recommended speed for each segment, and recommended pitch value.

[0010] During the navigation process of the ship implementing the collaborative optimization scheme, the ship energy consumption prediction model is updated online based on the real-time collected actual navigation environment data and ship energy consumption data, and the dynamic replanning of the collaborative optimization model is triggered to generate rolling optimization instructions to adjust the speed and trim.

[0011] In some embodiments, based on navigation environment forecast information and electronic nautical chart information, a route feature code incorporating spatiotemporal correlation features is constructed, including:

[0012] Based on the port of origin and port of destination in the voyage mission information, the initial navigable area is determined on the electronic nautical chart. Combined with wind, wave and current data in the navigation environment forecast information, the initial navigable area is dynamically corrected to generate a set of candidate routes containing environmental risk information.

[0013] For each candidate route in the candidate route set, waypoints are extracted along the path corresponding to the candidate route at preset spatial intervals. Based on electronic chart information and navigation environment forecast information, static geographical features and dynamic environmental features are associated with each waypoint. Static geographical features include water depth, distance from shore and turning angle of the route, while dynamic environmental features include predicted wind speed, predicted wave height, predicted current speed and its direction relative to the ship's heading.

[0014] Clustering analysis algorithms are used to perform pattern recognition on the waypoint feature sequences of candidate routes to determine key turning points and typical segment divisions on the routes;

[0015] Based on the results of key turning points and typical route segment division, the feature sequences of each candidate route are restructured and vectorized to generate route feature codes that can characterize the global spatiotemporal attributes and local environmental changes of the route.

[0016] In some embodiments, route feature encoding, real-time ship status information, and preset speed and trim decision variables are input into a physical information fusion ship energy consumption prediction model to generate segment energy consumption probability distributions and voyage cumulative energy consumption expectations corresponding to different decision schemes, including:

[0017] Based on the energy transfer mechanism between the ship's main engine, propeller and hull, a basic energy consumption mechanism model of the ship is constructed with speed, trim and navigation environment as input variables.

[0018] By using historical ship energy consumption data and corresponding navigation status datasets, a Gaussian process regression model is trained to obtain a data-driven ship energy consumption black box prediction model.

[0019] Using the output of the basic energy consumption mechanism model as prior knowledge constraints, a weighted fusion is performed with the data-driven black-box prediction model to construct a physical information fusion ship energy consumption prediction model. The model parameters are adaptively tuned using a Bayesian optimization algorithm.

[0020] The environmental information of the route segment represented by the route feature encoding, the draft and load information of the ship in real time status information, together with the preset combination of speed and trim decision variables, are input into the trained physical information fusion ship energy consumption prediction model.

[0021] The ship energy consumption prediction model outputs the predicted probability distribution of the ship's energy consumption per unit time in a single voyage under the current decision combination;

[0022] Based on the probability distribution of energy consumption per segment, rolling cumulative calculation is performed on the segment sequence defined by the route feature encoding to obtain the expected cumulative energy consumption and its fluctuation range for the entire voyage at different confidence levels.

[0023] In some embodiments, the output of the basic energy consumption mechanism model is used as a prior knowledge constraint and weightedly fused with a data-driven black-box prediction model to construct a physical information fusion ship energy consumption prediction model, including:

[0024] The predicted energy consumption value of the basic energy consumption mechanism model under a given input is transformed into a Gaussian distribution with the mean of the predicted energy consumption value and the variance of the preset mechanism uncertainty, which is used as the prior probability distribution.

[0025] The prediction output of the data-driven black-box prediction model under the same input is represented as a Gaussian process posterior distribution with the model prediction value as the mean and the model prediction variance as the uncertainty.

[0026] Based on the Bayesian inference framework, the prior probability distribution is fused with the posterior distribution of the Gaussian process to calculate its posterior probability distribution;

[0027] The mean function and covariance function of the posterior probability distribution jointly define the ship energy consumption prediction model based on physical information fusion. The mean function is a weighted combination of the predicted values ​​from the basic energy consumption mechanism model and the predicted values ​​from the data-driven black box prediction model. The weight coefficients are determined synchronously during the model training phase through a Bayesian optimization algorithm.

[0028] In some embodiments, with minimizing the expected cumulative energy consumption of a voyage as the core optimization objective, a three-variable collaborative optimization model is constructed by coupling voyage time constraints, main engine operating condition constraints, and ship stability safety boundaries, including:

[0029] Based on the segment sequence and its associated dynamic environmental features represented by the route feature encoding, route path decision variables are defined, and their solution space is the set of candidate routes corresponding to the encoding.

[0030] The planned speed and planned trim value for each segment are defined as continuous decision variables, and their feasible domain is jointly defined by the safe speed range of the ship's main engine, the design speed range, and the allowable trim range determined based on the ship's hydrostatic curve.

[0031] To minimize the expected cumulative energy consumption per voyage, an objective function for the collaborative optimization model is constructed.

[0032] Construct a multimodal coupling constraint set, which includes:

[0033] The total voyage time calculated based on the distance of each segment and the corresponding planned speed must not exceed the time window constraint of the latest arrival time required in the voyage mission information.

[0034] The amplitude of the ship's motion response calculated from the navigation environment forecast information under the combination of planned speed and planned trim value for each segment shall not exceed the seakeeping and stability constraints of the preset safety threshold.

[0035] The variation of the planned trim value between adjacent segments is constrained by the dynamic feasibility of the maximum adjustment rate of the ballast water system.

[0036] The decision variables, objective function, and multimodal coupling constraint set of the route path, planned speed and planned pitch value for each segment are integrated and defined as a three-variable collaborative optimization model.

[0037] In some embodiments, constructing a multimodal coupling constraint set includes:

[0038] Based on navigation environment forecast information and route feature coding, identify the different dominant navigation environment modes that may be encountered in the current voyage. The navigation environment modes include at least the open water mode and the restricted severe sea state mode.

[0039] For each identified navigation environment mode, configure constraint activation rules and constraint parameters corresponding to the dominant risk of that navigation environment mode;

[0040] In the open water mode, the time window constraint and the energy consumption fluctuation range constraint output by the ship energy consumption prediction model based on physical information fusion are activated and strengthened. At the same time, under the premise of meeting the inherent seakeeping and stability safety thresholds of the ship, a smaller safety margin is allowed to be applied in the optimization calculation.

[0041] Under the restricted severe sea state mode, the seakeeping and stability constraints are activated and strengthened, allowing for a larger safety margin in the optimization calculation, adjusting the safety threshold of the ship motion response amplitude to a more stringent value, and converting the time window constraint from a hard constraint to a soft constraint that allows violation under certain penalty costs.

[0042] During the optimization process, the multimodal coupling constraint set dynamically calls and applies the constraint configurations under the corresponding navigation environment mode based on the environmental characteristics of the flight segments traversed by the route path decision variables.

[0043] In some embodiments, a hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is used to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme, including:

[0044] A hierarchical solution architecture is constructed, in which the first-layer solver is based on a metaheuristic optimization algorithm, performs a global search in the route path decision space defined by the collaborative optimization model, generates and maintains a population containing multiple candidate routes;

[0045] For each candidate route in the population, the second-layer solver is invoked for processing, including:

[0046] The second-layer solver is based on a sequential decision-making algorithm. It models the sequence of segments along the candidate route as a sequential decision-making process, using the planned speed and planned pitch value of each segment as decision variables. The values ​​of the decision variables must meet the constraints defined in the collaborative optimization model. With the goal of minimizing the expected cumulative energy consumption of the voyages along the route, it performs segment-by-segment or global optimization to obtain the optimal speed and pitch value sequence that matches the candidate route.

[0047] The optimal speed and pitch sequence obtained by the second-layer solver for each candidate route, along with the corresponding expected cumulative energy consumption for each voyage, are fed back to the first-layer solver as the basis for evaluating the fitness value of the candidate route.

[0048] The first-layer solver selects, crosses over, and mutates candidate routes in the population based on fitness values ​​to generate a new generation of population, and iteratively executes the above process.

[0049] When the iteration process meets the preset convergence criterion, the calculation is terminated, and the candidate route with the best fitness value in the final population and its corresponding optimal speed and pitch value sequence are output together as the globally optimal collaborative optimization scheme.

[0050] In some embodiments, the second-layer solver, based on a sequential decision algorithm, models the sequence of segments along the candidate route as a sequential decision process. Using the planned speed and planned trim value for each segment as decision variables, and aiming to minimize the expected cumulative energy consumption of voyages along the route, it performs segment-by-segment or global optimization to obtain the optimal speed and trim value sequence matching the candidate route, including:

[0051] The candidate route segment sequence is constructed as a multi-stage decision graph, where each node corresponds to the initial state of a segment, and the directed edges between nodes represent the execution of a specific combination of speed and pitch values ​​for that segment.

[0052] The ship energy consumption prediction model based on physical information fusion calculates the expected energy consumption of a given route under the environmental characteristics and ship status of each possible decision combination, which serves as the cost corresponding to that decision edge.

[0053] From the starting node to the ending node of the multi-stage decision graph, the dynamic programming algorithm is applied to search for the path that minimizes the sum of costs of each flight segment, i.e., the expected value of the cumulative energy consumption of the flight.

[0054] The optimal path obtained through the search is formed by combining the speed and trim values ​​of each decision edge, arranged in the order of the flight segments, to create a sequence of optimal speed and trim values ​​that matches the candidate route.

[0055] In some embodiments, during the navigation of a ship executing a collaborative optimization scheme, based on real-time collected actual navigation environment data and ship energy consumption data, the ship energy consumption prediction model is updated online with adaptive parameters, and dynamic replanning of the collaborative optimization model is triggered to generate rolling optimization instructions to adjust speed and trim, including:

[0056] During the voyage, the ship continuously collects measured data on the navigation environment, actual fuel consumption data of the main engine, and actual motion status data of the ship to form an online verification dataset;

[0057] The prediction output of the ship energy consumption prediction model corresponding to the current state is compared with the online validation dataset and physical information, and the prediction residual sequence of the model is calculated.

[0058] When the statistical characteristics of the model's predicted residual sequence exceed the preset model drift threshold, the online learning process is initiated. Using the latest online validation dataset within the sliding time window, some or all of the adjustable parameters of the physical information fusion ship energy consumption prediction model are incrementally updated to achieve adaptive correction of the model parameters.

[0059] Based on the updated physical information fusion ship energy consumption prediction model, and combined with the latest navigation environment forecast information and real-time ship status, the current ship position is used as the new starting point and the original destination port is used as the end point. The collaborative optimization model and solution process are re-executed to carry out local or global replanning.

[0060] The replanning process generates an updated collaborative optimization scheme from the current moment to the end of the voyage. The recommended speed and recommended trim values ​​for the upcoming next decision cycle are extracted from the updated collaborative optimization scheme and issued to the ship control system as rolling optimization instructions for execution.

[0061] In a second aspect, the present invention also provides an energy efficiency improvement system based on route-speed-margin collaborative optimization, applicable to the method described in the first aspect. The system includes a data sensing and acquisition module, a route feature processing module, a ship energy consumption prediction module, a collaborative optimization modeling and solving module, an online learning and dynamic replanning module, and an instruction execution and interface module. The data sensing and acquisition module is configured to acquire voyage mission information, real-time ship status information, navigation environment forecast information, and electronic chart information of the target ship, and to collect actual navigation environment data and ship energy consumption data in real time during navigation. The route feature processing module is connected to the data sensing and acquisition module and is configured to construct a route feature code containing spatiotemporal correlation features based on navigation environment forecast information and electronic chart information. The ship energy consumption prediction module is connected to the route feature processing module and the data sensing and acquisition module, and internally deploys a ship energy consumption prediction model based on physical information fusion. It is configured to receive the route feature code, real-time ship status information, and preset speed and margin decision variables, and generate segment energy consumption probability distributions and voyage cumulative energy consumption expectations corresponding to different decision schemes. The collaborative optimization modeling and solving module... The modeling and solving module is connected to the ship energy consumption prediction module. It is configured to minimize the expected cumulative energy consumption of a voyage as the core optimization objective, coupling voyage time constraints, main engine operating condition constraints, and ship stability safety boundaries to construct a three-variable collaborative optimization model for the route path, planned speed for each segment, and planned trim value. A hybrid intelligent solving algorithm combining metaheuristics and sequential decision-making is used to solve the model, outputting a globally optimal collaborative optimization scheme containing a recommended route path sequence, recommended speed for each segment, and recommended trim value. The online learning and dynamic replanning module is connected to the data sensing and acquisition module, the ship energy consumption prediction module, and the collaborative optimization modeling and solving module. It is configured to perform online adaptive parameter updates to the ship energy consumption prediction model based on real-time collected actual navigation environment data and ship energy consumption data during voyage, and trigger dynamic replanning in the collaborative optimization modeling and solving module to generate rolling optimization instructions. The instruction execution and interface module is connected to the collaborative optimization modeling and solving module and the online learning and dynamic replanning module. It is configured to send the globally optimal collaborative optimization scheme or rolling optimization instructions to the ship control system or display terminal.

[0062] Compared with existing technologies, the present invention, employing the above technical solution, has the following advantages: It constructs route feature codes based on navigation environment forecast information and electronic chart information, and combines real-time ship status information and decision variables to generate the expected cumulative energy consumption of the voyage using a ship energy consumption prediction model based on physical information fusion. With minimizing this expected value as the objective, it couples navigation time constraints, main engine operating condition constraints, and ship stability safety boundaries to construct and solve a three-variable collaborative optimization model for the route path, planned speed for each segment, and planned trim value, outputting a collaborative optimization scheme. During navigation, the model's parameters are updated online based on real-time data, triggering dynamic replanning. This invention achieves global voyage energy consumption optimization through a three-variable collaborative optimization framework and physical information fusion prediction, and effectively improves the adaptability and robustness of the optimization scheme in dynamic environments through online learning and rolling optimization mechanisms. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram of steps S101 to S106 of the energy efficiency improvement method described in the specific implementation method;

[0065] Figure 2 This is a schematic diagram of steps S201 to S204 of the energy efficiency improvement method described in the specific implementation embodiment;

[0066] Figure 3 This is a schematic diagram of steps S301 to S306 of the energy efficiency improvement method described in the specific implementation method;

[0067] Figure 4 This is a schematic diagram of steps S401 to S404 of the energy efficiency improvement method described in the specific implementation embodiment;

[0068] Figure 5 This is a schematic diagram of steps S501 to S505 of the energy efficiency improvement method described in the specific implementation. Detailed Implementation

[0069] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see Figure 1 In a first aspect, this embodiment provides an energy efficiency improvement method based on route-speed-curve coordinated optimization, including:

[0071] S101. Obtain the target vessel's voyage mission information, real-time vessel status information, and navigation environment forecast information;

[0072] S102. Based on navigation environment forecast information and electronic nautical chart information, construct route feature codes that include spatiotemporal correlation features;

[0073] S103. Input the route feature code, real-time ship status information and preset speed and trim decision variables into the physical information fusion ship energy consumption prediction model to generate the segment energy consumption probability distribution and voyage cumulative energy consumption expectation corresponding to different decision schemes.

[0074] S104. Taking the minimization of the expected cumulative energy consumption of the voyage as the core optimization objective, and coupling the voyage time constraint, main engine operating condition constraint and ship stability safety boundary, a three-variable collaborative optimization model is constructed, consisting of the route path, the planned speed of each segment and the planned pitch value.

[0075] S105. A hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is adopted to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme. The collaborative optimization scheme includes the recommended route sequence, the recommended speed for each segment, and the recommended pitch value.

[0076] S106. During the navigation process of the ship executing the collaborative optimization scheme, based on the real-time collected actual navigation environment data and ship energy consumption data, the ship energy consumption prediction model is updated online with adaptive parameters, and the dynamic replanning of the collaborative optimization model is triggered to generate rolling optimization instructions to adjust the speed and trim.

[0077] In step S101, the voyage mission information includes macro-level mission parameters such as the port of origin, port of destination, planned arrival time window, cargo load, and voyage economic objectives. Real-time ship status information covers parameters reflecting the ship's immediate operational status, including current draft, load condition, main engine speed, actual speed, ship position, and ballast tank status. Navigation environment forecast information refers to predicted data from meteorological and oceanographic forecasting agencies on dynamic environmental parameters such as wind speed and direction, wave height and direction, and current speed and direction for the planned navigation area over a future period. This information collectively forms the data foundation for subsequent optimization decisions, ensuring that the optimization plan is closely integrated with the specific navigation mission, the ship's own conditions, and external environmental conditions.

[0078] In step S102, the route feature coding process aims to fuse discrete electronic chart information with dynamic environmental forecast information to form a structured digital representation of a candidate route in both spatial and temporal dimensions. Electronic chart information provides static geospatial data, such as channel boundaries, water depth, and the location of navigational obstructions. Navigation environmental forecast information provides dynamic sea state data that changes over time. When constructing the route feature code, a candidate route needs to be discretized into a series of waypoints, and each waypoint is associated with its geographical location, static chart features, and the dynamic environmental features of that location during the prediction period. Through this association, the final route feature code can simultaneously represent the spatial direction of the route and the environmental conditions expected to be encountered along the route at different points in time (spatiotemporal correlation), thus providing rich contextual information for subsequent energy consumption prediction.

[0079] In step S103, the physical information fusion ship energy consumption prediction model integrates the physical mechanisms of ship navigation with data-driven methods. Real-time ship status information (such as draft) and preset speed and trim decision variables jointly determine the ship's navigation attitude and resistance in a specific environment. Route feature encoding provides the environmental context for each segment. After receiving these inputs, the model, through internally fused mechanisms and data knowledge, outputs the probability distribution of ship energy consumption within a single future segment, rather than a single deterministic value. This reflects a quantitative consideration of the uncertainty of the navigation environment. Based on the energy consumption probability distribution of each segment, cumulative calculations are performed along the entire route to obtain the expected value of the cumulative energy consumption required to complete the entire voyage and its possible fluctuation range. This step transforms the complex multivariate coupling relationship into a quantifiable energy consumption expectation index, providing a clear objective function for global optimization.

[0080] In step S104, the optimization objective of the three-variable co-optimization model directly follows from the previous step, namely, minimizing the expected cumulative energy consumption of the voyage. To achieve this objective, the model simultaneously optimizes the route path, the planned speed for each segment, and the planned trim value as decision variables. The trim value refers to the difference between the bow and stern drafts, which is optimized by adjusting the distribution of ballast water in each compartment to change the ship's attitude, thereby affecting its resistance and propulsion efficiency. Constraints ensure the feasibility and safety of the optimization scheme: the voyage time constraint requires the total voyage time to not exceed the latest arrival time required by the mission; the main engine operating condition constraint limits the feasible range of speed to avoid overloading or inefficient operation of the main engine; and the ship's stability safety boundary limits the range of trim value variation to ensure safe navigation under various sea conditions. By placing these three variables, which are often considered independently, within the same optimization framework and coupling them with multiple practical constraints, the model can explore the globally optimal energy efficiency scheme.

[0081] In step S105, for the aforementioned complex high-dimensional, nonlinear, and multi-constraint optimization model, a hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is employed. Metaheuristic algorithms (such as genetic algorithms and particle swarm optimization) excel at global exploration in a broad, discrete decision space (such as different route choices), avoiding getting trapped in local optima. Sequential decision-making algorithms (such as dynamic programming) are adept at handling continuous decision problems with sequential dependencies (such as optimizing speed and trim segment by segment along a fixed route). Preferably, the hybrid solution strategy is implemented in layers: the upper layer uses metaheuristic algorithms to search among numerous candidate routes; for each evaluated candidate route, the lower layer calls sequential decision-making algorithms to quickly solve for the optimal speed and trim sequence along that route, and feeds back the corresponding cumulative energy consumption expectation value to the upper layer as a basis for evaluating the route's merits. Through this collaborative solution, the final output is a complete collaborative optimization scheme, including a recommended global route and specific speed and trim suggestions for each segment of that route.

[0082] In step S106, since there will inevitably be discrepancies between the actual environment and the forecast information during actual navigation, and the ship's state is also constantly changing, this step compares the actual navigation environment data (such as measured wind, waves, and currents) and ship energy consumption data (such as actual main engine fuel consumption) with the predicted values ​​of the ship energy consumption prediction model in real time. When the deviation continues to exceed a certain threshold, it indicates that the model's prediction accuracy has decreased due to environmental changes. At this time, the online learning process is initiated to fine-tune the parameters of the prediction model using the latest measured data, so that the model can adapt to the new navigation conditions. On this basis, taking the ship's current position as a new starting point, and combining the updated prediction model and the latest environmental forecast, the optimization process that started from step S102 is re-executed to perform local or global replanning. In the updated scheme generated by the replanning, the speed and trim commands for the next decision cycle (such as the next few hours) will be issued as rolling optimization commands to the ship control system for execution, thereby realizing closed-loop optimization control based on real-time feedback.

[0083] This embodiment places three key decision variables—route, speed, and trim—within a unified collaborative optimization framework. It uses minimizing the expected cumulative energy consumption for the entire voyage as the global objective for joint solution. By constructing a physical information fusion-based energy consumption prediction model, it effectively quantifies the impact of environmental uncertainties and multivariate coupling on energy consumption. A hybrid intelligent solution algorithm is designed to solve the problem of solving high-dimensional complex optimization models. Finally, by introducing online learning and dynamic replanning mechanisms, the static offline optimization scheme acquires adaptive capabilities to cope with dynamically changing environments. This embodiment explores deeper energy-saving potential at the global voyage level, achieving a systematic improvement in ship energy efficiency while ensuring navigation safety and time requirements.

[0084] Please see Figure 2 In some embodiments, based on navigation environment forecast information and electronic nautical chart information, a route feature code containing spatiotemporal correlation features is constructed, including:

[0085] S201. Based on the port of origin and port of destination in the voyage mission information, determine the initial navigable area on the electronic nautical chart, and combine the wind, wave and current data in the navigation environment forecast information to dynamically correct the initial navigable area and generate a set of candidate routes containing environmental risk information.

[0086] S202. For each candidate route in the candidate route set, extract waypoints along the path corresponding to the candidate route at preset spatial intervals, and associate static geographical features and dynamic environmental features with each waypoint based on electronic chart information and navigation environment forecast information. Static geographical features include water depth, distance from shore and turning angle of the route, while dynamic environmental features include predicted wind speed, predicted wave height, predicted current speed and its direction relative to the ship's heading.

[0087] S203. Use clustering analysis algorithm to perform pattern recognition on the waypoint feature sequence of candidate routes to determine the key turning points and typical segment divisions on the routes.

[0088] S204. Based on the results of key turning points and typical route segment division, the feature sequences of each candidate route are restructured and vectorized to generate route feature codes that can characterize the global spatiotemporal attributes and local environmental changes of the route.

[0089] In step S201, the initial navigable area is a preliminary delineation of the waters between the port of origin and the port of destination, based on electronic chart information and considering factors such as safe water depth, channel restrictions, and the location of obstructions. Wind, wave, and current data from the navigation environment forecast are used to dynamically correct this area. Specifically, the system assesses sea areas in the forecast data that exceed preset safety thresholds (e.g., wave height exceeding the ship's seakeeping limit, excessive countercurrent speed), and excludes these high-risk areas from the initial navigable area or marks them as high-cost areas. Through this dynamic correction, the final set of candidate routes not only considers static geographical constraints but also avoids sea areas with adverse environmental conditions during the forecast period, ensuring that route selection is based on both safety and energy efficiency.

[0090] In step S202, to construct a refined route representation, waypoints need to be set at fixed intervals (e.g., every 1 nautical mile) along each candidate route. For each waypoint, its static geographic features are extracted from the electronic chart: water depth data is directly read; the distance from the shore is obtained by calculating the geometric distance between the point and the nearest coastline; the turning angle of the segment is obtained by calculating the angle between the point and the direction of the line connecting the preceding and following waypoints, used to characterize the curvature of the route. Simultaneously, predicted environmental parameters for the waypoint's location at the ship's expected arrival time are extracted from navigation environment forecast information, including the magnitudes of wind speed, wave height, and current speed, and the angles between the directions of these environmental elements and the ship's planned course for that segment (e.g., headwind, with the waves) are calculated, thus forming a dynamic environmental feature vector. This step assigns rich static and dynamic attribute labels to each location point on the route.

[0091] In step S203, clustering analysis algorithms (such as K-means or DBSCAN) are used for unsupervised learning of the waypoint feature sequence to automatically identify the route structure. The algorithm's input is the feature vectors of all waypoints (containing geographical and environmental features). By calculating the similarity between feature vectors (such as Euclidean distance), the algorithm aggregates consecutive waypoints with similar features into the same cluster. The boundary points of these clusters typically correspond to locations where route characteristics change significantly, such as transitioning from open water to narrow channels, from downstream to upstream areas, or areas of abrupt environmental changes. These locations can be identified as key turning points or segment division points. The waypoint sequence between adjacent key turning points is then divided into a typical segment. This step, through a data-driven approach, discretizes continuous routes into a sequence of segments with relatively uniform characteristics, providing structured processing units for subsequent optimization.

[0092] In step S204, based on the key turning points and typical segments identified in the previous step, the original discrete waypoint feature sequence is structurally reorganized. Specifically, the features of all waypoints within a typical segment are aggregated and statistically analyzed (e.g., calculating the average water depth, average and variance of environmental features, segment length, and total turning angle within the segment) to generate a summary vector representing the overall characteristics of the segment. Then, in the order of the segments, these summary vectors and the feature vectors of the key turning points are sequentially concatenated to form a structured long vector. This long vector is the final route feature encoding, which includes both the overall spatial path information of the route (reflected through the segment sequence and turning points) and the environmental statistical characteristics and trends (spatiotemporal correlation) of each local segment, thus representing a complete candidate route in a compact and information-rich form.

[0093] This embodiment generates a safe candidate set through dynamic correction, integrates spatiotemporal information through refined waypoint annotation, and introduces cluster analysis to achieve intelligent segmentation of the route. Finally, it generates a structured feature code, which is not only a simple description of the route path, but also a quantitative carrier that integrates multi-dimensional information such as geography, environment, and navigation risks. This provides crucial and high-quality input features for subsequent high-precision energy consumption prediction and collaborative optimization, effectively improving the perception capability and decision reliability of the entire energy efficiency optimization system.

[0094] Please see Figure 3 In some embodiments, route feature encoding, real-time ship status information, and preset speed and trim decision variables are input into a physical information fusion ship energy consumption prediction model to generate segment energy consumption probability distributions and voyage cumulative energy consumption expectations corresponding to different decision schemes, including:

[0095] S301. Based on the energy transfer mechanism between the ship's main engine, propeller and hull, construct a basic energy consumption mechanism model for ships with speed, trim and navigation environment as input variables.

[0096] S302. Using historical ship energy consumption data and the corresponding navigation status dataset, train the Gaussian process regression model to obtain a data-driven ship energy consumption black box prediction model.

[0097] S303. Using the output of the basic energy consumption mechanism model as prior knowledge constraints, a weighted fusion is performed with the data-driven black-box prediction model to construct a physical information fusion ship energy consumption prediction model. The model parameters are adaptively optimized using a Bayesian optimization algorithm.

[0098] S304. Input the route feature encoding-represented segment environmental information, the ship's real-time status information including draft and load, together with the preset speed and trim decision variables, into the trained physical information fusion ship energy consumption prediction model.

[0099] S305. The ship energy consumption prediction model outputs the predicted probability distribution of the ship's energy consumption per unit time in a single voyage under the current decision combination.

[0100] S306. Based on the probability distribution of energy consumption of flight segments, the flight segment sequence defined by the feature encoding of the flight route is rolled and accumulated to obtain the expected cumulative energy consumption of the entire flight at different confidence levels and its fluctuation range.

[0101] In this embodiment, the basic energy consumption mechanism model of a ship is established based on the principles of ship hydrodynamics and propulsion. This model takes ship speed, trim value, and navigation environment information (such as the effects of wind, waves, and currents on the hull) provided by route feature encoding as input. By calculating the total resistance of the ship under a given attitude and environment, and considering the propeller propulsion efficiency and main engine fuel consumption characteristic curves, it finally outputs a theoretical estimate of energy consumption per unit time. This model reflects the physical laws of energy transfer, but its accuracy is limited by the simplification of the model and the accuracy of the input parameters.

[0102] The data-driven black-box prediction model for ship energy consumption is constructed using Gaussian process regression. The historical dataset used for training contains a large amount of data on ship status (such as draft and load), environmental conditions, speed, trim, and corresponding actual main engine fuel consumption recorded from historical voyages. By learning the complex nonlinear relationships in this data, the Gaussian process regression model can predict energy consumption under new input combinations and output the prediction results in the form of a probability distribution, where the mean represents the predicted energy consumption and the variance characterizes the uncertainty of the prediction.

[0103] The physical information fusion-based ship energy consumption prediction model is constructed by weightedly fusing the two types of models mentioned above. Specifically, the deterministic values ​​output by the mechanistic model are treated as prior knowledge with a certain degree of uncertainty, while the probability distribution output by the data-driven model is used as the observation likelihood, and fusion is performed within a Bayesian inference framework. The key fusion weight parameters in the model are adaptively tuned during the training phase using a Bayesian optimization algorithm. This optimization process aims to minimize the prediction error of the model on the validation set, automatically finding the optimal parameter combination so that the fused model possesses both physical rationality and data adaptability.

[0104] The trained fusion model, during actual prediction, receives segment environmental information from route feature encoding, the ship's real-time draft and load status, and a set of decision variables to be evaluated: speed and trim. After internal processing, the model outputs a probability distribution of the ship's energy consumption per unit time within that segment, such as a Gaussian distribution with a certain mean and variance, which represents the range of energy consumption prediction under environmental uncertainty.

[0105] After obtaining the energy consumption probability distribution for each flight segment, the energy consumption distributions of each segment are accumulated along the entire route, either by time or distance. Since the prediction for each segment is a probability distribution, the accumulation calculation needs to consider its correlation, typically employing methods such as Monte Carlo simulation or convolution operations, to ultimately obtain the probability distribution of the total energy consumption required to complete the entire voyage. From this total energy consumption distribution, its expected value can be extracted as the expected cumulative energy consumption for the voyage, and the energy consumption fluctuation range at different confidence levels can be calculated.

[0106] This embodiment combines a mechanistic model reflecting general laws with a data-driven model characterizing specific ship historical behavior, and utilizes Bayesian optimization for parameter adaptation, significantly improving the accuracy and reliability of energy consumption prediction under different navigation conditions. The probability distribution form of the model output provides input for quantifying uncertainty in subsequent optimization, enabling optimization decisions to balance performance expectations and risks, which is an important foundation for achieving global collaborative optimization.

[0107] Please see Figure 4 In some embodiments, the output of the basic energy consumption mechanism model is used as prior knowledge constraint and weightedly fused with a data-driven black-box prediction model to construct a physical information fusion ship energy consumption prediction model, including:

[0108] S401. The predicted energy consumption value of the basic energy consumption mechanism model under a given input is transformed into a Gaussian distribution with the mean of the predicted energy consumption value and the variance of the preset mechanism uncertainty, which is used as the prior probability distribution.

[0109] S402. The prediction output of the data-driven black-box prediction model under the same input is characterized as a Gaussian process posterior distribution with the model prediction value as the mean and the model prediction variance as the uncertainty.

[0110] S403. Based on the Bayesian inference framework, the prior probability distribution is fused with the Gaussian process posterior distribution to calculate its posterior probability distribution.

[0111] S404, the mean function and covariance function of the posterior probability distribution jointly define the ship energy consumption prediction model of physical information fusion. The mean function is a weighted combination of the predicted values ​​of the basic energy consumption mechanism model and the predicted values ​​of the data-driven black box prediction model. The weight coefficients are determined by synchronous optimization during the model training phase using a Bayesian optimization algorithm.

[0112] In this embodiment, the deterministic energy consumption prediction value of the basic energy consumption mechanism model is assigned an uncertainty measure, namely, a preset mechanism uncertainty. This preset mechanism uncertainty reflects the inherent imprecision of the mechanism model due to simplification assumptions or parameter errors, and its value can be set based on historical model validation data or expert experience. By using this deterministic energy consumption prediction value as the mean and the square of the preset mechanism uncertainty as the variance, a Gaussian distribution is constructed. This distribution serves as the prior probability distribution in Bayesian inference, representing a preliminary estimate of the possible energy consumption values ​​based on physical knowledge before the observation data is available.

[0113] Data-driven black-box prediction models (Gaussian process regression models) naturally output predictions in the form of a probability distribution, namely the Gaussian process posterior distribution. This distribution uses the predicted values ​​learned by the model from the data as its mean, and quantifies the uncertainty of the prediction by the prediction variance calculated by the model itself. The magnitude of this variance depends on the similarity between the input points and the historical training data; the higher the similarity, the smaller the variance and the higher the prediction confidence.

[0114] When fusing data using the Bayesian inference framework, the aforementioned prior probability distribution and the Gaussian process posterior distribution are considered as two sources of information regarding the same true energy consumption value. The Bayesian formula is used to calculate the posterior probability distribution after considering both types of information simultaneously. The mean of this posterior distribution is the optimal estimate of energy consumption after fusing the mechanistic prior and data likelihood, while its covariance characterizes the uncertainty of the fused estimate.

[0115] The mean function and covariance function of the posterior probability distribution jointly define the ship energy consumption prediction model based on physical information fusion. Specifically, the mean function is a weighted combination of the predictions from the mechanistic model and the data-driven model. The crucial weight coefficients are not fixed but are determined synchronously during the overall model training phase by a Bayesian optimization algorithm, aiming to minimize the overall prediction error. This achieves overall adaptive tuning of the fusion weights and the model's internal parameters.

[0116] This embodiment achieves physical information fusion through Bayesian inference, probabilizing the deterministic output of the mechanistic model and treating it as prior knowledge. This probabilistic output is then rigorously mathematically fused with the probability distribution output by the data-driven model. This embodiment not only combines the advantages of both types of models but also dynamically determines the optimal fusion weights through Bayesian optimization. This allows the final model to adaptively balance physical laws and historical experience based on specific data conditions, resulting in more accurate and reliable energy consumption probability predictions. This lays a solid model foundation for subsequent risk-based optimization decisions.

[0117] Please see Figure 5 In some embodiments, with minimizing the expected cumulative energy consumption of a voyage as the core optimization objective, a three-variable collaborative optimization model is constructed by coupling voyage time constraints, main engine operating condition constraints, and ship stability safety boundaries, including:

[0118] S501. Based on the segment sequence and associated dynamic environmental features represented by the route feature encoding, define route path decision variables, whose solution space is the set of candidate routes corresponding to the encoding.

[0119] S502. Define the planned speed and planned trim value for each segment as continuous decision variables. Their feasible domain is jointly defined by the safe speed range of the ship's main engine, the design speed range, and the allowable trim range determined based on the ship's hydrostatic curve.

[0120] S503. Construct the objective function of the collaborative optimization model with the goal of minimizing the expected cumulative energy consumption per voyage;

[0121] S504. Construct a multimodal coupling constraint set, which includes:

[0122] The total voyage time calculated based on the distance of each segment and the corresponding planned speed must not exceed the time window constraint of the latest arrival time required in the voyage mission information.

[0123] The amplitude of the ship's motion response calculated from the navigation environment forecast information under the combination of planned speed and planned trim value for each segment shall not exceed the seakeeping and stability constraints of the preset safety threshold.

[0124] The variation of the planned trim value between adjacent segments is constrained by the dynamic feasibility of the maximum adjustment rate of the ballast water system.

[0125] S505, the decision variables of the route path, the planned speed and planned pitch value of each segment, the objective function, and the multimodal coupling constraint set are integrated and defined as a three-variable collaborative optimization model.

[0126] In this embodiment, the route path decision variable is a discrete choice variable, and its range of choices is directly defined by the set of candidate routes corresponding to the route feature encoding. Each candidate route corresponds to a unique path decision, which determines the sequence of segments the ship will traverse and the dynamic environmental characteristics associated with each segment.

[0127] Planned speed and planned trim are defined as continuous decision variables for each segment. The lower limit of the feasible region for planned speed is typically determined by the speed corresponding to the lowest stable rotational speed of the main engine, while the upper limit is determined by the speed corresponding to the maximum continuous power of the main engine or the ship's design speed. The feasible region for planned trim needs to be determined based on the ship's hydrostatic curve or stability manual. This curve describes the permissible safe range of the bow-stern draft difference (i.e., trim) under different draft conditions to ensure sufficient ship stability. The trim is calculated by subtracting the stern draft from the bow draft.

[0128] The objective function of the collaborative optimization model is directly set as minimizing the expected cumulative energy consumption of the voyage. This expected value is calculated by the physical information fusion ship energy consumption prediction model based on the input route path, planned speed and trim value of each segment, reflecting the expected energy consumption for completing the entire voyage under environmental uncertainty.

[0129] Multimodal coupling constraints are used to ensure the feasibility and safety of the optimization scheme. Time window constraints calculate the total sailing time by summing the quotient of the distance of each segment and the corresponding planned speed (i.e., sailing time), and require that it not exceed the latest arrival time required by the mission. Seakeeping and stability constraints require calculating the amplitude of the ship's roll, pitch, and other motions under specific speed, trim, and environmental conditions based on navigation environment forecast information (wind, waves) and using ship motion response theory or empirical formulas, and require that these amplitudes be below a preset threshold to ensure the safety of the ship and cargo. Dynamic feasibility constraints consider practical operability, limiting the variation in planned trim values ​​between adjacent segments to ensure that it does not exceed the maximum adjustment capacity of the ship's ballast water system within the corresponding time interval.

[0130] Integrating the decision variables, objective function, and multimodal coupling constraint set defined above constitutes a complete three-variable co-optimization model for flight path, speed, and pitch. Mathematically, this model is a complex optimization problem involving discrete and continuous variables, and nonlinear objectives and constraints.

[0131] This embodiment clarifies the specific forms and feasible regions of the three decision variables, sets minimizing the expected cumulative energy consumption of the voyage as the direct objective, and systematically integrates multiple practical constraints such as time, main engine safety, ship motion safety, and operational feasibility. These constraints are not set in isolation, but rather take into account the coupled effects between variables (such as the combined influence of speed and trim on motion response), thereby constructing an optimization problem framework that closely reflects the complexity and safety requirements of actual navigation, providing a precise mathematical model foundation for subsequent intelligent solutions.

[0132] In some embodiments, constructing a multimodal coupling constraint set includes:

[0133] Based on navigation environment forecast information and route feature coding, identify the different dominant navigation environment modes that may be encountered in the current voyage. The navigation environment modes include at least the open water mode and the restricted severe sea state mode.

[0134] For each identified navigation environment mode, configure constraint activation rules and constraint parameters corresponding to the dominant risk of that navigation environment mode;

[0135] In the open water mode, the time window constraint and the energy consumption fluctuation range constraint output by the ship energy consumption prediction model based on physical information fusion are activated and strengthened. At the same time, under the premise of meeting the inherent seakeeping and stability safety thresholds of the ship, a smaller safety margin is allowed to be applied in the optimization calculation.

[0136] Under the restricted severe sea state mode, the seakeeping and stability constraints are activated and strengthened, allowing for a larger safety margin in the optimization calculation, adjusting the safety threshold of the ship motion response amplitude to a more stringent value, and converting the time window constraint from a hard constraint to a soft constraint that allows violation under certain penalty costs.

[0137] During the optimization process, the multimodal coupling constraint set dynamically calls and applies the constraint configurations under the corresponding navigation environment mode based on the environmental characteristics of the flight segments traversed by the route path decision variables.

[0138] In this embodiment, the identification of navigation environment patterns is based on navigation environment forecast information and route feature encoding. The system analyzes the intensity, spatial distribution, and relative relationship between forecasted environmental parameters such as wind, waves, and currents and the route, and combines this with electronic chart information (such as water area openness and distance from shore) to classify different segments of the voyage into different dominant navigation environment patterns. For example, segments with relatively calm winds and waves and open waters are identified as open water patterns; segments with relatively strong winds and waves, or navigating in narrow, poorly sheltered waters, are identified as restricted, severe sea state patterns.

[0139] For each identified navigation environment pattern, its corresponding constraint activation rules and constraint parameters need to be pre-configured. The constraint activation rules determine which constraints will be activated or given higher priority when the ship navigates through a segment belonging to that pattern during the optimization process. The constraint parameters specifically define the strictness or calculation method of these constraints.

[0140] In open water mode, due to the relatively favorable environment, navigation safety risks mainly stem from the ability to arrive on time. Therefore, the optimization strategy in this mode focuses on both economy and punctuality. Specifically, time window constraints are activated and strengthened, requiring strict adherence to arrival times. Simultaneously, energy consumption fluctuation range constraints output by the ship energy consumption prediction model based on physical information fusion may also be activated to control the uncertainty risk of energy consumption. While meeting the ship's inherent seakeeping and stability safety thresholds (which are determined by the ship's own performance and remain consistent across different modes), a smaller safety margin can be applied in the optimization calculations, i.e., a more aggressive search can be conducted near the constraint boundaries to uncover greater energy-saving potential.

[0141] In restricted and severe sea state modes, navigation safety becomes the primary concern. Therefore, optimization strategies in this mode focus on risk control. Seakeeping and stability constraints are activated and strengthened, ensuring sufficient margin for ship motion response by applying greater safety margins. Although the ship's inherent safety threshold remains unchanged, the "stricter numerical values" are reflected in the use of more conservative allowable values ​​in optimization calculations (e.g., limiting the actual motion response amplitude to a lower percentage of the threshold). At the same time, the importance of time window constraints is relatively reduced, transforming from hard constraints that must be strictly followed into soft constraints that allow for violations but incur corresponding penalties (such as being factored into the objective function), reflecting the trade-off between safety and timeliness in severe environments.

[0142] During the optimization process, the multimodal coupling constraint set is not applied statically. The system dynamically calls and applies the constraint configurations for the corresponding modes based on the flight path determined by the route decision variables and the environmental modes of each segment of that path. This enables the optimization model to intelligently adapt to the risk characteristics of different stages of the journey, achieving differentiated and refined management of constraints.

[0143] This embodiment introduces a multimodal coupled constraint set, enabling the optimization model to possess environmentally adaptive constraint management capabilities. It dynamically adjusts the stringency and priority of constraints based on the risk level of the navigation environment, prioritizing economic optimization in open waters and ensuring safety in harsh sea conditions. This dynamic and differentiated constraint processing mechanism overcomes the shortcomings of traditional single-constraint models, which are either overly conservative or excessively risky in complex and variable navigation environments. It allows the collaborative optimization scheme to better balance multiple objectives such as energy efficiency, time, and safety on a global scale, improving the practicality and robustness of the scheme.

[0144] In some embodiments, a hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is used to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme, including:

[0145] A hierarchical solution architecture is constructed, in which the first-layer solver is based on a metaheuristic optimization algorithm, performs a global search in the route path decision space defined by the collaborative optimization model, generates and maintains a population containing multiple candidate routes;

[0146] For each candidate route in the population, the second-layer solver is invoked for processing, including:

[0147] The second-layer solver is based on a sequential decision-making algorithm. It models the sequence of segments along the candidate route as a sequential decision-making process, using the planned speed and planned pitch value of each segment as decision variables. The values ​​of the decision variables must meet the constraints defined in the collaborative optimization model. With the goal of minimizing the expected cumulative energy consumption of the voyages along the route, it performs segment-by-segment or global optimization to obtain the optimal speed and pitch value sequence that matches the candidate route.

[0148] The optimal speed and pitch sequence obtained by the second-layer solver for each candidate route, along with the corresponding expected cumulative energy consumption for each voyage, are fed back to the first-layer solver as the basis for evaluating the fitness value of the candidate route.

[0149] The first-layer solver selects, crosses over, and mutates candidate routes in the population based on fitness values ​​to generate a new generation of population, and iteratively executes the above process.

[0150] When the iteration process meets the preset convergence criterion, the calculation is terminated, and the candidate route with the best fitness value in the final population and its corresponding optimal speed and pitch value sequence are output together as the globally optimal collaborative optimization scheme.

[0151] In this embodiment, a hierarchical solution architecture is constructed to address the high complexity of the collaborative optimization model. The first-layer solver is responsible for global exploration in the discrete decision space of the flight path. It operates based on metaheuristic optimization algorithms (such as genetic algorithms), initializes a population consisting of multiple different candidate routes, and improves the quality of the population through iterative evolution.

[0152] For each candidate route in the population, its merits need to be evaluated, which depends on matching it with the optimal speed and trim sequence. Therefore, a second-layer solver is invoked for processing. The second-layer solver, based on a sequential decision algorithm (such as dynamic programming), models the navigation process along the fixed route as a multi-stage sequential decision problem. Each segment is considered a stage, and the decision variables for that stage are the planned speed and planned trim values, which must satisfy all relevant constraints in the model. The optimization objective is to minimize the expected cumulative energy consumption for each voyage along this fixed route. Using a sequential decision algorithm, the optimal speed and trim values ​​for each segment of the route, as well as the corresponding minimum expected cumulative energy consumption for each voyage, can be efficiently solved.

[0153] The optimal speed and pitch sequence output by the second-layer solver, along with the corresponding expected cumulative energy consumption for each voyage, is fed back to the first-layer solver. This expected cumulative energy consumption is used as the primary basis for evaluating the fitness of the candidate route; the smaller the expected value, the higher the fitness.

[0154] The first-layer solver performs selection, crossover, and mutation operations on the current population based on the fitness values ​​of each candidate route. The selection operation retains routes with high fitness; the crossover operation exchanges partial path information between two routes to generate a new route; and the mutation operation randomly perturbs a local path of a route. These operations generate a new generation of population, thereby exploring the possibility of new routes.

[0155] The two-layer solution process described above is performed iteratively. In each generation, the first-layer solver generates a new population of candidate routes, while the second-layer solver calculates the optimal speed and trim sequence for each new route and computes its fitness. The computation terminates when the preset maximum number of generations is reached, or when the optimal fitness value in the population no longer significantly improves over multiple consecutive generations, satisfying the convergence criterion. Finally, the candidate routes with the best fitness from each generation, along with their corresponding optimal speed and trim value sequences, are output as the globally optimal collaborative optimization scheme.

[0156] This embodiment decomposes the complex global optimization problem through a layered architecture: the upper-layer metaheuristic algorithm extensively searches the route space, while the lower-layer sequential decision algorithm accurately solves for speed and pitch for each fixed route. The two layers are tightly coupled through fitness evaluation. This embodiment combines the global exploration capability of the metaheuristic algorithm with the precise optimization capability of the sequential decision algorithm under fixed paths, effectively overcoming the difficulty of traditional single algorithms in simultaneously handling discrete path selection and continuous speed and attitude optimization, thus reliably and efficiently solving for high-quality global collaborative optimization solutions.

[0157] In some embodiments, the second-layer solver, based on a sequential decision algorithm, models the sequence of segments along the candidate route as a sequential decision process. Using the planned speed and planned trim value for each segment as decision variables, and aiming to minimize the expected cumulative energy consumption of voyages along the route, it performs segment-by-segment or global optimization to obtain the optimal speed and trim value sequence matching the candidate route, including:

[0158] The candidate route segment sequence is constructed as a multi-stage decision graph, where each node corresponds to the initial state of a segment, and the directed edges between nodes represent the execution of a specific combination of speed and pitch values ​​for that segment.

[0159] The ship energy consumption prediction model based on physical information fusion calculates the expected energy consumption of a given route under the environmental characteristics and ship status of each possible decision combination, which serves as the cost corresponding to that decision edge.

[0160] From the starting node to the ending node of the multi-stage decision graph, the dynamic programming algorithm is applied to search for the path that minimizes the sum of costs of each flight segment, i.e., the expected value of the cumulative energy consumption of the flight.

[0161] The optimal path obtained through the search is formed by combining the speed and trim values ​​of each decision edge, arranged in the order of the flight segments, to create a sequence of optimal speed and trim values ​​that matches the candidate route.

[0162] In this embodiment, the construction of a multi-stage decision graph is the foundation of sequential decision-making. The sequence of candidate route segments is viewed as multiple interconnected stages, with the starting point of each segment defined as a decision node. Directed edges between nodes represent the selection of a specific speed and trim value combination when entering the next segment from the end state (node) of the previous segment. By discretizing the continuous decision space (speed, trim) into a finite number of feasible decision combinations, multiple decision edges pointing to the next node can be generated for each node, thus forming a graph describing all possible decision paths.

[0163] Assigning a cost to each decision edge is key to optimization, and the calculation of the cost relies on a ship energy consumption prediction model based on physical information fusion. For a decision edge, the corresponding segment environmental characteristics (derived from route feature encoding) and the ship's initial state (e.g., draft) are known, and the decision variables (speed and trim) are determined. This information is input into the energy consumption prediction model, and the model outputs the expected energy consumption for executing this decision on that segment. This expected value is defined as the cost of that decision edge, quantifying the energy consumption expected by choosing this decision combination.

[0164] From the starting node (flight start point) to the ending node (flight end point) of the multi-stage decision graph, there are numerous paths connected by edges from different decisions. Dynamic programming can efficiently search for the path with the minimum total cost. Dynamic programming recursively calculates the minimum cumulative cost to reach the destination from each node, starting from the ending node, and records the corresponding optimal subsequent decision. By backtracking forward, the complete path from the starting point to the destination that minimizes the sum of the costs of each segment can be found.

[0165] This optimal path consists of a series of sequentially connected decision edges. Each edge corresponds to a flight segment and a specific combination of speed and trim values ​​used in that segment. Extracting and arranging these decision combinations in the order of the flight segments forms the optimal speed and trim value sequence matching the candidate route. This sequence ensures that, from a global perspective, the expected cumulative energy consumption of the flight is minimized when traveling along this fixed route.

[0166] In this embodiment, the second-layer solver uses dynamic programming to solve the speed and trim sequence under a fixed route. By constructing a multi-stage decision graph, the continuous optimization problem is transformed into a graph search problem. The dynamic programming algorithm is used to efficiently find the global optimal solution. It can accurately handle the sequential dependencies and state transitions between routes, ensuring that the evaluation (fitness value) of each route provided to the upper layer is based on its current optimal speed and trim strategy, thereby guaranteeing the accuracy and efficiency of the hierarchical collaborative solution.

[0167] In some embodiments, during the navigation of a ship executing a collaborative optimization scheme, based on real-time collected actual navigation environment data and ship energy consumption data, the ship energy consumption prediction model is updated online with adaptive parameters, and dynamic replanning of the collaborative optimization model is triggered to generate rolling optimization instructions to adjust speed and trim, including:

[0168] During the voyage, the ship continuously collects measured data on the navigation environment, actual fuel consumption data of the main engine, and actual motion status data of the ship to form an online verification dataset;

[0169] The prediction output of the ship energy consumption prediction model corresponding to the current state is compared with the online validation dataset and physical information, and the prediction residual sequence of the model is calculated.

[0170] When the statistical characteristics of the model's predicted residual sequence exceed the preset model drift threshold, the online learning process is initiated. Using the latest online validation dataset within the sliding time window, some or all of the adjustable parameters of the physical information fusion ship energy consumption prediction model are incrementally updated to achieve adaptive correction of the model parameters.

[0171] Based on the updated physical information fusion ship energy consumption prediction model, and combined with the latest navigation environment forecast information and real-time ship status, the current ship position is used as the new starting point and the original destination port is used as the end point. The collaborative optimization model and solution process are re-executed to carry out local or global replanning.

[0172] The replanning process generates an updated collaborative optimization scheme from the current moment to the end of the voyage. The recommended speed and recommended trim values ​​for the upcoming next decision cycle are extracted from the updated collaborative optimization scheme and issued to the ship control system as rolling optimization instructions for execution.

[0173] In this embodiment, the online verification dataset consists of measured data continuously collected during navigation. The measured navigation environment data includes on-site observations of wind speed and direction, wave height and direction, and current speed and direction; actual main engine fuel consumption data is read from the ship's fuel flow meter; and actual ship motion data may include actual speed, trim, roll angle, etc. These data are aligned by timestamps to form a dataset used for model verification and updating.

[0174] The model prediction residual sequence is obtained through comparative calculation. The ship's current state (e.g., draft, load) and actual environmental data are input into the physical information fusion ship energy consumption prediction model to obtain the predicted energy consumption per unit time under that state. This predicted value is compared with the actual main engine fuel consumption data (converted to unit time) measured during the same period; the difference is the prediction residual for that moment. The residuals over a continuous period constitute a sequence.

[0175] The model drift threshold is a preset statistical criterion used to determine whether a model has experienced significant performance degradation. For example, it can be set that if a certain combination of the mean and variance of the residual sequence exceeds a certain threshold for multiple consecutive periods, it is considered to have exceeded the drift threshold. Once exceeded, the online learning process is triggered.

[0176] The online learning process utilizes the latest online validation dataset accumulated within a sliding time window (e.g., the last few hours) to incrementally update the adjustable parameters of the physical information fusion ship energy consumption prediction model. The update process can employ online learning algorithms such as recursive least squares and stochastic gradient descent, aiming to minimize the prediction error within the window, fine-tuning the model parameters to make the model predictions more closely match the current actual navigation conditions and ship status, thus achieving adaptive correction.

[0177] Based on the updated energy consumption prediction model and the latest navigation environment forecast information (covering the future time period from the current position to the destination port) and the latest real-time status of the vessel, the entire process from route feature encoding to collaborative optimization model solving is re-executed, using the current position as the new voyage starting point. Due to changes in the starting point and the latest environmental information, this planning may result in different routes, speeds, and trim strategies than the original plan, thus achieving dynamic replanning.

[0178] The updated collaborative optimization scheme obtained from the replanning is a complete new scheme covering the period from the current moment to the end of the voyage. From this new scheme, the recommended speed and recommended trim values ​​for the next decision cycle (the next 4 hours) are extracted and used as rolling optimization instructions. These instructions are issued to the ship's automatic control system or displayed to the navigator to adjust the ship's actual navigation operations, thus forming a closed-loop control of "monitoring-updating-replanning-execution".

[0179] This embodiment realizes a closed-loop and dynamic evolution of the energy efficiency optimization scheme. Through online monitoring and adaptive model updates, the system can sense and adapt to model prediction deviations and environmental changes. Through periodic or triggered dynamic replanning, the optimization scheme can be continuously refreshed based on the latest information, overcoming the limitations of static offline schemes. This ensures that the ship can continuously follow a near-globally optimal energy efficiency trajectory throughout the voyage, significantly improving the robustness and long-term effectiveness of the optimization scheme in real dynamic environments.

[0180] In a second aspect, this embodiment also provides an energy efficiency improvement system based on route-speed-margin collaborative optimization, applicable to the method described in the first aspect. The system includes a data sensing and acquisition module, a route feature processing module, a ship energy consumption prediction module, a collaborative optimization modeling and solving module, an online learning and dynamic replanning module, and an instruction execution and interface module. The data sensing and acquisition module is configured to acquire the target ship's voyage mission information, real-time ship status information, navigation environment forecast information, and electronic chart information, and to collect actual navigation environment data and ship energy consumption data in real time during the voyage. The route feature processing module is connected to the data sensing and acquisition module and is configured to construct a route feature code containing spatiotemporal correlation features based on navigation environment forecast information and electronic chart information. The ship energy consumption prediction module is connected to the route feature processing module and the data sensing and acquisition module, and internally deploys a ship energy consumption prediction model based on physical information fusion. It is configured to receive the route feature code, real-time ship status information, and preset speed and margin decision variables, and generate segment energy consumption probability distributions and voyage cumulative energy consumption expectations corresponding to different decision schemes. The modeling and solving module is connected to the ship energy consumption prediction module. It is configured to minimize the expected cumulative energy consumption of a voyage as the core optimization objective, coupling voyage time constraints, main engine operating condition constraints, and ship stability safety boundaries to construct a three-variable collaborative optimization model for the route path, planned speed for each segment, and planned trim value. A hybrid intelligent solving algorithm combining metaheuristics and sequential decision-making is used to solve the model, outputting a globally optimal collaborative optimization scheme containing a recommended route path sequence, recommended speed for each segment, and recommended trim value. The online learning and dynamic replanning module is connected to the data sensing and acquisition module, the ship energy consumption prediction module, and the collaborative optimization modeling and solving module. It is configured to perform online adaptive parameter updates to the ship energy consumption prediction model based on real-time collected actual navigation environment data and ship energy consumption data during voyage, and trigger dynamic replanning in the collaborative optimization modeling and solving module to generate rolling optimization instructions. The instruction execution and interface module is connected to the collaborative optimization modeling and solving module and the online learning and dynamic replanning module. It is configured to send the globally optimal collaborative optimization scheme or rolling optimization instructions to the ship control system or display terminal.

[0181] This system employs a modular design to clearly divide and integrate the complex energy efficiency optimization process. Each module collaborates closely based on data flow and logical relationships, collectively achieving a complete technical closed loop from multi-source information perception, intelligent predictive modeling, global collaborative optimization to online adaptive execution. This embodiment provides a stable, efficient, and engineerable implementation platform for the aforementioned methods, enabling the intelligent and dynamic improvement of ship energy efficiency to operate reliably in actual navigation.

[0182] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: By constructing a three-variable collaborative optimization model for the route path, planned speed, and planned trim value, traditionally fragmented decision variables are placed under a unified framework for global joint optimization, overcoming the local optimum problem caused by independent or sequential optimization of dependent variables in existing methods. This allows for the exploration of deeper energy-saving potential at the entire voyage level. By constructing a ship energy consumption prediction model that integrates physical mechanisms and data-driven approaches, and outputting the probability distribution of voyage segment energy consumption, a quantitative characterization of the uncertainty of the navigation environment is achieved, providing a more reliable risk perception basis for optimization decisions. The invention also employs metaheuristics and sequential decision-making... The combined hybrid intelligent solution algorithm effectively solves the problem of solving high-dimensional, nonlinear, and multi-constraint collaborative optimization models, ensuring the reliable output of high-quality global optimal solutions. Furthermore, by introducing an online parameter adaptive update and dynamic replanning mechanism based on real-time data, the static optimization scheme is equipped with the ability to cope with changes in the actual dynamic navigation environment, forming a closed-loop control of "prediction-optimization-execution-feedback", which significantly improves the adaptability, robustness and long-term effectiveness of the optimization scheme in actual navigation. Finally, under the premise of strictly meeting multiple constraints such as navigation time, main engine operating conditions and ship stability and safety, a systematic and intelligent improvement of the ship's energy efficiency for the entire voyage is achieved.

[0183] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0184] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0185] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An energy efficiency improvement method based on route-speed-curve coordinated optimization, characterized in that, include: Obtain voyage mission information, real-time ship status information, and navigation environment forecast information for the target vessel; Based on the navigation environment forecast information and electronic nautical chart information, a route feature code containing spatiotemporal correlation features is constructed; The route feature encoding, real-time ship status information, and preset speed and trim decision variables are input into a physical information fusion ship energy consumption prediction model to generate segment energy consumption probability distributions and voyage cumulative energy consumption expectations corresponding to different decision schemes, including: Based on the energy transfer mechanism between the ship's main engine, propeller and hull, a basic energy consumption mechanism model of the ship is constructed with speed, trim and navigation environment as input variables. By using historical ship energy consumption data and corresponding navigation status datasets, a Gaussian process regression model is trained to obtain a data-driven ship energy consumption black box prediction model. Using the output of the basic energy consumption mechanism model as prior knowledge constraints, a weighted fusion is performed with the data-driven black-box prediction model to construct a physical information fusion ship energy consumption prediction model. The model parameters are adaptively tuned using a Bayesian optimization algorithm. The environmental information of the route segment represented by the route feature encoding, the draft and load information of the ship in real time status information, together with the preset combination of speed and trim decision variables, are input into the trained physical information fusion ship energy consumption prediction model. The ship energy consumption prediction model outputs the predicted probability distribution of the ship's energy consumption per unit time in a single voyage under the current decision combination; Based on the probability distribution of energy consumption per segment, rolling cumulative calculation is performed on the segment sequence defined by the route feature encoding to obtain the expected cumulative energy consumption of the entire voyage at different confidence levels and its fluctuation range. With minimizing the expected cumulative energy consumption of the voyage as the core optimization objective, and coupling sailing time constraints, main engine operating condition constraints, and ship stability safety boundaries, a three-variable collaborative optimization model is constructed, consisting of the route path, planned speed for each segment, and planned trim value. Based on the segment sequence and its associated dynamic environmental features represented by the route feature encoding, route path decision variables are defined, and their solution space is the set of candidate routes corresponding to the encoding. The planned speed and planned trim value for each segment are defined as continuous decision variables, and their feasible domain is jointly defined by the safe speed range of the ship's main engine, the design speed range, and the allowable trim range determined based on the ship's hydrostatic curve. To minimize the expected cumulative energy consumption per voyage, an objective function for the collaborative optimization model is constructed. Construct a multimodal coupling constraint set, which includes: The total voyage time calculated based on the distance of each segment and the corresponding planned speed must not exceed the time window constraint of the latest arrival time required in the voyage mission information. The amplitude of the ship's motion response calculated from the navigation environment forecast information under the combination of planned speed and planned trim value for each segment shall not exceed the seakeeping and stability constraints of the preset safety threshold. The variation of the planned trim value between adjacent segments is constrained by the dynamic feasibility of the maximum adjustment rate of the ballast water system. The decision variables, objective function, and multimodal coupling constraint set of the route path, planned speed and planned pitch value for each segment are integrated and defined as a three-variable collaborative optimization model. A hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is used to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme, which includes a recommended route sequence, recommended speed for each segment, and recommended pitch value. During the ship's navigation in accordance with the aforementioned collaborative optimization scheme, the ship's energy consumption prediction model is updated online based on real-time collected actual navigation environment data and ship energy consumption data. This triggers dynamic replanning of the collaborative optimization model and generates rolling optimization commands to adjust speed and trim.

2. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 1, characterized in that, Based on the aforementioned navigation environment forecast information and electronic nautical chart information, a route feature code incorporating spatiotemporal correlation features is constructed, including: Based on the port of origin and port of destination in the voyage mission information, the initial navigable area is determined on the electronic nautical chart, and the initial navigable area is dynamically corrected by combining the wind, wave and current data in the navigation environment forecast information to generate a set of candidate routes containing environmental risk information. For each candidate route in the candidate route set, waypoints are extracted along the path corresponding to the candidate route at preset spatial intervals. Based on the electronic chart information and the navigation environment forecast information, static geographical features and dynamic environmental features are associated with each waypoint. The static geographical features include water depth, distance from the shore and turning angle of the route. The dynamic environmental features include predicted wind speed, predicted wave height, predicted current speed and its direction relative to the ship's heading. Clustering analysis algorithms are used to perform pattern recognition on the waypoint feature sequences of the candidate routes to determine key turning points and typical segment divisions on the routes; Based on the key turning points and typical route segment division results, the feature sequences of each candidate route are restructured and vectorized to generate the route feature code that can characterize the global spatiotemporal attributes and local environmental changes of the route.

3. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 1, characterized in that, Using the output of the basic energy consumption mechanism model as prior knowledge constraints, a weighted fusion is performed with the data-driven black-box prediction model to construct the physical information fusion ship energy consumption prediction model, including: The predicted energy consumption value of the basic energy consumption mechanism model under a given input is transformed into a Gaussian distribution with the mean of the predicted energy consumption value and the variance of the preset mechanism uncertainty, which is used as the prior probability distribution. The prediction output of the data-driven black-box prediction model under the same input is characterized as a Gaussian process posterior distribution with the model prediction value as the mean and the model prediction variance as the uncertainty. Based on the Bayesian inference framework, the prior probability distribution is fused with the posterior distribution of the Gaussian process to calculate its posterior probability distribution; The mean function and covariance function of the posterior probability distribution jointly define the ship energy consumption prediction model fused with physical information. The mean function is a weighted combination of the predicted values ​​of the basic energy consumption mechanism model and the predicted values ​​of the data-driven black-box prediction model. The weight coefficients are determined by the Bayesian optimization algorithm during the model training phase.

4. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 1, characterized in that, Construct a multimodal coupling constraint set, including: Based on the navigation environment forecast information and the route feature code, identify the different dominant navigation environment modes that may be encountered in the current voyage. The navigation environment modes include at least the open water mode and the restricted severe sea state mode. For each identified navigation environment mode, configure constraint activation rules and constraint parameters corresponding to the dominant risk of that navigation environment mode; In the open water mode, the time window constraint and the energy consumption fluctuation range constraint output by the ship energy consumption prediction model based on the fusion of the physical information are activated and strengthened. At the same time, under the premise of meeting the inherent seakeeping and stability safety thresholds of the ship, a smaller safety margin is allowed to be applied in the optimization calculation. Under the restricted severe sea state mode, the seakeeping and stability constraints are activated and strengthened, allowing for a larger safety margin in the optimization calculation, adjusting the safety threshold of the ship motion response amplitude to a more stringent value, and converting the time window constraint from a hard constraint to a soft constraint that allows violation under certain penalty costs. During the optimization process, the multimodal coupling constraint set dynamically calls and applies constraint configurations under the corresponding navigation environment mode based on the environmental characteristics of the flight segments traversed by the route path decision variables.

5. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 1, characterized in that, A hybrid intelligent solution algorithm combining metaheuristics and sequential decision-making is used to solve the collaborative optimization model and output the globally optimal collaborative optimization scheme, including: A hierarchical solution architecture is constructed, in which the first-layer solver is based on a metaheuristic optimization algorithm, performs a global search in the route path decision space defined by the collaborative optimization model, generates and maintains a population containing multiple candidate routes; For each candidate route in the population, the second-layer solver is invoked for processing, including: The second-layer solver is based on a sequential decision algorithm. It models the sequence of segments along the candidate route as a sequential decision process, using the planned speed and planned pitch value of each segment as decision variables. The values ​​of the decision variables must meet the constraints defined in the collaborative optimization model. With the goal of minimizing the expected cumulative energy consumption of the voyage along the candidate route, it performs segment-by-segment or global optimization to obtain the optimal speed and pitch value sequence that matches the candidate route. The optimal speed and pitch sequence obtained by the second-layer solver for each candidate route, along with the corresponding expected cumulative energy consumption for each voyage, are fed back to the first-layer solver as a basis for evaluating the fitness value of the candidate route. The first-layer solver performs selection, crossover, and mutation operations on candidate routes in the population based on the fitness value to generate a new generation of population, and iteratively executes the above process. When the iteration process meets the preset convergence criterion, the calculation is terminated, and the candidate route with the best fitness value in the final population and its corresponding optimal speed and pitch value sequence are output together as the globally optimal collaborative optimization scheme.

6. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 5, characterized in that, The second-layer solver, based on a sequential decision-making algorithm, models the sequence of segments along the candidate route as a sequential decision-making process. Using the planned speed and planned trim value for each segment as decision variables, and aiming to minimize the expected cumulative energy consumption of the voyage along the candidate route, it performs segment-by-segment or global optimization to obtain the optimal speed and trim value sequence matching the candidate route, including: The candidate route segment sequence is constructed as a multi-stage decision graph, where each node corresponds to the initial state of a segment, and the directed edges between nodes represent the execution of a specific speed and pitch decision combination in that segment. Based on the ship energy consumption prediction model fused with the physical information, the expected energy consumption of the ship segment caused by each possible decision combination is calculated under the given environmental characteristics and ship status of the segment, and is used as the cost corresponding to the decision edge. From the starting node to the ending node of the multi-stage decision graph, a dynamic programming algorithm is applied to search for the path that minimizes the sum of costs for each flight segment, i.e., the expected value of the cumulative energy consumption of the flight. The speed and trim value decision combinations corresponding to each decision edge traversed by the optimal path obtained by the search are arranged in the order of the flight segments to form the optimal speed and trim value sequence that matches the candidate route.

7. The energy efficiency improvement method based on route-speed-curve coordinated optimization according to claim 1, characterized in that, During the ship's navigation using the aforementioned collaborative optimization scheme, based on real-time collected actual navigation environment data and ship energy consumption data, the ship energy consumption prediction model undergoes online parameter adaptive updates, triggering dynamic replanning of the collaborative optimization model and generating rolling optimization commands to adjust speed and trim, including: During the voyage, the ship continuously collects measured data on the navigation environment, actual fuel consumption data of the main engine, and actual motion status data of the ship to form an online verification dataset; The online verification dataset is compared with the prediction output of the ship energy consumption prediction model corresponding to the current state, which is fused with the physical information, and the model prediction residual sequence is calculated. When the statistical characteristics of the model's predicted residual sequence exceed the preset model drift threshold, the online learning process is initiated. Using the latest online validation dataset within the sliding time window, some or all of the adjustable parameters of the physical information fusion ship energy consumption prediction model are incrementally updated to achieve adaptive correction of the model parameters. Based on the updated physical information fusion ship energy consumption prediction model, and combined with the latest navigation environment forecast information and real-time ship status, the current ship position is used as the new starting point and the original destination port is used as the endpoint. The collaborative optimization model and solution process are re-executed to carry out local or global replanning. The replanning process generates an updated collaborative optimization scheme from the current moment to the end of the voyage. The recommended speed and recommended trim value for the upcoming next decision cycle are extracted from the updated collaborative optimization scheme and issued to the ship control system as the rolling optimization command for execution.

8. An energy efficiency improvement system based on route-speed-curve coordinated optimization, characterized in that, The system applicable to the method of any one of claims 1 to 7, the system comprising: The data sensing and acquisition module is configured to acquire voyage mission information, real-time ship status information, navigation environment forecast information and electronic nautical chart information of the target ship, and to collect actual navigation environment data and ship energy consumption data in real time during the voyage. The route feature processing module is connected to the data sensing and acquisition module and is configured to construct route feature codes containing spatiotemporal correlation features based on the navigation environment forecast information and electronic nautical chart information. The ship energy consumption prediction module is connected to the route feature processing module and the data sensing and acquisition module. It has a physical information fusion ship energy consumption prediction model deployed inside. It is configured to receive the route feature code, real-time ship status information and preset speed and trim decision variables, and generate the segment energy consumption probability distribution and voyage cumulative energy consumption expectation corresponding to different decision schemes. The collaborative optimization modeling and solving module is connected to the ship energy consumption prediction module. It is configured to minimize the expected cumulative energy consumption of the voyage as the core optimization objective, coupled with the voyage time constraint, main engine operating condition constraint and ship stability safety boundary, to construct a three-variable collaborative optimization model of route path, planned speed and planned trim value for each segment, and to solve the model using a hybrid intelligent solving algorithm that combines metaheuristics and sequential decision-making, outputting a globally optimal collaborative optimization scheme that includes a recommended route path sequence, recommended speed and recommended trim value for each segment. The online learning and dynamic replanning module is connected to the data sensing and acquisition module, the ship energy consumption prediction module, and the collaborative optimization modeling and solving module. It is configured to perform online parameter adaptive updates on the ship energy consumption prediction model fused with physical information based on real-time collected actual navigation environment data and ship energy consumption data during navigation, and trigger the collaborative optimization modeling and solving module to perform dynamic replanning to generate rolling optimization instructions. The instruction execution and interface module is connected to the collaborative optimization modeling and solving module and the online learning and dynamic replanning module, and is configured to send the global optimal collaborative optimization scheme or the rolling optimization instruction to the ship control system or display terminal.

Images