Ship hybrid power scheduling method and system based on digital twinning and distributed collaboration

Abstract

This invention belongs to the field of marine propulsion technology and provides a method and system for scheduling hybrid power systems based on digital twins and distributed collaboration. The method includes collecting and processing multi-source data to obtain multi-dimensional synchronous data vectors, constructing a multi-step prediction model to obtain predicted multi-dimensional synchronous data vectors, building an operating cost model and a preset strategy library to generate power control commands for optimal hybrid power allocation, and constructing a virtual digital twin model and a distributed collaborative model. The method and system for scheduling hybrid power systems based on digital twins and distributed collaboration provided by this invention increase the safety and reliability of decision execution by performing simulation operation and safety verification before decision execution through the digital twin model. Simultaneously, the distributed collaborative model enables each power unit to have local autonomous decision-making capabilities to cope with sudden extreme situations, improving the robustness and fault tolerance of hybrid power scheduling in dynamic marine environments.

Description

Technical Field

[0001] This invention belongs to the field of marine propulsion technology, and specifically relates to a method and system for scheduling hybrid power systems of ships based on digital twins and distributed collaboration. Background Technology

[0002] Currently, marine propulsion systems have evolved from traditional single-power systems to modern hybrid power systems. These systems introduce new power sources such as batteries and fuel cells, forming a multi-energy collaborative power supply system with traditional diesel engines. The key technology for marine hybrid power systems is the allocation and scheduling of power sources across the hybrid system. However, current mainstream hybrid power control systems generally adopt a centralized control architecture, where a central controller processes power system sensor data and calculates power allocation commands. This architecture suffers from the following problems and shortcomings: First, the central controller generally relies on pre-set operating condition logic, making it difficult to handle complex and dynamic marine environments. Second, the central controller manages all sensing, decision-making, and execution of the hybrid power system. When faced with sudden severe sea conditions or equipment disturbances, significant response delays occur in the data acquisition, transmission, decision-making, and execution links, causing the hybrid power system to fail to respond promptly to changes in instantaneous power demand, severely impacting the ship's maneuverability in severe sea conditions. Third, the allocation and scheduling of marine hybrid power relies entirely on the central controller, lacking effective fault-tolerance and safety verification mechanisms. In some cases, this can even lead to a sharp decline in power performance or loss of control, seriously affecting the safety of ship operation.

[0003] Therefore, it is necessary to design a ship hybrid power scheduling method and system that can at least solve some of the above problems and defects. Summary of the Invention

[0004] To address the above technical problems, this invention proposes a ship hybrid power scheduling method and system based on digital twins and distributed collaboration. By using a digital twin model to perform simulation operation and safety verification before decision execution, the safety and reliability of decision execution are increased. At the same time, the distributed collaboration model enables each power unit to have local autonomous decision-making capabilities to cope with sudden extreme situations, thereby improving the robustness and fault tolerance of ship hybrid power scheduling in dynamic marine environments.

[0005] The technical solution of this invention is:

[0006] This invention proposes a ship hybrid power scheduling method based on digital twins and distributed collaboration, comprising the following steps:

[0007] Collect and acquire multi-source data including sea state and environmental data, ship dynamic status data and ship motion status data, and perform data cleaning and synchronization alignment on the multi-source data to obtain a multi-dimensional synchronized data vector.

[0008] Extract sea state and ship motion data from the multidimensional synchronous data vector, perform fusion and correction processing to obtain corrected wave feature data, generate a quantized sea state level based on the corrected wave feature data, and update the multidimensional synchronous data vector.

[0009] A multi-step prediction model is constructed based on deep learning algorithms. Multi-dimensional synchronous data vectors are input into the multi-step prediction model to obtain predicted multi-dimensional synchronous data vectors for future time periods.

[0010] An operating cost model including multi-objective optimization and a preset strategy library including several weight coefficient vectors are constructed. The operating cost model generates power control commands for optimal allocation of hybrid power based on the predicted multidimensional synchronous data vector and the weight coefficient vector.

[0011] Construct a virtual digital twin model corresponding to the ship, initialize it by inputting a multidimensional synchronous data vector, and receive a predicted multidimensional synchronous data vector to load the predicted sea state environment for future time periods;

[0012] A distributed collaborative model is constructed based on a multi-agent reinforcement learning framework, comprising multiple agents deployed at various power units of the hybrid power system. The power control commands are transmitted sequentially through the operating cost model, the virtual digital twin model, and the distributed collaborative model.

[0013] Preferably, the sea environment data includes wave height, wave direction, wave period, wind direction, and wind speed data; the ship power status data includes real-time operating status data of the fuel power unit, battery power unit, and fuel cell power unit; and the ship motion status data includes sailing speed, sailing course, roll parameters, pitch parameters, and heave parameters.

[0014] Preferably, the synchronization alignment process includes the following steps:

[0015] Receive multi-source data after data cleaning and processing, and mark it with a high-precision system timestamp;

[0016] Establish corresponding processing buffers based on different data stream acquisition frequencies to manage data streams of different frequencies;

[0017] Interpolation and extrapolation methods are used to align all data to the synchronous clock, generating a multi-dimensional synchronous data vector.

[0018] Preferably, the fusion correction process includes the following steps:

[0019] Extract the vector data corresponding to the roll, pitch, and heave parameters from the multidimensional synchronous data vector, and obtain the corresponding kinematic energy spectrum through spectral analysis;

[0020] The inferred wave direction spectrum is obtained by inverting the ship motion response function, and the inferred wave characteristic data including significant wave height, average wave period, and main wave direction are extracted from it.

[0021] An adaptive Kalman filter algorithm is used to fuse the inferred wave feature data and the sea state environment data to obtain corrected wave feature data.

[0022] Preferably, the operating cost model includes a total operating cost function:

[0023] ;

[0024] in, For fuel costs, As a penalty for emissions, As a penalty for power error, As a comfort penalty, , , , These are the corresponding weight coefficients, and ;

[0025] The weight coefficient vector consists of weight coefficients , , , The composition and weighting coefficients are adjusted based on the quantified sea state level, navigation position and stage, mission priority, equipment usage status, and regional environmental regulations.

[0026] Preferably, the virtual digital twin model receives the power control commands generated by the operating cost model to run simulations and record simulation data, while simultaneously conducting safety and performance evaluation and testing on the simulation data;

[0027] If an anomaly is detected during the assessment, a correction suggestion is generated and fed back to the operating cost model. The operating cost model then regenerates the corrected power control command based on the correction suggestion.

[0028] If no abnormalities are found in the assessment, the power control command will be passed to the distributed cooperative model for execution.

[0029] Preferably, the virtual digital twin model includes a ship motion and resistance model, a power system dynamic model, and an environmental interaction model;

[0030] The distributed collaborative model includes a fuel-powered intelligent agent, a battery pack intelligent agent, and a fuel cell intelligent agent.

[0031] Preferably, the working modes of multiple agents in the distributed collaborative model include a collaborative operation mode and an autonomous operation mode;

[0032] In collaborative operation mode, multiple intelligent agents receive power control commands sent through a virtual digital twin model to adjust the target output power of the corresponding power units;

[0033] In autonomous operation mode, multiple agents autonomously generate locally optimal instructions based on their own reinforcement learning training framework in order to autonomously adjust the target output power of the corresponding power unit.

[0034] Preferably, the distributed collaborative model is equipped with an anomaly identification and detection unit to detect abnormal situations, including environmental changes, equipment malfunctions, and communication interruptions, and to switch multiple agents between collaborative operation mode and autonomous operation mode based on the detection results.

[0035] Preferably, the present invention also provides a ship hybrid power dispatching system based on digital twins and distributed collaboration, comprising:

[0036] The data acquisition module is used to collect and acquire multi-source data, including sea state and environmental data, ship dynamic status data, and ship motion status data.

[0037] The data processing module is used to perform data cleaning, synchronization alignment, and fusion correction on the acquired multi-source data to obtain a multi-dimensional synchronized data vector.

[0038] The central decision-making module is used to receive multi-dimensional synchronous data vectors and generate power control commands for optimal allocation of hybrid power. The central decision-making module includes a multi-step prediction unit with a multi-step prediction model, a multi-objective optimization unit with an operating cost model, and an expert strategy library unit with weight coefficient vectors.

[0039] The digital twin module is used to construct a virtual digital twin model corresponding to the ship and receive power control commands issued by the central decision-making module to simulate the decision-making results and judge the feasibility of the power control commands based on the simulation results.

[0040] The distributed coordination module is used to build a distributed coordination model that includes multiple intelligent agents deployed at various power units of the hybrid power system, in order to receive power control commands sent through the digital twin module or autonomously generate locally optimal commands.

[0041] The present invention has the following advantages and effects compared with the prior art:

[0042] (1) A virtual simulation environment corresponding to the physical ship body is established by adopting a digital twin platform, so that the power control command is simulated before it is issued to the physical power system, thereby identifying and correcting some risk commands that may cause equipment overload or exceed limits in advance, playing a proactive prevention role, and achieving forward-looking safety verification, so as to increase the safety and reliability of the execution of decision commands.

[0043] (2) A distributed cooperative model is adopted, which includes multiple agents deployed at each power unit of the hybrid power system. This enables each power unit to have local autonomous decision-making capabilities. In the event of sudden extreme conditions, multiple distributed agents can still maintain the basic stable operation of the hybrid power system based on local observation and cooperative strategies, and achieve rapid and timely local power adjustment and response, thereby improving the robustness and fault tolerance of ship hybrid power scheduling in complex and uncertain marine environments.

[0044] (3) A preset strategy library with several weight coefficient vectors is adopted. The values ​​of the corresponding weight coefficients in the weight coefficient vectors are adjusted according to different strategies, such as the quantification of sea state level, navigation position stage, task priority, equipment usage status, regional environmental regulations, etc., so as to form a comprehensive, dynamic and detailed multi-response strategy, making the scheduling and allocation of hybrid power more comprehensive, refined and rational. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the ship hybrid power scheduling method based on digital twins and distributed collaboration in an embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram illustrating the principle of a ship hybrid power scheduling system based on digital twins and distributed collaboration in an embodiment of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the present invention, specific embodiments will now be described in further detail. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention.

[0048] Example:

[0049] like Figure 1 As shown, this invention provides a ship hybrid power scheduling method based on digital twins and distributed collaboration, which specifically includes the following steps:

[0050] The system collects multi-source data, including sea state environmental data, ship dynamic status data, and ship motion status data. Sea state environmental data includes wave height, direction, and period data at the bow of the ship acquired via 3D wave radar, and wind direction and speed data acquired via meteorological sensors. The acquisition frequency for wave height, direction, and period data is generally in the range of 0.1-0.2 Hz, while the acquisition frequency for wind direction and speed data is generally in the range of 1-5 Hz. Ship dynamic status data includes real-time operating parameters of several hybrid power units acquired via sensors within the power system. Specifically, this includes real-time operating parameters of fuel cell power units (such as diesel engines) such as speed, torque, fuel flow, and exhaust temperature; real-time operating parameters of battery pack power units such as total voltage, total current, cell temperature, and total internal resistance; and real-time operating parameters of fuel cell power units such as stack voltage, stack current, stack temperature, and hydrogen flow rate. The acquisition frequency for these data is generally in the range of 1-100 Hz. Ship motion status data includes ship speed, heading (including COG and HDG), and position parameters (including latitude and longitude information) collected through global positioning systems (such as GPS or BeiDou) and gyrocompasses, as well as motion attitude parameters collected through inertial measurement units (IMUs). The motion attitude parameters specifically include the ship's roll, pitch, and heave parameters, which are calculated from the three-axis acceleration and three-axis angular velocity in the raw data collected by the inertial measurement units. The data acquisition frequency of the above-mentioned ship speed, rudder angle, and position parameters is generally in the range of 1-10Hz, while the data acquisition frequency of motion attitude parameters is generally in the range of 50-100Hz.

[0051] Data cleaning and synchronization alignment are performed on multi-source data to obtain a multi-dimensional synchronized data vector. The data cleaning process specifically includes outlier detection and missing value handling. After synchronization alignment, the data is smoothed (e.g., using a Savitzky-Golay filter to remove high-frequency noise).

[0052] Furthermore, the synchronization alignment process specifically includes the following steps: receiving data packets of multi-source data, including sea state environmental data, ship dynamic status data, and ship motion status data, after data cleaning and processing, and marking them with a system timestamp with nanosecond precision; establishing corresponding processing buffers according to the acquisition frequency of different data streams to manage data streams of different frequencies; and using interpolation and extrapolation methods to uniformly align all data to the synchronization clock, forming a multi-dimensional synchronization data vector to solve the spatiotemporal inconsistency problem of multi-source data and ensure that the data processed subsequently is based on a unified time reference.

[0053] Extract the data vectors corresponding to the sea state environment data and ship motion state data from the multidimensional synchronous data vector, perform fusion correction processing to obtain corrected wave feature data, generate a quantified sea state level based on the corrected wave feature data, update the multidimensional synchronous data vector, and record the quantified sea state level into the multidimensional synchronous data vector.

[0054] Furthermore, the fusion correction process specifically includes the following steps:

[0055] The vector data corresponding to the roll, pitch, and heave parameters of the ship's motion state data are extracted from the multidimensional synchronous data vector, filtered to remove non-wave-induced motion during the ship's navigation, and the corresponding motion energy spectrum is obtained through spectral analysis (such as fast Fourier transform).

[0056] The inferred wave direction spectrum is obtained by inverting the ship's motion response function (RAO), and the corresponding inferred wave characteristic data, including significant wave height, mean wave period, and main wave direction, are extracted from it. It should be noted that the ship's motion response function (RAO) corresponds one-to-one with the ship itself and can be obtained through ship model experiments or CFD calculations.

[0057] The wave height, wave direction, and wave period in the inferred wave characteristic data and sea state environment data are fused using an adaptive Kalman filter algorithm or a weighted average algorithm to obtain corrected wave characteristic data.

[0058] It should be noted that the fusion process requires consideration of data reliability and confidence levels. The confidence level is determined dynamically based on the signal-to-noise ratio of the 3D wave radar data acquisition and the completeness of the ship motion state data. Furthermore, the determination of the quantified sea state level involves comparing the significant wave height in the corrected wave characteristic data with a standard sea state scale (such as the Douglas Sea State Table or the WMO Sea State Scale) to determine the corresponding quantified sea state level.

[0059] A multi-step prediction model is constructed based on deep learning algorithms. Multi-dimensional synchronous data vectors are input into the model to obtain predicted multi-dimensional synchronous data vectors for a future time period. The multi-step prediction model is trained on historical ship operation data. Its inputs include multi-source data from the current operating state, including sea state data, ship dynamic state data, and ship operational state data. It outputs corresponding predicted multi-dimensional synchronous data vectors. Specifically, in this embodiment, the multi-step prediction model is built using the LSTM deep learning algorithm framework and trained on historical ship operation data. It is applicable to multi-step (time step) prediction of multivariate time series. Given the widespread and mature application of the Long Short-Term Memory (LSTM) algorithm in capturing temporal dependencies between variables, its structure and principles will not be elaborated upon further here.

[0060] An operating cost model incorporating multi-objective optimization and a pre-defined strategy library containing several weighted coefficient vectors are constructed. The current dominant strategy is selected from the pre-defined strategy library. The operating cost model generates power control commands for optimal allocation and scheduling of hybrid power based on predicted multi-dimensional synchronous data vectors and the weighted coefficient vectors in the current dominant strategy. These power control commands include the target output power of the fuel cell power unit, battery pack power unit, and fuel cell power unit, which are time-sequential control command sequences within a future time domain.

[0061] Specifically, the operating cost model includes a total operating cost function:

[0062]

[0063] in, Fuel cost can be calculated using the fuel consumption rate, which is specifically obtained by calculating real-time operating status parameters collected by the fuel power unit. Its corresponding optimization objective is to minimize fuel consumption. Emission penalties can be obtained through statistical calculations of carbon dioxide emissions, nitrogen and sulfur emissions, and particulate emissions, with the corresponding optimization objective being to minimize operational emissions. The power error penalty can be calculated by squaring the difference between the required power and the actual power to penalize drastic power changes, ensuring that the hybrid power system meets the ship's propulsion requirements and responds quickly to load changes. Its corresponding optimization objective is to minimize power fluctuations. The comfort penalty can be obtained by estimating acceleration changes, vibration levels, and noise levels, specifically calculated using parameters such as engine speed and power. This aims to smooth power output, improve comfort and stability, and avoid shock loads; the corresponding optimization objective is to minimize shock changes. Furthermore, This is done to normalize the weighting coefficients.

[0064] Furthermore, the weight coefficient vector in the preset strategy library corresponds to the values ​​of multiple weight coefficients in the total operating cost function; that is, the weight coefficient vector includes the aforementioned weight coefficients. , , , This allows for dynamic adjustment of weighting coefficients based on the current dominant strategy. Specifically, the values ​​of the weighting coefficient vectors in the preset strategy library are adjusted according to factors such as the quantified sea state level, navigation position stage, task priority, equipment usage status, and regional environmental regulations. This results in a comprehensive, dynamic, and granular range of response strategies, making the scheduling and allocation of hybrid power more comprehensive, refined, and rational. For example, when the quantified sea state level is Douglas 0-3, it is considered calm sea state, and the weighting coefficients are significantly increased. Prioritize fuel efficiency optimization; when the quantified sea state level is Douglas 4-5, it is classified as a moderate sea state, and the weighting coefficient is increased accordingly. and To ensure that the dynamic response is acceptable despite some economic cost-efficiency loss; when the quantified sea state level is Douglas 6 or above, it is classified as a severe sea state, and the weighting coefficient is significantly increased in this case. and To ensure reliable power delivery under harsh sea conditions, safety and comfort are prioritized. Adaptive adjustments are made for each navigation position stage: when determining that the vessel is navigating in the open ocean, the weighting coefficient is significantly increased. To maximize fuel cost-effectiveness; when determining that the navigation is along the coast, increase the weighting coefficient. and To meet nearshore emission regulations while ensuring the ability to handle complex channel conditions; when determining that operations are being conducted upon arrival at port, the weighting factor is significantly increased. To meet the port's zero-emission requirements, the adaptive adjustment for task priority is as follows: when in the midst of a routine transport task, the weighting coefficient is significantly increased. To minimize transportation costs; and to significantly increase the weighting coefficient during emergency missions (such as search and rescue, medical evacuation, etc.). To ensure the response speed of the power system; when in the process of scientific research missions, the weighting coefficient is increased. and To ensure a stable environment for scientific research while minimizing pollution to the research area's ecological environment, the following adaptive adjustments are made to the equipment's operating status: When the ship's propulsion system is new, the weighting coefficients can be configured in a balanced manner to optimize performance; when the ship's propulsion system is under moderate wear or nearing maintenance, the weighting coefficients should be increased. And reduce the weighting coefficient This allows for moderate performance degradation to avoid overload and thus protect equipment safety. Adaptive adjustments for regional environmental regulations are as follows: when in an emission control area, the weighting factor is significantly increased. To meet stringent emission standards, such as switching to clean energy modes; when located in ecologically sensitive areas, the weighting coefficient should be increased. and This aims to reduce the impact of noise and emissions on sensitive ecosystems. It should be noted that the adaptive adjustments to the weighting coefficients described above are only examples from a pre-defined strategy library. In practical applications, appropriate control strategies can be added through self-learning or manual intervention. Furthermore, a reasonable operating range is set for each weighting coefficient to avoid system oscillations caused by extreme or abrupt changes, thereby improving the stability and reliability of the system.

[0065] A virtual digital twin model completely corresponding to the ship is constructed. Current multidimensional synchronous data vectors are input into the virtual digital twin model for initialization to simulate the ship's real-time state. Simultaneously, sea state environmental data from the predicted multidimensional synchronous data vectors are received to load and simulate predicted sea state environments for future time periods. Specifically, the virtual digital twin model includes a ship motion and resistance model, a power system dynamic model, and an environmental interaction model. The ship motion and resistance model is a high-precision hydrodynamic model built based on CFD calculations and empirical formulas. The power system dynamic model is built based on algorithms such as the universal characteristics of internal combustion engines (diesel engines), battery electrochemical models, and fuel cell efficiency curves. The environmental interaction model is built based on models of the impact of environmental loads such as waves, sea winds, and ocean currents on the ship, thus constructing a virtual digital twin model completely corresponding to the physical ship hull.

[0066] Based on a multi-agent reinforcement learning framework, a distributed collaborative model is constructed, which includes multiple agents deployed at various power units of the hybrid power system. Power control commands are transmitted sequentially through the operating cost model, the virtual digital twin model, and the distributed collaborative model.

[0067] Specifically, the virtual digital twin model receives power control commands generated by the operating cost model to run the simulation and record the simulation data, while simultaneously performing safety and performance evaluation on the simulation data. If an anomaly is detected, a correction suggestion is generated and fed back to the operating cost model, which then regenerates the corrected power control commands based on the suggestion. If no anomaly is detected, the power control commands are transmitted to the distributed collaborative model for execution. The safety and performance evaluation specifically includes close monitoring of the dynamic changes of key parameters in the simulation data, such as the load fluctuation trend of the engine in the fuel cell power unit, the change trajectory of the battery's SOC, and the operating status of the fuel cell. This is to detect and determine whether any safety hazards occur during the simulation, such as excessive engine torque operation, battery overcharging / over-discharging, excessive temperature, or power device overload, as well as whether the energy efficiency of the hybrid power system deviates from expectations, such as fuel economy assessment and emission level assessment.

[0068] Furthermore, the virtual digital twin model continuously compares the simulation data with the actual feedback data (real data detected by sensors) in the multidimensional synchronous data vector, and automatically adjusts the parameters of the virtual digital model using the recursive least squares method, thereby maintaining the consistency and high fidelity between the virtual digital twin model and the physical ship hull.

[0069] Specifically, in this embodiment, the distributed collaborative model includes a fuel-powered intelligent agent, a battery pack intelligent agent, and a fuel cell intelligent agent. The fuel-powered intelligent agent monitors parameters such as the rotational speed, torque, fuel flow rate, and exhaust temperature of the fuel-powered unit, and has the ability to regulate power output. Its corresponding learning and training optimization objectives include fuel efficiency, emission levels, and load stability. The battery pack intelligent agent monitors parameters such as battery SOC, voltage, current, cell temperature, and internal resistance, and has the ability to manage charging and discharging power and protect battery health. Its corresponding learning and training optimization objectives include SOC maintenance, cycle life, and charging and discharging efficiency. The fuel cell intelligent agent monitors and controls parameters such as hydrogen flow rate, stack voltage, and stack temperature, and has the ability to ensure the safe and stable operation of the stack. Its corresponding learning and training optimization objectives include hydrogen consumption optimization and lifespan protection. Through collaborative and complementary strategies among the various intelligent agents, goal conflicts and resource competition among multiple agents are effectively suppressed, enabling them to better adapt to the collaborative control and scheduling of various power units in complex marine environments.

[0070] Furthermore, the working modes of multiple agents in the distributed collaborative model include collaborative operation mode and autonomous operation mode:

[0071] In collaborative operation mode, multiple agents receive power control commands transmitted through a virtual digital twin model to adjust the target output power of the corresponding power units and update multi-source data. In this mode, the multiple agents corresponding to each power unit strictly follow a global optimization strategy to achieve overall optimal hybrid power scheduling. Furthermore, the multiple agents can perform small-scale optimizations of the power control commands within the compliant command framework through their internal control algorithms to further improve the accuracy and stability of power output control.

[0072] In autonomous operation mode, multiple agents autonomously generate locally optimal instructions based on their own reinforcement learning training framework to autonomously adjust the target output power of corresponding power units. Simultaneously, they update multi-source data to ensure the basic functionality of the hybrid power system and avoid affecting its stability. Understandably, in this mode, the locally optimal instructions generated autonomously by multiple agents have local action protection boundaries, preventing large-scale over-adjustments to avoid impacting the overall stability of the power system. Furthermore, this mode has certain triggering conditions; for example, it only triggers an abnormal response and automatically switches to autonomous operation mode when the system detects a sudden situation or when the virtual digital twin model responds untimely.

[0073] Furthermore, the distributed collaborative model is equipped with an anomaly identification and detection unit, which can detect abnormal situations such as sudden environmental changes (such as sudden severe sea conditions, extreme weather, etc.), equipment abnormalities (such as sensor failure, overheating and overload, etc.), and communication interruptions. It can switch multiple agents to autonomous operation mode, coordinate information exchange and data adjustment between multiple agents, and automatically switch to collaborative operation mode after the abnormal state is eliminated.

[0074] Furthermore, refer to Figure 2 As shown, the present invention also provides a ship hybrid power dispatching system based on digital twins and distributed collaboration, which specifically includes:

[0075] The data acquisition module is used to collect multi-source data, including sea state environmental data, ship dynamic status data, and ship motion status data. Specifically, the data acquisition module includes a sea state environmental data acquisition unit (such as 3D wave radar, meteorological sensors, etc.), a ship dynamic status data acquisition unit (such as sensors inside the power system, etc.), and a ship motion status data acquisition unit (such as GPS, gyrocompass, inertial measurement sensors, etc.) to collect sea state environmental data, ship dynamic status data, and ship motion status data.

[0076] The data processing module is used to perform data cleaning, synchronization alignment, and fusion correction on the acquired multi-source data to obtain a multi-dimensional synchronized data vector.

[0077] The central decision-making module receives multi-dimensional synchronous data vectors and generates power control commands for optimal hybrid power allocation. This module includes a multi-step prediction unit, a multi-objective optimization unit, and an expert strategy library unit. The multi-step prediction unit uses a deep learning-based multi-step prediction model to obtain predicted multi-dimensional synchronous data vectors for future time periods based on the current multi-dimensional synchronous data vector. The multi-objective optimization unit uses a total operating cost function that incorporates multi-objective optimization to adjust and optimize multiple objectives, including fuel cost, emission penalties, power error penalties, and comfort penalties, to obtain the optimal power scheduling scheme with the minimum total operating cost. The expert strategy library unit uses several weighted coefficient vectors to dynamically and adaptively adjust multiple weighted coefficients in the total operating cost function, resulting in a power scheduling allocation scheme that better suits the current environment and conditions. These weighted coefficient vectors are adjusted based on factors such as quantified sea state level, navigation position stage, mission priority, equipment usage status, and regional environmental regulations, forming comprehensive, dynamic, and granular response strategies that make hybrid power scheduling allocation more comprehensive, refined, and rational.

[0078] The digital twin module is used to construct a virtual digital twin model corresponding to the ship and receive power control commands from the central decision-making module. It simulates the decision-making results to determine the feasibility of the power control commands. In other words, the power control commands sent by the central decision-making module do not directly act on the ship's physical power system, but are pre-simulated in the virtual digital twin model to determine whether the power control commands are reasonable, compliant, and achieve the preset ideal effects, thereby improving the safety and reliability of the overall system decision-making.

[0079] A distributed collaborative module is used to construct a distributed collaborative model comprising multiple agents deployed at various power units of the hybrid power system. This model receives power control commands sent via the digital twin module or autonomously generates locally optimal commands. All agents are built upon a multi-agent reinforcement learning framework. Specifically, in this embodiment, the distributed collaborative model includes a fuel cell agent, a battery pack agent, and a fuel cell agent. By setting up a distributed architecture with multiple agents, system crashes caused by single-point failures can be avoided. This ensures that basic operational capabilities are maintained even when the central decision-making module or digital twin module fails, providing redundant safety control guarantees. While maintaining central global optimization, it also features rapid local response, effectively solving the inherent defects of traditional single-central control architectures, such as slow response and insufficient robustness, in complex marine environments.

[0080] In summary, the ship hybrid power scheduling method and system based on digital twins and distributed collaboration provided by this invention increases the safety and reliability of decision execution by performing simulation operation and safety verification before decision execution through a digital twin model. At the same time, the distributed collaboration model enables each power unit to have local autonomous decision-making capabilities to cope with sudden extreme situations, thereby improving the robustness and fault tolerance of ship hybrid power scheduling in dynamic marine environments.

[0081] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. All equivalent changes and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A ship hybrid power scheduling method based on digital twins and distributed collaboration, characterized in that, Includes the following steps: Collect and acquire multi-source data including sea state and environmental data, ship dynamic status data and ship motion status data, and perform data cleaning and synchronization alignment on the multi-source data to obtain a multi-dimensional synchronized data vector. Extract sea state and ship motion data from the multidimensional synchronous data vector, perform fusion and correction processing to obtain corrected wave feature data, generate a quantized sea state level based on the corrected wave feature data, and update the multidimensional synchronous data vector. A multi-step prediction model is constructed based on deep learning algorithms. Multi-dimensional synchronous data vectors are input into the multi-step prediction model to obtain predicted multi-dimensional synchronous data vectors for future time periods. An operating cost model including multi-objective optimization and a preset strategy library including several weight coefficient vectors are constructed. The operating cost model generates power control commands for optimal allocation of hybrid power based on the predicted multidimensional synchronous data vector and the weight coefficient vector. Construct a virtual digital twin model corresponding to the ship, initialize it by inputting a multidimensional synchronous data vector, and receive a predicted multidimensional synchronous data vector to load the predicted sea state environment for future time periods; A distributed collaborative model is constructed based on a multi-agent reinforcement learning framework, comprising multiple agents deployed at various power units of the hybrid power system. The power control commands are transmitted sequentially through the operating cost model, the virtual digital twin model, and the distributed collaborative model.

2. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 1, characterized in that: The sea environment data includes wave height, wave direction, wave period, wind direction, and wind speed data. The ship's power status data includes real-time operating status data of the fuel power unit, battery power unit, and fuel cell power unit. The ship's motion status data includes sailing speed, sailing course, roll parameters, pitch parameters, and heave parameters data.

3. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 1, characterized in that, The synchronization alignment process includes the following steps: Receive multi-source data after data cleaning and processing, and mark it with a high-precision system timestamp; Establish corresponding processing buffers based on different data stream acquisition frequencies to manage data streams of different frequencies; Interpolation and extrapolation methods are used to align all data to the synchronous clock, generating a multi-dimensional synchronous data vector.

4. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 2, characterized in that, The fusion correction process includes the following steps: Extract the vector data corresponding to the roll, pitch, and heave parameters from the multidimensional synchronous data vector, and obtain the corresponding kinematic energy spectrum through spectral analysis; The inferred wave direction spectrum is obtained by inverting the ship motion response function, and the inferred wave characteristic data including significant wave height, average wave period, and main wave direction are extracted from it. An adaptive Kalman filter algorithm is used to fuse the inferred wave feature data and the sea state environment data to obtain corrected wave feature data.

5. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 1, characterized in that: The operating cost model includes a total operating cost function: ； in, For fuel costs, As a penalty for emissions, As a penalty for power error, As a comfort penalty, , , , These are the corresponding weight coefficients, and ; The weight coefficient vector consists of weight coefficients , , , The composition and weighting coefficients are adjusted based on the quantified sea state level, navigation position and stage, mission priority, equipment usage status, and regional environmental regulations.

6. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 1, characterized in that: The virtual digital twin model receives the power control commands generated by the operating cost model, runs the simulation, and records the simulation data. At the same time, it performs safety and performance evaluation and testing on the simulation data. If an anomaly is detected during the assessment, a correction suggestion is generated and fed back to the operating cost model. The operating cost model then regenerates the corrected power control command based on the correction suggestion. If no abnormalities are found in the assessment, the power control command will be passed to the distributed cooperative model for execution.

7. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 2, characterized in that: The virtual digital twin model includes a ship motion and resistance model, a dynamic model of the power system, and an environmental interaction model; The distributed collaborative model includes a fuel-powered intelligent agent, a battery pack intelligent agent, and a fuel cell intelligent agent.

8. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 1, characterized in that: The working modes of multiple agents in the distributed collaborative model include a collaborative operation mode and an autonomous operation mode. In collaborative operation mode, multiple intelligent agents receive power control commands sent through a virtual digital twin model to adjust the target output power of the corresponding power units; In autonomous operation mode, multiple agents autonomously generate locally optimal instructions based on their own reinforcement learning training framework in order to autonomously adjust the target output power of the corresponding power unit.

9. The ship hybrid power scheduling method based on digital twin and distributed collaboration according to claim 8, characterized in that: The distributed collaborative model is equipped with an anomaly identification and detection unit to detect abnormal situations, including environmental changes, equipment malfunctions, and communication interruptions, and to switch multiple agents between collaborative operation mode and autonomous operation mode based on the detection results.

10. A ship hybrid power dispatching system based on digital twins and distributed collaboration, characterized in that, include: The data acquisition module is used to collect and acquire multi-source data, including sea state and environmental data, ship dynamic status data, and ship motion status data. The data processing module is used to perform data cleaning, synchronization alignment, and fusion correction on the acquired multi-source data to obtain a multi-dimensional synchronized data vector. The central decision-making module is used to receive multi-dimensional synchronous data vectors and generate power control commands for optimal allocation of hybrid power. The central decision-making module includes a multi-step prediction unit with a multi-step prediction model, a multi-objective optimization unit with an operating cost model, and an expert strategy library unit with weight coefficient vectors. The digital twin module is used to construct a virtual digital twin model corresponding to the ship and receive power control commands issued by the central decision-making module to simulate the decision-making results and judge the feasibility of the power control commands based on the simulation results. The distributed coordination module is used to build a distributed coordination model that includes multiple intelligent agents deployed at various power units of the hybrid power system, in order to receive power control commands sent through the digital twin module or autonomously generate locally optimal commands.

Images