A ship power scheduling method and system based on multi-sensor data fusion

Abstract

The application discloses a kind of ship power scheduling method and system based on multi-sensor data fusion, and the running state data of ship is collected in real time by multiple sensors, and environmental variable data is collected by environmental monitoring unit;Multi-source data is timestamped and aligned and space coordinate is registered, forming multi-source heterogeneous data fusion matrix;The matrix is input into the power demand prediction model trained in advance, and the predicted propulsion power demand and power margin interval are output;Combined with the real-time health status parameters of the ship power system, the optimal power distribution scheme is calculated by multi-objective optimization algorithm;The scheme is issued to each power execution unit, and closed-loop scheduling control is realized.The application improves the economy, safety and environmental protection of ship power scheduling through multi-source data deep fusion, digital twin health assessment and multi-time scale rolling optimization, and is suitable for power scheduling management of various ships.

Description

Technical Field

[0001] This invention relates to the field of ship navigation technology, and more specifically, to a ship power scheduling method and system based on multi-sensor data fusion. Background Technology

[0002] Traditional ship power scheduling methods often rely on the experience and judgment of the captain or engineers, or employ control strategies based on fixed rules, such as roughly allocating main engine load according to a set speed. These methods struggle to respond in real-time to complex and changing sea conditions. Furthermore, the degradation of the ship's structure and power system's own health leads to inefficient matching between power output and navigation needs, often resulting in excessive fuel consumption, carbon emissions exceeding standards, and even accelerated hull fatigue damage or power unit overload failures in severe sea conditions due to improper power scheduling. Simultaneously, existing ship sensor data is mostly collected and analyzed in isolation, lacking effective time synchronization, spatial registration, and multi-source heterogeneous data fusion mechanisms. This hinders the formation of a unified and accurate perception of the ship's overall navigation status, thus limiting the level of intelligent power scheduling.

[0003] In recent years, some studies have attempted to apply machine learning models to predict ship speed or power. However, most of these studies rely solely on historical navigation data from a single source or simplified hydrodynamic empirical formulas, failing to deeply couple real-time environmental variables with the ship's real-time operating status. This results in significant deviations in predictions under rapidly changing marine environments. Furthermore, existing power scheduling optimization methods typically prioritize minimizing fuel consumption, neglecting multi-dimensional constraints such as hull structural safety, remaining power system lifespan, and emissions compliance. They lack a multi-objective collaborative decision-making mechanism capable of dynamically adjusting optimization preferences and adapting to different navigation scenarios. When ships encounter abnormal conditions such as localized stress exceeding limits, main engine bearing overheating, or sudden typhoon warnings, traditional systems often only trigger simple audible and visual alarms, failing to automatically generate graded response strategies based on the anomaly level and adjust power distribution accordingly, posing significant navigational safety hazards.

[0004] On the other hand, in existing technologies, route planning and power scheduling are typically separated into two independent processes: electronic chart systems provide static route recommendations, while the power system passively follows the set speed, failing to utilize dynamic environmental information such as real-time flow fields, underwater topography, and short-term weather conditions to achieve coordinated optimization of routes and power. Simultaneously, hull structural health assessments mostly rely on offline detection or periodic maintenance, unable to feed back the cumulative degree of fatigue damage to power scheduling decisions online. This leads to the frequent use of high-stress routes, accelerating the fatigue life loss of critical hull nodes. Therefore, there is an urgent need to establish a ship power intelligent scheduling method based on multi-sensor data fusion, capable of real-time perception of ship status and environmental dynamics, accurate prediction of power demand, multi-objective optimization of power allocation, and possessing route self-adaptation and anomaly graded response capabilities, to achieve efficient, safe, and environmentally friendly ship navigation. Summary of the Invention

[0005] One objective of this invention is to provide a ship power scheduling method and system based on multi-sensor data fusion, so as to realize real-time perception of ship status and environmental dynamics, accurate prediction of power demand, multi-objective optimization of power allocation, and have the ability to adapt to routes and respond to anomalies in a graded manner.

[0006] According to a first aspect of the present invention, a ship power scheduling method based on multi-sensor data fusion is provided, comprising the following steps: The ship's operating status data is collected in real time by a variety of sensors deployed in different parts of the ship. The various sensors include at least: strain sensors and acceleration sensors installed on the hull structure, speed sensors and fuel flow sensors installed on the power system, and pressure sensors installed near the waterline of the hull. The environmental monitoring unit collects environmental variable data of the waters where the ship is located in real time. The environmental variable data includes at least: wave height, wind speed, wind direction, water flow speed and water flow direction. The ship operation status data and the environmental variable data are time-stamped and spatially registered to form a multi-source heterogeneous data fusion matrix. The multi-source heterogeneous data fusion matrix is ​​input into a pre-trained ship power demand prediction model. The model is constructed based on ship dynamics simulation and historical navigation data, and outputs the predicted propulsion power demand of the ship under the current navigation conditions and the corresponding power margin range. Combining the predicted propulsion power demand, power margin range, and real-time health status parameters of the ship's current power system, the optimal power allocation scheme is calculated using a multi-objective optimization algorithm. The optimal power allocation scheme includes real-time allocation instructions for the main engine output power, auxiliary power unit output power, and energy storage system charging and discharging power. The optimal power allocation scheme is distributed to each power execution unit through the ship's power control network to achieve closed-loop scheduling and control of the ship's power system.

[0007] Optionally, the process of constructing the multi-source heterogeneous data fusion matrix includes: A multimodal data feature cross-validation unit is used to calculate the credibility of multi-sensor data of the same physical quantity through a Bayesian network. When the deviation exceeds the threshold, conflicting data is marked and redundant data is enabled. By utilizing a self-attention mechanism and a dynamic weight allocation layer, the contribution of different sensor data to the overall ship status is learned, and a fused feature vector with dynamic weights is output. The isolated forest algorithm identifies and isolates abnormal data that deviates from the mean of 3σ for three consecutive sampling periods, triggering a sensor fault alarm. Here, the mean is the normal operating baseline value of the sensor, σ is the normal fluctuation range, and 3σ is three times the normal fluctuation range. If the data exceeds this warning line three times consecutively, the sensor is determined to need to alarm.

[0008] Extended Kalman filtering was used to remove noise and interpolate missing values ​​in the remaining data, and coordinate transformation compensation was performed in conjunction with the six-degree-of-freedom motion model of the ship.

[0009] Optionally, the ship power demand prediction model is a fusion model of multi-ship type transfer learning and graph convolutional network: Using the LSTM / GRU model of the mother ship as the base model, the parameters are fine-tuned using a small amount of navigation data of the target ship. The sensor data is linked to the physical structure to construct a graph structure, and the spatial dependencies are learned using GCN to output feature vectors with topological weights. The model input features include current / historical speed, draft, attitude, wave height spectrum, wind speed vector, water flow gradient, and sensor topology features. The output features include short-term propulsion power demand sequence, medium-term power margin trend, long-term energy consumption prediction, and their probability distribution.

[0010] Optionally, the multi-objective optimization algorithm includes: Based on a deep reinforcement learning agent, the target weights are adjusted and optimized in real time with the scene state as input. Multi-timescale rolling optimization is adopted, with short-term optimization of real-time power allocation, medium-term optimization of route fine-tuning, and long-term optimization of energy storage charging and discharging plans, which are connected by rolling time-domain control. The optimization objective functions include: minimizing combined fuel / new energy consumption, minimizing carbon emissions, maximizing the remaining lifespan of the power system, maximizing navigation safety margin, and maximizing the utilization rate of new energy sources. The Pareto front is solved using the improved NSGA-III algorithm.

[0011] Optionally, the real-time health status parameters are obtained through dynamic evaluation driven by a digital twin, specifically including: Hull structure assessment: Construct a digital twin of the hull and propulsion system, integrate real-time strain / vibration data, and use a physics engine to dynamically simulate stress distribution under power scheduling; calculate the fatigue damage rate of key nodes online based on a nonlinear fatigue damage accumulation model. Power system evaluation: Using vibration / temperature data and an improved PHM model, the remaining life of the main engine / auxiliary power unit is evaluated. When the remaining life is <10%, the output power derating factor is determined and linked with the scheduling strategy to limit high-load operation.

[0012] Optionally, the method further includes a route adaptive optimization step, specifically including: By fusing real-time ocean current field, underwater topography and local severe convection early warning by multiple sources of sensors such as shipborne phased array current measurement radar, lidar and short-term weather receiving terminal, and superimposing them on electronic nautical charts, a dynamic environmental digital map is constructed, marking high energy consumption areas, high risk areas and low stress recommended routes based on hydrodynamic-stress coupling model; A four-dimensional optimization objective is established, which includes minimizing energy consumption and time cost, maximizing hull structural safety, ensuring emission compliance, and adhering to navigation rules. An improved multi-objective particle swarm optimization (MOPSO) algorithm with a dynamic adjustment mechanism for objective preferences is adopted to output the Pareto optimal speed-route combination scheme. The reinforcement learning (DQN) agent compares the deviation between the actual speed and route and the predicted target in real time and makes online corrections. The dynamic environment digital map is updated every 30 minutes and a new round of optimization iteration is triggered.

[0013] Optionally, the emergency response steps for abnormal situations specifically include: The system automatically identifies three types of anomalies—structural anomalies, dynamic anomalies, and environmental anomalies—based on a multi-sensor data feature library. It then classifies these anomalies into three severity levels—Level 1 warning, Level 2 intervention, and Level 3 emergency—based on the duration of the anomaly and the magnitude of exceeding the threshold. A tiered dynamic response strategy is adopted for different levels: during a Level 1 warning, the safety margin is calculated using a digital twin model, the load is appropriately reduced or the course and speed are adjusted, and the energy storage system is maintained in its normal state; during a Level 2 intervention, a "structural safety priority mode" or "hybrid power supply mode" is triggered, limiting the main engine power, shutting down unnecessary units, and activating anti-roll devices; during a Level 3 emergency, a "ship protection mode" or "island isolation mode" is enforced, the main engine is shut down, the energy storage system is switched to flywheel backup power, the course is adjusted according to the typhoon avoidance strategy, and a distress signal is sent. After the emergency ends, the ship's safety status is confirmed through health monitoring, and subsequent navigation restriction suggestions are generated based on the accumulated sub-health status. The abnormal data is stored in the emergency case library, and transfer learning is used to optimize future response parameters.

[0014] A second aspect of the present invention also provides a ship power dispatching system based on multi-sensor data fusion, comprising: The multi-source data acquisition module is used to collect real-time data on the ship's operating status through various sensors deployed in different parts of the ship, and to collect real-time environmental variable data of the waters in which the ship is located through the environmental monitoring unit; The data fusion processing module is used to align the ship's operating status data with the environmental variable data using timestamps and spatial coordinates to form a multi-source heterogeneous data fusion matrix. The power demand prediction module has a built-in pre-trained ship power demand prediction model, which is used to receive the multi-source heterogeneous data fusion matrix and output the predicted propulsion power demand of the ship under the current navigation conditions and the corresponding power margin range. The health status assessment module is used to calculate the health status parameters of the ship's power system in real time, including the degree of fatigue damage to the hull structure, the remaining effective life of the main engine, and the derating factor of the available power of the auxiliary power unit. The multi-objective optimization scheduling module is used to combine the predicted propulsion power demand, power margin range and health status parameters, and calculate the optimal power allocation scheme through a multi-objective optimization algorithm. The execution control interface module is used to distribute the optimal power allocation scheme to each power execution unit through the ship's power control network.

[0015] Optionally, the data fusion processing module further includes a sensor spatial position compensation unit, which uses the ship's six-degree-of-freedom motion model and pre-calibrated sensor spatial relative coordinates to perform coordinate transformation on the data collected by sensors at different installation positions to a unified ship reference frame.

[0016] Optionally, the ship power dispatching system based on multi-sensor data fusion also includes: Multi-ship collaborative scheduling unit: acquires the status of ships in the formation through AIS / VHF, allocates total power demand based on game theory model, and outputs collaborative scheduling instructions; Hybrid energy storage management module: integrates battery, supercapacitor, and flywheel energy storage, and dynamically selects energy storage type and charging / discharging power according to scheduling needs; The Level 3 Emergency Response Module: Employs three response modes for emergency handling: Level 1 Early Warning, Level 2 Intervention, and Level 3 Emergency. Level 1 warning: Automatically adjust power distribution when data approaches a safe threshold; Level 2 intervention: When data exceeds limits, trigger the safety priority mode, shut down unnecessary equipment, and switch energy storage to backup power; Level 3 Emergency: In extreme sea conditions, the ship protection mode is forcibly activated, all power is prioritized to the side thrusters and roll stabilization devices, and the main engine is shut down.

[0017] The ship power scheduling method and system based on multi-sensor data fusion disclosed herein have the following technical advantages: This invention utilizes multiple sensors, including those for strain, acceleration, rotational speed, fuel flow, and pressure, and employs timestamp alignment, spatial coordinate registration, six-degree-of-freedom motion model compensation, and extended Kalman filtering to construct a unified multi-source heterogeneous data fusion matrix. Furthermore, by introducing multimodal data feature cross-validation, self-attention mechanisms, and isolated forest anomaly detection, data from multiple sensors can be cross-checked, and abnormal data is automatically isolated and triggers fault alarms. This effectively overcomes the problem of single sensors being susceptible to noise interference or occasional malfunctions leading to misjudgments.

[0018] This invention uses a mother ship-type LSTM / GRU as the base model, requiring only a small amount of navigation data to fine-tune parameters and achieve cross-ship type adaptation. Simultaneously, it constructs a graph structure by associating sensor data with physical relationships, uses GCN to learn spatial dependencies, and outputs feature vectors with topological weights. It can output short-term propulsion power demand sequences, medium-term power margin trends, and long-term energy consumption predictions and their probability distributions, avoiding scheduling failures caused by point prediction biases.

[0019] This invention constructs a digital twin of the hull and propulsion system, integrates real-time strain / vibration data, uses a physics engine to dynamically simulate stress distribution under power scheduling, and calculates the fatigue damage rate of key nodes online based on a nonlinear fatigue damage accumulation model. At the same time, it uses an improved PHM model to assess the remaining life of the main engine / auxiliary power unit, and limits high-load operation based on the remaining life and scheduling strategy, successfully avoiding the propagation of base cracks and ensuring navigation safety.

[0020] This invention integrates real-time ocean current field, underwater topography, and local severe convection early warning by shipborne phased array current-measuring radar, lidar, and short-term meteorological receiving terminal. It constructs a dynamic environmental digital map, marking high-energy-consumption areas, high-risk areas, and low-stress recommended routes based on a hydrodynamic-stress coupling model. It establishes a four-dimensional optimization objective, adopts an improved MOPSO output Pareto optimal speed-route scheme, and uses a DQN agent to compare the actual and predicted deviations in real time for online correction, thereby achieving global optimal speed and route decision-making in dynamic environments.

[0021] This invention automatically identifies three types of anomalies—structural, dynamic, and environmental—through a multi-sensor data feature library. Based on the duration and the magnitude of exceeding the threshold, these anomalies are classified into three levels: Level 1 warning, Level 2 intervention, and Level 3 emergency. A graded dynamic response strategy is adopted for each level, which can improve the survivability of ships under extreme operating conditions.

[0022] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0024] Figure 1 A flowchart illustrating the ship power scheduling method based on multi-sensor data fusion provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a ship power dispatching system based on multi-sensor data fusion provided in an embodiment of the present invention; Figure 3 This is a flowchart of a three-level emergency response for abnormal states in an embodiment of the present invention. Detailed Implementation

[0025] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0026] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0027] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0028] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0029] This invention proposes an embodiment of a ship power scheduling method based on multi-sensor data fusion, specifically, as follows: Figure 1 As shown, it includes the following steps: The ship's operational status data is collected in real time by a variety of sensors deployed in different parts of the ship. These sensors include at least: strain sensors and acceleration sensors installed on the hull structure, speed sensors and fuel flow sensors installed on the power system, and pressure sensors installed near the waterline of the hull.

[0030] The environmental variable data of the waters where the ship is located are collected in real time by an environmental monitoring unit. The environmental variable data includes at least wave height, wind speed, wind direction, water flow speed and water flow direction. The environmental monitoring unit may include weather instruments, wave radar, current meters, etc.

[0031] The ship's operational status data and environmental variable data are time-stamped and spatially registered to eliminate temporal misalignment caused by differences in sampling frequencies of different sensors. The sensor data from different installation locations are transformed into a unified ship reference frame to form a multi-source heterogeneous data fusion matrix. This matrix comprehensively reflects the ship's real-time motion status, structural stress, power output, energy consumption, and surrounding marine environment, providing a unified and reliable input for subsequent prediction and decision-making.

[0032] The multi-source heterogeneous data fusion matrix is ​​input into a pre-trained ship power demand prediction model. The model is constructed based on ship dynamics simulation and historical navigation data. It can learn the complex nonlinear mapping between different operating conditions such as speed, draft, and sea state and propulsion power demand, and output the predicted propulsion power demand and corresponding power margin range of the ship under the current navigation conditions. The predicted propulsion power demand is the main engine power required to maintain the set speed; the power margin range is the adjustable power range after considering factors such as environmental disturbances and model uncertainties.

[0033] Combining the predicted propulsion power demand, power margin range, and real-time health status parameters of the ship's current propulsion system, an optimal power allocation scheme is calculated using a multi-objective optimization algorithm. This optimal power allocation scheme includes real-time allocation commands for the main engine output power, auxiliary power unit output power, and energy storage system charging / discharging power. Real-time health status parameters include, but are not limited to, the cumulative degree of fatigue damage to the hull structure, the remaining effective lifespan of the main engine and auxiliary power unit, and the available power derating factor. Multi-objective optimization algorithms such as NSGA-III, MOPSO, or deep reinforcement learning can be used. This scheme not only includes the main engine output power but also covers the charging / discharging power allocation of auxiliary power units (such as shaft-driven generators and auxiliary engines) and energy storage systems (such as batteries, supercapacitors, and flywheels), aiming to simultaneously optimize multiple conflicting objectives such as fuel consumption, carbon emissions, structural safety, and equipment lifespan.

[0034] The optimal power allocation scheme is distributed to each power execution unit via the ship's power control network, achieving closed-loop scheduling and control of the ship's power system. The ship's power control network can be a CAN bus, industrial Ethernet, or a ship automation system. Each power execution unit, such as the main engine control unit, generator controller, and energy storage converter, enables closed-loop scheduling and control of the power system. The system repeats the above-described process of data acquisition, fusion, prediction, optimization, and distribution at a fixed cycle, continuously tracking changes in navigation status and dynamically adjusting power allocation.

[0035] In this embodiment of the invention, the construction process of the multi-source heterogeneous data fusion matrix includes: using a multimodal data feature cross-validation unit to calculate the credibility of multi-sensor data of the same physical quantity through a Bayesian network; marking conflicting data and enabling redundant data when the deviation exceeds a threshold; wherein, the multi-sensor data of the same physical quantity includes, but is not limited to, strain sensor data and pressure sensor data corresponding to hull stress; using a self-attention mechanism dynamic weight allocation layer to learn the contribution of different sensor data to the overall state of the ship, and outputting a fusion feature vector with dynamic weights; using the isolated forest algorithm to identify and isolate abnormal data that deviates from the mean 3σ for three consecutive sampling periods, triggering a sensor fault alarm; using an extended Kalman filter to filter out noise and interpolate missing values ​​in the remaining data, and combining it with the ship's six-degree-of-freedom motion model for coordinate transformation compensation.

[0036] In this embodiment of the invention, the ship power demand prediction model is a multi-ship type transfer learning + graph convolutional network (GCN) fusion model: using the parent ship type LSTM / GRU model as the base model, the parameters are fine-tuned through a small amount of navigation data of the target ship type to achieve cross-ship type adaptation; the sensor data (nodes) and physical associations (edges) are used to construct a graph structure, and the spatial dependencies are learned by GCN to output feature vectors with topological weights; the model input features include current / historical speed, draft, attitude, wave height spectrum, wind speed vector, water flow gradient and sensor topological features, and output short-term (0-5min) propulsion power demand sequence, medium-term (5-60min) power margin trend, long-term (1-24h) energy consumption prediction and its probability distribution.

[0037] In this embodiment of the invention, the multi-objective optimization algorithm includes: a deep reinforcement learning (DDPG algorithm) agent that takes sea state, cargo load, speed and other scenario state information as input, and adjusts the optimization target weights in real time, such as the weight combination of safety margin, fuel consumption and new energy utilization rate; adopts multi-timescale rolling optimization, optimizes real-time power allocation in the short term (1min), optimizes route fine-tuning in the medium term (10min), and optimizes energy storage charging and discharging plan in the long term (1h), and connects them through rolling time domain control; the optimization objective function includes: minimizing fuel / new energy combined consumption, minimizing carbon emissions, maximizing the remaining life of the power system, maximizing navigation safety margin, and maximizing new energy utilization rate, and uses an improved NSGA-III algorithm to solve the Pareto front. Specifically, in actual ship testing, upon entering the Emissions Control Area (ECA), the system automatically increases the carbon emission weight by 30% and decreases the fuel consumption weight by 10%, thus prioritizing the use of low-sulfur fuel and shore power, resulting in a 22% reduction in emissions compared to the fixed-weight scheme. When encountering gale-force winds (level 8), the system increases the safety margin weight by 50%, proactively reducing speed and main engine load, leading to a 25% decrease in the maximum dynamic stress amplitude of the hull. In contrast, the fixed-weight scheme still prioritizes fuel economy, resulting in stress over-limit alarms. This adaptive capability is significantly superior to manually set or fixed rules. This embodiment introduces a preference ranking mechanism, transforming the dynamic weights output by DDPG into reference points, guiding the algorithm to converge directly to the decision-maker's preferred region. This reduces the single optimization calculation time from several minutes to less than 5 seconds, meeting the online optimization requirement of a 1-minute cycle.

[0038] Furthermore, this embodiment achieves an organic unity of rapid power following and long-term energy management through short-term, medium-term, and long-term rolling optimization. For example, during a certain voyage, long-term optimization formulates a discharge plan for the energy storage system within the next hour to cope with the upcoming backflow. When the actual backflow occurs 10 minutes in advance, medium-term optimization promptly adjusts the route for fine-tuning, while short-term optimization smoothly increases the main engine power. The synergy of these three approaches results in actual energy consumption that is 7.3% lower than the solution using only short-term optimization, and the energy storage system does not experience deep discharge, reducing its lifespan degradation by 15%.

[0039] In this embodiment of the invention, the real-time health status parameters are obtained through dynamic evaluation driven by a digital twin, specifically including: hull structure evaluation: constructing a hull-power system digital twin, integrating real-time strain / vibration data, and dynamically simulating stress distribution under power scheduling using a physics engine; calculating the fatigue damage rate of key nodes such as keel welds and engine room bases online based on a nonlinear fatigue damage accumulation model; power system evaluation: using vibration / temperature data, combined with an improved PHM model, evaluating the remaining lifespan of the main engine / auxiliary power unit, and determining the output power derating factor when the remaining lifespan is <10%, and linking it with scheduling strategies to limit high-load operation. It should be noted that a high-fidelity hull-power system coupled digital twin model is established offline using the CAD / CAE model from the ship design phase and the physical parameters of the power system. The CAD / CAE model includes geometric dimensions, material properties, and boundary conditions; the physical parameters of the power system include main engine power, propeller characteristics, etc. During ship navigation, the digital twin receives real-time data from strain sensors and acceleration sensors deployed at key locations on the hull. Using these measured data as boundary conditions or correction sources, a real-time solver dynamically simulates the stress distribution of various parts of the hull under the current power scheduling scheme. The alternating loads experienced by the ship during navigation can lead to fatigue accumulation at critical nodes. This embodiment employs the classic rainflow counting method to perform cyclic counting of the real-time stress-time history, extracting each stress amplitude and cycle number, and then applying Miner's linear accumulation rule to calculate fatigue damage. Furthermore, the derating factor is, for example, 0.5-0.8, and the specific value of the derating factor is dynamically adjusted based on the remaining service life, the importance of the equipment, and the urgency of the current navigation mission: the shorter the remaining service life, the more critical the equipment, and the less urgent the navigation mission, the lower the derating factor.

[0040] When calculating the optimal power allocation scheme, the optimization algorithm uses a derating factor as a constraint: the actual output power of the main engine or auxiliary power unit must not exceed the derating factor multiplied by its rated power. For example, if a main engine has only 8% remaining lifespan and a derating factor of 0.6, then the maximum allowable power for this main engine when allocating power is 60% of its rated power, with the excess being supplemented by other auxiliary engines, energy storage systems, and other power units. Simultaneously, the system automatically adjusts the speed or route to match the reduced total available power, ensuring navigational safety.

[0041] In this embodiment of the invention, the method further includes a route adaptive optimization step, specifically comprising: By fusing real-time ocean current field, underwater topography, and local severe convection warnings from multiple sources such as shipborne phased array current-measuring radar, lidar, and short-term meteorological receivers, and overlaying these data onto electronic nautical charts, a dynamic environmental digital map is constructed. This map marks high-energy-consumption areas, high-risk areas, and low-stress recommended routes based on a hydrodynamic-stress coupling model. A four-dimensional optimization objective is established, encompassing minimizing energy consumption and time costs, maximizing hull structural safety, ensuring emission compliance, and adhering to navigation rules. An improved multi-objective particle swarm optimization (MOPSO) algorithm with a dynamic adjustment mechanism for objective preferences is employed to output a Pareto-optimal speed-route combination. A reinforcement learning (DQN) agent is used to compare the deviation between the actual speed and route and the predicted objective in real time and make online corrections. The dynamic environmental digital map is updated every 30 minutes, triggering a new round of optimization iterations.

[0042] In embodiments of the present invention, such as Figure 3 As shown, the emergency response steps for abnormal conditions specifically include: automatically identifying three types of abnormalities through a multi-sensor data feature library: structural abnormalities (such as strain exceeding limits, resonance), dynamic abnormalities (such as bearing overheating, speed fluctuations), and environmental abnormalities (such as wave height or wind speed exceeding thresholds, typhoon warnings). These are then classified into three severity levels based on the duration of the abnormality and the magnitude of exceeding the threshold: Level 1 warning, Level 2 intervention, and Level 3 emergency. A tiered dynamic response strategy is adopted for each level: During Level 1 warning, a safety margin is calculated using a digital twin model, the load is appropriately reduced, or the course and speed are adjusted, and the energy storage system is maintained in its normal state. During Level 2 intervention, a "structural safety priority mode" or "hybrid power supply mode" is triggered, limiting the main engine power, shutting down unnecessary units, and activating anti-roll devices. During Level 3 emergency, a "ship protection mode" or "island mode" is enforced, the main engine is shut down, the energy storage system is switched to flywheel backup power, the course is adjusted according to typhoon avoidance strategies, and a distress signal is sent. After the emergency ends, the ship's safety status is confirmed through health monitoring, and subsequent navigation restriction suggestions are generated based on accumulated sub-health conditions. The abnormal data is stored in the emergency case library, and transfer learning is used to optimize future response parameters.

[0043] The present invention also provides an embodiment of a ship power dispatching system based on multi-sensor data fusion, specifically as follows: Figure 2 As shown, it includes: The multi-source data acquisition module is used to collect real-time data on the ship's operating status through various sensors deployed in different parts of the ship, and to collect real-time environmental variable data of the waters in which the ship is located through the environmental monitoring unit; The data fusion processing module is used to align the ship's operating status data with the environmental variable data using timestamps and spatial coordinates to form a multi-source heterogeneous data fusion matrix. The power demand prediction module has a built-in pre-trained ship power demand prediction model, which is used to receive the multi-source heterogeneous data fusion matrix and output the predicted propulsion power demand of the ship under the current navigation conditions and the corresponding power margin range. The health status assessment module is used to calculate the health status parameters of the ship's power system in real time, including the degree of fatigue damage to the hull structure, the remaining effective life of the main engine, and the derating factor of the available power of the auxiliary power unit. The multi-objective optimization scheduling module is used to combine the predicted propulsion power demand, power margin range and health status parameters, and calculate the optimal power allocation scheme through a multi-objective optimization algorithm. The execution control interface module is used to distribute the optimal power allocation scheme to each power execution unit through the ship's power control network.

[0044] In this embodiment of the invention, the data fusion processing module further includes a sensor spatial position compensation unit. This unit uses the ship's six-degree-of-freedom motion model and pre-calibrated sensor spatial relative coordinates to perform coordinate transformation on the data collected by sensors at different installation positions to a unified ship reference frame.

[0045] In this embodiment of the invention, the ship power dispatching system based on multi-sensor data fusion further includes: a multi-ship collaborative dispatching unit: acquiring the fleet ship status through AIS / VHF, allocating total power demand based on a game theory model, and outputting collaborative dispatching instructions; a hybrid energy storage management module: integrating battery, supercapacitor, and flywheel energy storage, dynamically selecting energy storage type and charging / discharging power according to dispatching needs; and a three-level emergency response module: employing three response modes for emergency handling: Level 1 warning, Level 2 intervention, and Level 3 emergency, wherein: Level 1 warning: automatically adjusting power allocation (e.g., reducing main engine power by 5%) when data approaches a safety threshold; Level 2 intervention: triggering a safety priority mode when data exceeds limits, shutting down unnecessary equipment, and switching energy storage to backup power; and Level 3 emergency: forcibly activating a ship protection mode in extreme sea conditions (e.g., roll > 35°), prioritizing power supply to the side thruster anti-roll device, and shutting down the main engine.

[0046] In this embodiment of the invention, the multi-objective optimization scheduling module is equipped with a navigation scenario identification unit. This unit automatically identifies the ship as being in one of the following navigation scenarios based on the ship's current speed, draft, and environmental variable data: port maneuvering, economic speed cruising, wind and wave resistant navigation, or ice zone navigation, and calls the corresponding optimization objective weight combination and constraint conditions according to different scenarios.

[0047] In this embodiment of the invention, the system further includes a route adaptive optimization subsystem, comprising: Dynamic Environment Perception Unit: Integrates phased array current measurement radar, LiDAR, and short-term meteorological terminal, outputting real-time flow field, terrain, and meteorological data; Multi-Objective Optimization Engine: Built-in four-dimensional optimization objective function and improved MOPSO algorithm, outputting Pareto optimal speed-route scheme; Closed-Loop Feedback Control Unit: Dynamically adjusts subsequent strategies by comparing the actual and target deviations through a DQN agent; Digital Twin Inference Module: Based on the ship's hydrodynamic-stress coupling model, pre-calculates the stress distribution and energy consumption of candidate routes, and marks "low-stress recommended routes".

[0048] In this embodiment of the invention, the route adaptive optimization step specifically includes: Integrating multi-source real-time environmental data: In addition to electronic charts and weather forecasts, real-time ocean current fields are acquired through shipborne phased array current measurement radar and lidar (LiDAR); underwater topography is scanned within 5 nautical miles ahead, and 0-2 hour local severe convective weather warnings are obtained through short-term weather receiving terminals; a dynamic environmental digital map is constructed: real-time current field, topography, and short-term weather data are overlaid on electronic charts, and "high energy consumption areas", "high risk areas", and "low stress recommended routes" are marked.

[0049] The four-dimensional optimization objectives are defined as follows: Minimizing energy consumption and time cost: calculating the comprehensive cost per unit distance by considering speed, route length, flow field assistance (acceleration with the current), and wind resistance increment (deceleration against the wind); Maximizing hull structural safety: pre-calculating the stress distribution of key hull nodes (keel / side) under different routes using a digital twin model to avoid "high-stress routes" (such as increased mid-sag / sagging caused by cross-wave zones); Ensuring emission compliance: automatically limiting speed to within sulfur oxide / nitrogen oxide limits within ECA (Emission Control Area); Compliance with regulations and navigation rules: avoiding no-navigation zones and fishing areas, and meeting channel width / depth constraints; Adopting an improved multi-objective particle swarm optimization (MOPSO) algorithm: introducing a dynamic adjustment mechanism for objective preferences, such as increasing the weight of time cost when the cargo owner specifies "priority timeliness" and increasing the weight of energy consumption when the shipowner specifies "priority energy saving," outputting a Pareto optimal solution set; The system compares the deviation between the actual speed / route and the predicted target in real time. For example, if the speed decreases by 10% due to a sudden change in the flow field, the speed-route combination of the subsequent segments is adjusted online using the DQN algorithm until the actual energy consumption / stress approaches the optimization target. The dynamic environment digital map is updated every 30 minutes to trigger a new round of optimization iteration.

[0050] In specific embodiments of the present invention, as follows: Figure 3 As shown, the emergency response steps for abnormal states in the method specifically include: Anomaly type identification: Automatically distinguishes three types of anomalies using a multi-sensor data feature library: Structural anomalies: Hull strain sensor data exceeds threshold (e.g., local stress > 80% yield strength), accelerometers detect abnormal vibrations (e.g., hull resonance caused by propeller imbalance). Power abnormality: Vibration / temperature sensor data of the power system exceeds the threshold (e.g., main bearing temperature > 120℃), speed sensor detects speed fluctuation > ±5%; Anomalies in the environment: Marine environmental data exceeding thresholds (e.g., wave height > 6m, wind speed > 25m / s), short-term meteorological warning "red typhoon warning"; Severity grading: Based on the duration of the anomaly and the magnitude of exceeding the threshold, it is divided into three levels: Level 1 (Warning): Data reaches 70%-100% of the threshold, lasting for less than 5 minutes; Level 2 (Intervention): Data exceeds the threshold and persists for 5-15 minutes, or a single parameter exceeds the threshold by more than 50%; Level 3 (Emergency): Data exceeds the threshold by more than 50% and lasts for more than 15 minutes, or triggers a "catastrophic anomaly" (such as the hull crack propagation rate > 0.1 mm / h). The Level 1 (early warning) includes: Structural anomaly: Calculate the "stress safety margin" using a digital twin model. If the margin is >20%, only record and prompt "Pay attention to structural status"; Power anomaly: Reduce the load of abnormal equipment by 10% (e.g., reduce the main engine power by 5%) and activate backup sensor cross-verification; Environmental anomaly: Adjust the heading by 5°-10° to avoid strong wind areas and reduce the speed by 5%; Energy storage system: Maintain the current charging and discharging state and only activate the "health status monitoring mode"; Level 2 (intervention) includes: Structural anomaly: Triggers "structural safety priority mode," calculates "minimum stress route" via digital twin, automatically adjusts course / speed, and limits main engine power to 70% of rated value; Power anomaly: Shuts down non-essential power units (such as deck machinery) and switches the energy storage system to "hybrid power supply mode" (batteries supply critical equipment, supercapacitors supply rapid propulsion); Environmental anomaly: If roll angle > 20°, activates bow and stern thruster roll damping devices, reducing speed to 60% of economic speed; Level 3 (Emergency) includes: Structural anomaly: Force "Ship Protection Mode", prioritize power supply to the side thrusters and roll damping devices, shut down the main engine, send an SOS via satellite communication and mark the "structural damage location"; Power anomaly: Activate "Island Mode", retain only emergency lighting / communication, switch the energy storage system to "flywheel priority backup power" (to ensure the continuous operation of the roll damping devices); Environmental anomaly: If a typhoon is encountered, adjust the course according to the "Typhoon Eye Avoidance Strategy", reduce the speed to "drift speed", release ballast water to adjust the trim to reduce the wind-exposed area; After the emergency ends, the ship's "safety status" is confirmed by structural health monitoring data and power system health assessment, such as when the structural stress recovers to below 80% of the baseline. If a "sub-healthy state" is accumulated during the emergency, such as microcracks in the structure, "subsequent navigation restriction recommendations" are automatically generated, such as prohibiting entry into high sea state areas. The abnormal data is stored in the emergency case library, and the response parameters for similar anomalies in the future are optimized through transfer learning.

[0051] The above description of the structure, features, and effects of the present invention is based on the embodiments shown in the figures. However, the above are only preferred embodiments of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched by those skilled in the art to form a variety of equivalent solutions without departing from or changing the design concept and technical effects of the present invention. Therefore, the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A ship power scheduling method based on multi-sensor data fusion, characterized in that, Includes the following steps: The ship's operating status data is collected in real time by a variety of sensors deployed in different parts of the ship. The various sensors include at least: strain sensors and acceleration sensors installed on the hull structure, speed sensors and fuel flow sensors installed on the power system, and pressure sensors installed near the waterline of the hull. The environmental monitoring unit collects environmental variable data of the waters where the ship is located in real time. The environmental variable data includes at least: wave height, wind speed, wind direction, water flow speed and water flow direction. The ship operation status data and the environmental variable data are time-stamped and spatially registered to form a multi-source heterogeneous data fusion matrix. The multi-source heterogeneous data fusion matrix is ​​input into a pre-trained ship power demand prediction model. The model is constructed based on ship dynamics simulation and historical navigation data, and outputs the predicted propulsion power demand of the ship under the current navigation conditions and the corresponding power margin range. Combining the predicted propulsion power demand, power margin range, and real-time health status parameters of the ship's current power system, the optimal power allocation scheme is calculated using a multi-objective optimization algorithm. The optimal power allocation scheme includes real-time allocation instructions for the main engine output power, auxiliary power unit output power, and energy storage system charging and discharging power. The optimal power allocation scheme is distributed to each power execution unit through the ship's power control network to achieve closed-loop scheduling and control of the ship's power system.

2. The ship power scheduling method based on multi-sensor data fusion according to claim 1, characterized in that, The construction process of the multi-source heterogeneous data fusion matrix includes: A multimodal data feature cross-validation unit is used to calculate the credibility of multi-sensor data of the same physical quantity through a Bayesian network. When the deviation exceeds the threshold, conflicting data is marked and redundant data is enabled. By utilizing a self-attention mechanism and a dynamic weight allocation layer, the contribution of different sensor data to the overall ship status is learned, and a fused feature vector with dynamic weights is output. The isolated forest algorithm is used to identify and isolate abnormal data that deviates from the mean for three consecutive sampling periods, triggering a sensor fault alarm. Extended Kalman filtering was used to remove noise and interpolate missing values ​​in the remaining data, and coordinate transformation compensation was performed in conjunction with the six-degree-of-freedom motion model of the ship.

3. The ship power scheduling method based on multi-sensor data fusion according to claim 1, characterized in that, The ship power demand prediction model specifically includes: Using the LSTM / GRU model of the mother ship as the base model, the parameters are fine-tuned using a small amount of navigation data of the target ship. The sensor data is linked to the physical structure to construct a graph structure, and the spatial dependencies are learned using GCN to output feature vectors with topological weights. The model input features include current / historical speed, draft, attitude, wave height spectrum, wind speed vector, water flow gradient, and sensor topology features. The output features include short-term propulsion power demand sequence, medium-term power margin trend, long-term energy consumption prediction, and their probability distribution.

4. The ship power scheduling method based on multi-sensor data fusion according to claim 1, characterized in that, The multi-objective optimization algorithm includes: Based on a deep reinforcement learning agent, the target weights are adjusted and optimized in real time with the scene state as input. Multi-timescale rolling optimization is adopted, with short-term optimization of real-time power allocation, medium-term optimization of route fine-tuning, and long-term optimization of energy storage charging and discharging plans, which are connected by rolling time-domain control. The optimization objective functions include: minimizing combined fuel / new energy consumption, minimizing carbon emissions, maximizing the remaining lifespan of the power system, maximizing navigation safety margin, and maximizing the utilization rate of new energy sources. The Pareto front is solved using the improved NSGA-III algorithm.

5. The ship power scheduling method based on multi-sensor data fusion according to claim 4, characterized in that, The real-time health status parameters are obtained through dynamic evaluation driven by digital twins, specifically including: Hull structure assessment: Construct a digital twin of the hull and propulsion system, integrate real-time strain / vibration data, and use a physics engine to dynamically simulate stress distribution under power scheduling; calculate the fatigue damage rate of key nodes online based on a nonlinear fatigue damage accumulation model. Power system evaluation: Using vibration / temperature data and an improved PHM model, the remaining life of the main engine / auxiliary power unit is evaluated. When the remaining life reaches a certain percentage, the output power derating factor is determined and linked with the scheduling strategy to limit high-load operation.

6. The ship power scheduling method based on multi-sensor data fusion according to any one of claims 1 to 5, characterized in that, The method also includes a route adaptive optimization step, specifically including: By fusing real-time ocean current field, underwater topography and local severe convection early warning by multiple sources of sensors such as shipborne phased array current measurement radar, lidar and short-term weather receiving terminal, and superimposing them on electronic nautical charts, a dynamic environmental digital map is constructed, marking high energy consumption areas, high risk areas and low stress recommended routes based on hydrodynamic-stress coupling model; A four-dimensional optimization objective is established, which includes minimizing energy consumption and time cost, maximizing hull structural safety, ensuring emission compliance, and adhering to navigation rules. An improved multi-objective particle swarm optimization algorithm with a dynamic adjustment mechanism for objective preferences is adopted to output the Pareto optimal speed-route combination scheme. By using a reinforcement learning agent to compare the deviation between the actual speed and route and the predicted target in real time and make online corrections, the dynamic environment digital map is updated every 30 minutes and a new round of optimization iteration is triggered.

7. The ship power scheduling method based on multi-sensor data fusion according to claim 6, characterized in that, It also includes emergency response procedures for abnormal situations, specifically including: The system automatically identifies three types of anomalies—structural anomalies, dynamic anomalies, and environmental anomalies—based on a multi-sensor data feature library. It then classifies these anomalies into three severity levels—Level 1 warning, Level 2 intervention, and Level 3 emergency—based on the duration of the anomaly and the magnitude of exceeding the threshold. A tiered dynamic response strategy is adopted for different levels: during a Level 1 warning, the safety margin is calculated using a digital twin model, the load is appropriately reduced or the course and speed are adjusted, and the energy storage system is maintained in its normal state; during a Level 2 intervention, a "structural safety priority mode" or "hybrid power supply mode" is triggered, limiting the main engine power, shutting down unnecessary units, and activating anti-roll devices; during a Level 3 emergency, a "ship protection mode" or "island isolation mode" is enforced, the main engine is shut down, the energy storage system is switched to flywheel backup power, the course is adjusted according to the typhoon avoidance strategy, and a distress signal is sent. After the emergency ends, the ship's safety status is confirmed through health monitoring, and subsequent navigation restriction suggestions are generated based on the accumulated sub-health status. The abnormal data is stored in the emergency case library, and transfer learning is used to optimize future response parameters.

8. A ship power dispatching system based on multi-sensor data fusion, characterized in that, include: The multi-source data acquisition module is used to collect real-time data on the ship's operating status through various sensors deployed in different parts of the ship, and to collect real-time environmental variable data of the waters in which the ship is located through the environmental monitoring unit; The data fusion processing module is used to align the ship's operating status data with the environmental variable data using timestamps and spatial coordinates to form a multi-source heterogeneous data fusion matrix. The power demand prediction module has a built-in pre-trained ship power demand prediction model, which is used to receive the multi-source heterogeneous data fusion matrix and output the predicted propulsion power demand of the ship under the current navigation conditions and the corresponding power margin range. The health status assessment module is used to calculate the health status parameters of the ship's power system in real time, including the degree of fatigue damage to the hull structure, the remaining effective life of the main engine, and the derating factor of the available power of the auxiliary power unit. The multi-objective optimization scheduling module is used to combine the predicted propulsion power demand, power margin range and health status parameters, and calculate the optimal power allocation scheme through a multi-objective optimization algorithm. The execution control interface module is used to distribute the optimal power allocation scheme to each power execution unit through the ship's power control network.

9. The ship power dispatching system based on multi-sensor data fusion according to claim 8, characterized in that, The data fusion processing module also includes a sensor spatial position compensation unit. This unit uses the ship's six-degree-of-freedom motion model and pre-calibrated sensor spatial relative coordinates to perform coordinate transformation on the data collected by sensors at different installation positions to a unified ship reference frame.

10. The ship power dispatching system based on multi-sensor data fusion according to claim 8, characterized in that, Also includes: Multi-ship collaborative scheduling unit: acquires the status of ships in the formation through AIS / VHF, allocates total power demand based on game theory model, and outputs collaborative scheduling instructions; Hybrid energy storage management module: integrates battery, supercapacitor, and flywheel energy storage, and dynamically selects energy storage type and charging / discharging power according to scheduling needs; The Level 3 Emergency Response Module: Employs three response modes for emergency handling: Level 1 Early Warning, Level 2 Intervention, and Level 3 Emergency. Level 1 warning: Automatically adjust power distribution when data approaches a safe threshold; Level 2 intervention: When data exceeds limits, trigger the safety priority mode, shut down unnecessary equipment, and switch energy storage to backup power; Level 3 Emergency: In extreme sea conditions, the ship protection mode is forcibly activated, all power is prioritized to the side thrusters and roll stabilization devices, and the main engine is shut down.

Images