A method for switching fuel modes in dual-fuel hybrid ships based on NSDP and LSTM

By employing an adaptive switching control method combining NSDP and LSTM, and integrating intelligent network data for feedforward prediction, the problem of fuel mode switching for dual-fuel vessels under complex sea conditions was solved, achieving smooth and efficient fuel mode switching and improving the robustness and economy of the system.

CN123104881APending Publication Date: 2026-11-17HUANGGANG POLYTECHNIC COLLEGE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610980593.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-02
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to cope with strong disturbances and time-varying operating conditions during fuel mode switching in dual-fuel ships, and cannot achieve real-time collaborative optimization of multiple objectives. Furthermore, the control methods lack robustness under complex sea conditions, resulting in an unstable switching process and limited economic improvement.

Method used

An adaptive switching control method based on NSDP and LSTM is adopted, which integrates real-time navigation information, generates Pareto optimal switching strategy through NSDP multi-objective optimization, and uses a pre-trained LSTM network for real-time control. Combined with intelligent connected data for feedforward prediction, it achieves global optimal decision-making and online inference.

Benefits of technology

It enables smooth and efficient switching of fuel modes for dual-fuel ships under complex sea conditions, improves the system's anti-disturbance capability and operating condition adaptability, solves the shortcomings of traditional methods in multi-objective optimization and real-time performance, and improves the economy and environmental friendliness of the switching process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104881A_ABST
    Figure CN123104881A_ABST
Patent Text Reader

Abstract

This invention discloses a fuel mode switching method for dual-fuel hybrid electric vessels based on NSDP and LSTM. The method includes collecting real-time navigation information from the dual-fuel hybrid electric vessel, extracting several initial feature signals from the real-time navigation information, fusing all initial feature signals to obtain spatiotemporal fusion features, inputting the spatiotemporal fusion features into a pre-trained LSTM deep learning network, and outputting adjustment coefficients. The pre-trained LSTM deep learning network is trained using a Pareto optimal switching strategy set and a real-ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by an NSDP-based fuel mode switching multi-objective optimization model. Based on the adjustment coefficients, the fuel mode of the dual-fuel hybrid electric vessel is switched. This invention integrates real-time navigation information to construct an adaptive switching control method that combines NSDP multi-objective optimization and LSTM deep learning, which has significant theoretical and engineering value for improving the operational quality and economy of dual-fuel vessels.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of marine new energy power system control technology and green shipping energy conservation and emission reduction technology, specifically to a method for switching fuel modes of dual-fuel hybrid ships based on NSDP and LSTM. Background Technology

[0002] Currently, extensive research has been conducted on fuel mode switching and energy management control for hybrid-powered ships. Based on the controlled object, this research can be broadly categorized into two types: hybrid electric vehicle energy management and ship hybrid power system mode switching. Based on the technical approach, it can be divided into four categories: rule-based mode switching control, energy allocation based on optimization algorithms, adaptive control based on learning models, and cooperative control in intelligent connected environments. Existing technologies have formed relatively mature theoretical and methodological systems in areas such as power allocation in hybrid power systems, mode switching smoothness, and fuel economy optimization. However, significant shortcomings remain in the engineering applications of dual-fuel ships, strongly disturbed sea states, multi-objective real-time cooperative optimization, and the integration of intelligent connectivity and deep learning.

[0003] In the research on hybrid power system mode switching and energy management, existing achievements mainly focus on land-based equipment such as hybrid electric vehicles and construction machinery, with emphasis on multi-objective optimization of fuel economy, emissions, power, and battery life. Existing optimization methods are poorly adaptable to strong disturbances, time-varying operating conditions, and real-time changes in the external environment, making it difficult to directly transfer to moving vehicles such as ships that operate under long-term, variable, and highly disturbed conditions.

[0004] In terms of coordinated control during mode switching, existing technologies mainly fall into three categories: rule-based threshold switching, optimization-based trajectory planning, and model prediction-based real-time control. Rule-based methods are logically simple and easy to implement in engineering, but they rely on manual calibration and cannot adapt to changes in sea state, load, and fuel characteristics. The switching process is prone to problems such as speed overshoot, excessive vibration, and response lag. Optimization algorithms such as dynamic programming and particle swarm optimization can obtain the globally optimal switching trajectory, but they involve large computational loads, cannot be run online, and do not consider uncertainties caused by environmental disturbances. Model prediction control and sliding mode control methods have a certain degree of anti-interference capability, but they rely on accurate models and lack robustness under changes in dual-fuel combustion characteristics and random sea state disturbances, making it difficult to simultaneously guarantee stability, economy, and emission performance.

[0005] In the area of ​​dual-fuel engine mode switching control, existing research mainly focuses on combustion and injection control of gas-diesel and methanol-diesel dual-fuel engines, emphasizing fuel substitution rate, idling stability, emission optimization, and dynamic response. Studies show that dual-fuel engines are prone to torque fluctuations, speed oscillations, and combustion fluctuations at mode switching points, further exacerbating the switching shock. However, most existing controls only focus on the engine itself, failing to coordinate with the ship's propulsion system, load characteristics, and navigation environment. They also fail to integrate the switching process with energy management, battery life, and navigation safety for unified multi-objective optimization, resulting in frequent switching, poor ride smoothness, and limited economic improvement during actual navigation.

[0006] In the field of marine hybrid power system control, existing research mainly focuses on steady-state energy management, emphasizing the optimization of fuel consumption, emissions, and battery state of charge (SOC) maintenance. Research on the dynamic coordination of transient mode switching processes is relatively weak. Most ships employ simple threshold switching strategies, determining fuel mode and power source allocation based on load, speed, and battery capacity, without refined design of switching timing, torque coordination, and speed synchronization. In recent years, some studies have introduced reinforcement learning and LSTM methods into ship energy management, but most rely solely on historical data learning, failing to integrate real-time environmental information and lacking perception and prediction of wave, wind speed, traffic flow, and channel information. Consequently, their adaptive capabilities and robustness are insufficient to meet the demands of complex navigation. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM. By integrating real-time navigation information, an adaptive switching control method integrating NSDP multi-objective optimization and LSTM deep learning is constructed, which has important theoretical and engineering value for improving the operational quality and economy of dual-fuel ships.

[0008] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0009] According to a first aspect of this application, a method for switching fuel modes in a dual-fuel hybrid ship based on NSDP and LSTM is provided, comprising: Real-time navigation information of dual-fuel hybrid ships is collected, several initial feature signals are extracted from the real-time navigation information, and all initial feature signals are fused to obtain spatiotemporal fusion features. The spatiotemporal fusion features are input into a pre-trained LSTM deep learning network, and the output adjustment coefficients are obtained. The pre-trained LSTM deep learning network is trained by a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP. Based on the aforementioned adjustment coefficient, the fuel mode of the dual-fuel hybrid ship is switched.

[0010] In some embodiments of this application, based on the foregoing scheme, the step of denoising the initial feature signals using independent component analysis before fusing all initial feature signals includes: The initial feature signal is decomposed into several intrinsic mode components and a residual component by using ensemble empirical modes; Calculate the correlation coefficient between each intrinsic mode component and the initial characteristic signal one by one, and determine whether the intrinsic mode component is a noise component or a useful component based on the correlation coefficient; The noise components are purified using an independent component analysis algorithm; The initial characteristic signal of the denoised signal is obtained by reconstructing the signal from all purified noise components and all useful components.

[0011] In some embodiments of this application, based on the foregoing scheme, the purification of noise components using an independent component analysis algorithm includes: The noise component is represented as a combination of a noiseless speed signal and an unknown hybrid matrix. The combination form is preprocessed by centering and whitening to obtain one or more independent components; The signal separation vector is obtained by iteratively calculating for each independent component. After orthogonalizing and normalizing the signal separation vector, it is multiplied with the noise component to obtain the purified noise component.

[0012] In some embodiments of this application, based on the foregoing scheme, the real-time navigation information includes sea state information, traffic flow information, power operation information, and navigation information; The sea state information includes real-time wind speed, wind direction, significant wave height, water current speed and direction, air temperature, and water temperature. The traffic flow information includes the vessel's position, speed, course, distance to surrounding vessels, speed, course, nearest encounter distance, and channel congestion. The power operation information includes engine speed, output torque, coolant temperature, exhaust temperature, battery state of charge, diesel injection quantity, gas injection quantity, and mode switching command. The navigation information includes route type, speed limit, route gradient, estimated arrival time, and information on the division of idling zone / acceleration zone / cruising zone.

[0013] In some embodiments of this application, based on the aforementioned scheme, the spatiotemporal fusion features are input into a pre-trained LSTM deep learning network. The spatiotemporal fusion features include engine speed, battery state of charge, ship speed, real-time wind speed, significant wave height, distance to surrounding ships, and course gradient. Output adjustment coefficients, which include diesel injection adjustment coefficients and clean energy injection adjustment coefficients.

[0014] In some embodiments of this application, based on the aforementioned scheme, the LSTM deep learning network includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer, and an output layer connected in sequence. The input layer is used to output spatiotemporal fusion features, the first LSTM layer includes 128 neurons, the second LSTM layer includes 64 neurons, the fully connected layer includes 32 neurons, and the output layer is used to output adjustment coefficients.

[0015] In some embodiments of this application, based on the foregoing scheme, the method for generating the Pareto optimal switching strategy set includes: Construct a ternary minimization objective function:

[0016] In the formula, The stability index for the switching process is the speed overshoot rate; The economic indicators for the voyage are fuel consumption per unit distance. The environmental performance indicators for the navigation process are nitrogen oxide emission concentrations. The formula for calculating the stability index is:

[0017] in, Let the stability objective function be... For speed overshoot; This represents the engine's maximum speed during the switching process. The target rotational speed; The formula for calculating the economic indicators is as follows:

[0018] In the formula, Total fuel consumption per unit distance; This refers to diesel fuel consumption. For clean energy consumption, This is the calorific value conversion factor. This refers to the total distance traveled by the ship. The formula for calculating the environmental protection index is as follows:

[0019] In the formula, NO for mode switching process x Total emissions; For real-time NO x Emission concentration, To switch the start time, The time when the switch is completed; Establish constraints: Speed ​​constraint:

[0020] In the formula, This represents the change in engine speed; This is the upper limit of the rate of change of rotational speed, meaning that the increase or decrease in rotational speed per second must not exceed 30 revolutions per minute; Injection volume constraint: ,

[0021] In the formula, This refers to the real-time injection volume of the diesel injection actuator; This is the maximum permissible injection quantity for the diesel injection system; As a clean energy source, it represents the real-time injection volume of the injection actuator; This represents the maximum permissible injection volume for a clean energy injection system. SOC security constraints:

[0022] In the formula, State of charge of the ship's energy storage batteries; Switching time constraints:

[0023] In the formula, This represents the total duration of the fuel mode switch. Preset time; By time step Discretize the switching process into N decision stages: Define state variables:

[0024] In the formula, Let k be the system state vector at stage k. Let be the engine speed in the k-th stage; This represents the battery state of charge in the k-th stage. Let be the ship's speed in stage k; Let be the real-time wind speed at stage k; Define control variables:

[0025] In the formula, This is the control input vector for the k-th stage; This refers to the diesel injection coefficient; The gas (clean energy) injection coefficient; State transition equations are established based on engine dynamics and battery models:

[0026] In the formula, Let be the system state vector for the (k+1)th discrete decision stage; Let be the system state vector for the k-th discrete decision stage; This is the control input vector for the k-th discrete decision-making stage; Starting from the final stage, we work backward to calculate the non-dominated solutions for each stage, and retain the uniformly distributed Pareto optimal switching strategy set by sorting by congestion.

[0027] According to a second aspect of this application, a dual-fuel hybrid power ship fuel mode switching system based on NSDP and LSTM is provided, comprising: The first acquisition module is used to collect real-time navigation information of dual-fuel hybrid ships, extract several initial feature signals from the real-time navigation information, and fuse all the initial feature signals to obtain spatiotemporal fusion features. The second acquisition module is used to input the spatiotemporal fusion features into a pre-trained LSTM deep learning network and output adjustment coefficients. The pre-trained LSTM deep learning network is trained through a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP. The switching module is used to switch the fuel mode of the dual-fuel hybrid ship based on the adjustment coefficient.

[0028] According to a third aspect of this application, a computer-readable storage medium is provided that stores a computer program thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0029] According to a fourth aspect of this application, an electronic device is provided, comprising: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, cause the one or more processors to implement the method described above.

[0030] The beneficial effects of this application are as follows: (1) The fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM provided in this application breaks through the problem of "steady-state energy management as the main focus and insufficient transient switching control" in existing ship hybrid power systems, and realizes an integrated architecture of intelligent network perception, NSDP multi-objective optimization and LSTM real-time control. It not only relies on NSDP to complete multi-objective global optimal decision-making and generate Pareto optimal switching strategies that take into account stability, economy and emissions; but also realizes millisecond-level online inference and adaptive tracking based on LSTM; at the same time, it introduces intelligent network data to form feedforward prediction and builds a three-level collaborative mechanism of "global optimal - real-time control - operating condition prediction". This architecture upgrades from passive feedback to active optimal decision-making, solves the industry bottleneck that traditional rule control, PID control and single optimization algorithm cannot take into account both optimality and real-time performance, and is more suitable for the complex, strong disturbance and long time delay operation scenarios of dual-fuel ships.

[0031] (2) The fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM provided in this application proposes a multi-objective dynamic programming solution method for NSDP to achieve global optimum of the switching strategy. The non-dominated sorting dynamic programming (NSDP) system is systematically applied to the fuel mode switching problem of dual-fuel ships, and a method is established based on switching stability, fuel consumption, NO x The ternary multi-objective optimization model with emissions as the objective overcomes the limitations of traditional weighted methods in handling multi-objective conflicts. NSDP directly generates a Pareto optimal solution set through inverse recursion and non-dominated sorting, automatically selecting the optimal switching point, fuel allocation ratio, and switching rate under different operating conditions. Compared to methods such as dynamic programming, particle swarm optimization, and equivalent fuel minimization, NSDP requires no manual weight setting and does not rely on approximate models. It can obtain the global optimal solution under strong constraints, multivariates, and nonlinear conditions, significantly improving the economy, environmental friendliness, and stability of the switching process.

[0032] (3) This application provides a fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM. It proposes an NSDP-guided LSTM real-time adaptive control method and a two-stage control mode of "offline optimal solution and online deep learning" to solve the problems of traditional optimization algorithms being unable to run online and the insufficient accuracy of simple neural networks. The optimal policy generated by NSDP is used as a label to train the LSTM network, enabling the model to directly learn the optimal time-series control law, which has strong generalization ability and high control accuracy. LSTM can effectively capture the time-series dynamic characteristics of ships under large inertia, long time delay, and strong disturbances. Combined with the gating mechanism to suppress gradient vanishing, it can achieve rapid adaptive adjustment under speed, load, and sea state disturbances. Compared with existing methods based on rules, model prediction, and reinforcement learning, this method has a faster response, stronger robustness, and simpler calibration. It can be directly deployed on shipboard ECUs and better meets engineering requirements.

[0033] (4) The fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM provided in this application realizes advanced prediction of operating conditions and integrated adaptive adjustment of feedforward and feedback. It deeply integrates intelligent network technology into mode switching control, and realizes advanced prediction of operating conditions in the next 5 seconds through multi-source information such as AIS, Beidou, meteorology, route, and traffic flow, and constructs a control mechanism that combines feedforward compensation and feedback correction. The controller can adjust the switching timing and fuel injection amount in advance according to the route, slope, wind and waves, and traffic conditions, so as to avoid speed fluctuations and shocks caused by disturbances from the root. Compared with traditional feedback control without perception and prediction, this method greatly improves the system's anti-disturbance capability and operating condition adaptability, solves the problems of lag, large fluctuations, and frequent switching of ships in complex sea conditions, and upgrades mode switching from "post-correction" to "global pre-optimization".

[0034] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. Attached Figure Description

[0035] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and are intended to explain the invention, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for switching fuel modes in a dual-fuel hybrid ship based on NSDP and LSTM according to the present invention. Figure 2 This is a flowchart of the initial feature signal denoising process of the present invention; Figure 3 This is a framework diagram of a dual-fuel hybrid power ship fuel mode switching method based on NSDP and LSTM according to the present invention. Figure 4This is a schematic diagram illustrating the expected engine speed under three operating conditions in a specific embodiment of the present invention; Figure 5 This is a flowchart of the overall five-layer closed-loop architecture of the present invention; Figure 6 The flowchart of the adaptive control strategy for fuel mode switching of a dual-fuel hybrid ship based on NSDP and LSTM is shown below. Figure 7(a) is a schematic diagram of the fuel injection quantity based on PID control in a specific embodiment of the present invention; Figure 7(b) is a schematic diagram of the fuel injection quantity based on the rule-based switching strategy in a specific embodiment of the present invention; Figure 7(c) is a schematic diagram of the fuel injection quantity based on the strategy proposed in this invention in a specific embodiment of the invention; Figure 8(a) is a schematic diagram of the jet volume based on PID control in a specific embodiment of the present invention; Figure 8(b) is a schematic diagram of the jet volume of a rule-based switching strategy in a specific embodiment of the present invention; Figure 8(c) is a schematic diagram of the jet volume based on the strategy proposed in this invention in a specific embodiment of the invention; Figure 9 This is a comparison diagram of vibration amplitude under three different control strategy modes in a specific embodiment of the present invention; Figure 10 This is a schematic diagram of a dual-fuel hybrid power ship fuel mode switching system based on NSDP and LSTM according to the present invention. Figure 11 This is a schematic diagram of an electronic device. Detailed Implementation

[0036] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0037] It should be understood that the terms "comprising" and other similar expressions in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, or apparatus that includes a series of steps or units and is not limited to the listed steps or units. Furthermore, "first" and "second" are used to distinguish different objects and are not intended to describe a specific order.

[0038] According to the first aspect of this application, Figure 1 , Figure 3 and Figure 6 As shown, this embodiment provides a fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM. The method is divided into five main parts: an intelligent network sensing layer, a data fusion and preprocessing layer, a multi-objective optimization decision layer, a deep learning real-time control layer, and an execution and feedback layer. These layers work collaboratively to form a closed-loop adaptive control system. This embodiment of the fuel mode switching method for dual-fuel hybrid ships based on NSDP and LSTM includes: Step S1: Collect real-time navigation information of the dual-fuel hybrid power ship, extract several initial feature signals from the real-time navigation information, and fuse all the initial feature signals to obtain spatiotemporal fusion features.

[0039] Intelligent connectivity technology is a prerequisite for achieving accurate prediction and adaptive switching of ship operating conditions. This paper adopts a fusion data acquisition architecture combining shipborne IoT, BeiDou, AIS, and 5G / satellite communication.

[0040] In some embodiments of this example, the real-time navigation information includes sea state information, traffic flow information, power operation information, and navigation information. Initial feature signals of sea state, traffic flow, power operation, and navigation information are extracted from the real-time navigation information.

[0041] like Figure 5 As shown, the intelligent connected perception layer is responsible for collecting data in all dimensions and is the input foundation of the entire control system.

[0042] Environmental sensing unit: includes shipborne meteorological sensors, wave and current sensors, and visibility monitoring equipment, which acquire sea condition information such as wind speed, wind direction, significant wave height, current speed and direction, air temperature, and water temperature in real time.

[0043] Traffic sensing unit: It acquires traffic flow information such as the ship's position, speed, course, the position, speed, course, nearest encounter distance, and channel congestion through AIS, Beidou positioning, radar and cameras.

[0044] Ship Status Awareness Unit: Includes engine speed sensor, torque sensor, water temperature / exhaust temperature sensor, diesel / clean energy injection flow meter, battery SOC sensor, throttle opening sensor, etc., to realize full parameter monitoring of power system operating status such as engine speed, output torque, coolant temperature, exhaust temperature, SOC, diesel injection quantity, gas injection quantity, mode switching command, etc.

[0045] The navigation and route awareness unit (ROA) acquires prior information such as route type, speed limits, gradient, estimated arrival time, and idling / acceleration / cruising zone divisions based on the Electronic Chart Display and Information System (ECDIS) and route planning system. The data sampling frequency of the awareness layer is uniformly set at 10-20Hz, and timestamp alignment is used to ensure timing consistency. All awareness devices employ a unified timestamp synchronization mechanism, using the ship's master clock as a reference to eliminate time differences between multiple devices; the data format is uniformly converted to standard JSON format to ensure consistency in transmission and parsing.

[0046] Sensing data is aggregated to the edge computing unit via the shipboard local area network (CAN bus), and then uploaded to the data fusion layer via 5G / satellite communication. 5G is prioritized in near-shore waters, while satellite communication is automatically switched in offshore waters, with transmission latency controlled within 100ms. Furthermore, the raw sensor data is filtered using a Kalman filter algorithm to remove high-frequency random noise and environmental interference, retaining effective operating condition characteristic signals.

[0047] To address issues such as inconsistent sampling frequencies, transmission delays, and noise interference in real-time navigation information, such as... Figure 5 As shown, a data fusion and preprocessing layer is set up, and a spatiotemporal joint fusion strategy is adopted. Through the data fusion and preprocessing layer, real-time navigation information is denoised, synchronized, normalized and feature fused, and noise, delay and dimensional differences are eliminated.

[0048] Time alignment: Unify sampling time, perform linear interpolation on low-frequency data, perform moving average downsampling on high-frequency data, and use the moving window mean method to fill in missing values ​​for lost data during transmission; complete the time alignment of all dimensions of data based on timestamp.

[0049] The Min-Max normalization method is used to map all feature data to the [0,1] interval, eliminating the difference between units and numerical magnitudes. The calculation formula is as follows:

[0050] In the formula, These are the normalized eigenvalues; These are the maximum and minimum values ​​of this feature in the entire dataset; This is the original feature data.

[0051] Spatial feature fusion: Constructing a spatiotemporal feature matrix:

[0052] in, for Spatiotemporal feature matrix at any given time; for Initial sea state characteristic signal at time, for Initial characteristic signal of traffic flow at any time for Initial characteristic signal of dynamic operation at any time, for Initial characteristic signal for navigation at any time.

[0053] The 3σ criterion and sliding window detection are used to eliminate abrupt changes and outliers, ensuring input reliability.

[0054] In some implementations of this embodiment, such as Figure 2 As shown, independent component analysis is used to denoise the initial feature signal, resulting in denoised features, including: The initial characteristic signal is decomposed into several intrinsic mode components and a residual component using Ensemble Empirical Mode Decomposition (EEMD).

[0055] In the formula, The first decomposition obtained One Intrinsic Modal Component (IMF); The data sampling time point; For the decomposed first One residual component; The normalization factor is denoted by IMF, which stands for Intrinsic Mode Function. For the first One initial characteristic signal; Calculate the correlation coefficient between each intrinsic mode component and the initial characteristic signal. Determine whether the intrinsic mode component is a noise component or a useful component based on the correlation coefficient. The formula for calculating the correlation coefficient is: =

[0056] In the formula, For components With the initial characteristic signal The correlation coefficient; To be consistent with the initial characteristic signal The average value; For the first The average value of each intrinsic modal component; Sampling time; The noise components are purified using an independent component analysis algorithm; The denoising features are obtained by reconstructing the signal from all purified noise components and all useful components.

[0057] In this embodiment, the components are classified according to the magnitude of the correlation coefficient to distinguish between noise and valid information: Correlation coefficient 0–0.3: slight correlation, determined to be a noise component; Correlation coefficient 0.3–0.5: Real correlation, determined to be a noise component; Correlation coefficient 0.5–0.8: Significantly correlated, and considered a useful component; Correlation coefficient 0.8–1.0: Highly correlated, identified as a useful component; Let the noise component be denoted as The useful component is denoted as .

[0058] In some embodiments of this example, the purification of noise components using the Independent Component Analysis (ICA) algorithm includes: noise components It is represented as a combination of a noiseless speed signal and an unknown hybrid matrix; The combination form is preprocessed by centering and whitening to obtain one or more independent components; The signal separation vector is obtained by iteratively calculating for each independent component. After orthogonalizing and normalizing the signal separation vector, it is multiplied with the noise component to obtain the purified noise component. , that is, the pure intrinsic mode components.

[0059] The noise components purified by independent component analysis algorithm With useful components Signal reconstruction is performed to provide high-precision, low-interference, and reliable data support for subsequent fuel mode switching control.

[0060] During the fuel operation mode switching process of dual-fuel vessels, it is necessary to collect and analyze multiple operating parameters in real time, including engine speed, battery state of charge, vessel speed, real-time wind speed, significant wave height, distance to surrounding vessels, and course gradient. If the on-site data acquisition devices and sensors malfunction, or if there are measurement deviations or insufficient data accuracy, it can easily lead to deviations in fuel mode switching control and a significant decrease in control precision. Therefore, this embodiment uses high-precision sensors to collect information in real time. For example, vessel speed is mainly controlled by engine speed; the lower the engine speed, the lower the actual vessel speed. This embodiment uses high-precision speed sensors to collect engine speed information in real time, dynamically monitor the operating conditions of the power system, and use speed parameters as an important constraint for smooth fuel mode switching. Engine speed can directly reflect the actual load, operating conditions, and overall performance of the unit, accurately reflecting speed fluctuations, operational stability, and abnormal operating condition changes, providing a reliable operating condition basis for adaptive fuel mode switching.

[0061] Step S2: Input the spatiotemporal fusion features into a pre-trained LSTM deep learning network and output adjustment coefficients. The pre-trained LSTM deep learning network is trained using a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP.

[0062] This embodiment presents an adaptive control method for fuel mode switching in dual-fuel ships based on NSDP and LSTM in an intelligent connected environment. The overall system adopts a five-layer closed-loop architecture of "perception-decision-optimization-control-execution". Driven by intelligent connected data, the system uses NSDP multi-objective optimization as the decision-making core and LSTM deep learning as the real-time control carrier to achieve smooth, efficient, and adaptive switching between pure diesel, dual-fuel hybrid, and pure clean energy modes for dual-fuel engines. Figure 3 The diagram shown is a framework diagram of a dual-fuel hybrid power ship fuel mode switching method based on NSDP and LSTM according to the present invention.

[0063] In some implementations of this embodiment, the dual-fuel ship mode switching is essentially a multi-objective, multi-constraint, and multi-stage dynamic decision-making problem. Traditional methods struggle to simultaneously optimize stability, economy, and emissions. Non-dominated sorting dynamic programming (NSDP), capable of handling multi-objective optimization problems, is widely used to generate globally optimal Pareto solutions, achieving synergistic optimization of fuel consumption and battery wear. Deep learning methods, such as Long Short-Term Memory (LSTM) networks, leverage their temporal feature learning capabilities to learn control strategies from optimal trajectories, enabling real-time online inference.

[0064] In some implementations of this embodiment, such as Figure 5 As shown, a real-time control layer for deep learning is configured. The LSTM deep learning network includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer, and an output layer connected in sequence. The input layer is used to output spatiotemporal fusion features. The first LSTM layer includes 128 neurons, the second LSTM layer includes 64 neurons, and the fully connected layer includes 32 neurons. The output layer is used to output adjustment coefficients.

[0065] Specifically, the input layer is used to input 7-dimensional spatiotemporal fusion features, including engine speed, state of charge (SOC), ship speed, wind speed, wave height, distance to surrounding vessels, and course gradient; the output layer is used to output 2-dimensional control quantities, including diesel injection adjustment coefficients. Clean energy injection adjustment coefficient .

[0066] The network structure of the LSTM deep learning network consists of an input layer, an LSTM layer (128 neurons), an LSTM layer (64 neurons), a fully connected layer (32 neurons), and an output layer. The activation functions are ReLU and Sigmoid, the optimizer is Adam, the learning rate is 0.001, and the loss function is MSE.

[0067] In the model training mechanism, the training set consists of a Pareto optimal switching strategy set (200 sets) and a real ship navigation strategy set (1000 sets), which are divided into training and validation sets in a 7:3 ratio.

[0068] Early stopping is used to prevent overfitting, and the sliding window updates the input sequence every 5 seconds to maintain the model's adaptability.

[0069] The LSTM incorporates a 5-second future operating condition prediction module, which adjusts the diesel injection regulation coefficient and the clean energy injection regulation coefficient in advance based on intelligent network prior information, thereby adjusting the injection quantity to achieve: early torque reduction, reduced switching impact, early fuel supply, shortened switching time, early entry into the target mode operating point, and improved stability.

[0070] In some implementations of this embodiment, such as Figure 5 As shown, a multi-objective optimization decision layer is set up, and NSDP (non-dominated ranking dynamic programming) is introduced for global optimal solution. The method for generating the Pareto optimal switching strategy set is as follows: Construct a ternary minimization objective function:

[0071] In the formula, For the stability index of the switching process (speed overshoot rate); This refers to the economic indicators of the voyage (fuel consumption per unit distance). This refers to environmental indicators (nitrogen oxide emission concentration) during navigation.

[0072]

[0073] in, Let the stability objective function be... For speed overshoot; This represents the engine's maximum speed during the switching process. The target rotational speed.

[0074] fuel consumption per unit distance (Economic efficiency)

[0075] In the formula, Total fuel consumption per unit distance; This refers to diesel fuel consumption. For clean energy consumption, This is the calorific value conversion factor. This represents the ship's total voyage distance.

[0076] Nitrogen oxide emissions (Environmentally friendly)

[0077] In the formula, NO for mode switching process x Total emissions; For real-time NO x Emission concentration, To switch the start time, This is the time when the switch is complete.

[0078] Establish constraints: Speed ​​constraint:

[0079] In the formula, This represents the change in engine speed; This is the upper limit of the rate of change of rotational speed, meaning that the speed increase or decrease per second must not exceed 30 revolutions per minute.

[0080] Injection volume constraint: ,

[0081] In the formula, This refers to the real-time injection volume of the diesel injection actuator; This is the maximum permissible injection quantity for the diesel injection system; Real-time injection volume for clean energy (LNG / methanol, etc.) injection actuators; This is the maximum permissible injection volume for a clean energy injection system.

[0082] SOC security constraints:

[0083] In the formula, This refers to the state of charge of the ship's energy storage batteries.

[0084] Switching time constraints:

[0085] In the formula, This represents the total duration of the fuel mode switch.

[0086] By time step The switching process is discretized into N decision stages.

[0087] Define state variables:

[0088] In the formula, Let k be the system state vector at stage k. Let be the engine speed in the k-th stage; This represents the battery state of charge in the k-th stage. Let be the ship's speed in stage k; Let be the real-time wind speed at stage k.

[0089] Define control variables:

[0090] In the formula, This is the control input vector for the k-th stage; This refers to the diesel injection coefficient; This is the gas (clean energy) injection coefficient.

[0091] State transition equations are established based on engine dynamics and battery models:

[0092] In the formula, Let be the system state vector for the (k+1)th discrete decision stage; Let be the system state vector for the k-th discrete decision stage; This is the control input vector for the k-th discrete decision stage.

[0093] NSDP initialization: Set the population size to 200, the number of iterations to 100, and the crowding threshold to 0.5, and initialize the decision sequence.

[0094] Non-dominated sorting and reverse recursion: Recursively calculate the non-dominated solutions for each state point from the final stage, and retain the uniformly distributed Pareto optimal switching strategy set by sorting by congestion.

[0095] Output the optimal policy library: Finally, about 200 Pareto optimal switching policies covering different operating conditions are generated as training labels for LSTM.

[0096] Based on a pre-trained LSTM deep learning network, the temporal mapping relationship of the Pareto optimal switching strategy set is learned to generate diesel injection regulation coefficients and clean energy injection regulation coefficients.

[0097] Step S3: Based on the adjustment coefficient, switch the fuel mode of the dual-fuel hybrid ship.

[0098] like Figure 5 As shown, an execution and feedback layer is set up to drive various fuel injection and throttle valve actuators, complete the smooth switching of three fuel modes, and continuously collect operational feedback data such as speed, vibration, and emissions.

[0099] This embodiment can solve the multi-objective coordinated switching problem of dual-fuel ships under complex sea state disturbances, time-varying loads, and changes in dual fuel characteristics, and takes into account the integrated control of switching stability, fuel economy, emissions and equipment life.

[0100] In one specific embodiment, a Wärtsilä 46DF dual-fuel engine with a cylinder bore of 460mm, a stroke of 580mm, and a rated speed of 600rpm is selected. min -1 The cylinder output power is 1200kW, the piston speed is 11.6m / s, and the single-cylinder displacement is 96.41L. The intelligent network system includes AIS equipment, weather sensors, and a shipborne IoT module, with a data transmission latency of ≤100ms. A dual-fuel engine model and a fuel mode switching control model were established using Matlab / Simulink.

[0101] Three mode switching conditions are set: (1) Condition 1 is switching from pure diesel engine mode to dual-fuel hybrid mode; (2) Condition 2 is switching from dual-fuel hybrid mode to pure clean energy engine mode; (3) Condition 3 is switching from dual-fuel hybrid mode to pure diesel engine mode. The threshold values ​​of the expected engine speed corresponding to the three switching conditions are as follows: Figure 4 As shown.

[0102] The simulation curves of fuel injection quantity and jet fuel quantity changes under the three mode switching conditions are shown in Figures 7(a), 7(b), 7(c), 8(a), 8(b), and 8(c). As can be seen from Figures 7(a), 7(b), 7(c), 8(a), 8(b), and 8(c), the strategy proposed in this embodiment smoothly adjusts the diesel fuel injection quantity and clean energy jet fuel quantity. The parameter switching nodes are highly consistent with the theoretical set values, and there are no obvious abrupt changes, overshoots, or offsets during the switching process. The parameter control accuracy is significantly better than that of traditional control schemes.

[0103] The strategy proposed in this embodiment integrates intelligent network technology to achieve synchronous acquisition of multi-source data such as speed, load, and environmental conditions. Combined with adaptive signal decomposition and blind source separation algorithms for signal denoising, it ensures the accuracy of the control basis. The NSDP algorithm, aiming at switching stability, economy, and environmental friendliness, completes global optimization, pre-planning the optimal fuel injection ratio at each moment and outputting precise baseline control commands. Furthermore, leveraging the advantages of LSTM deep learning networks in processing time-series dynamic signals, it performs real-time dynamic compensation for system delays and minor disturbances in operating conditions, finely adjusting the injection quantity and jet quantity. With the synergistic effect of these multiple technologies, the system can accurately follow the mode switching logic to adjust injection parameters, thus achieving higher injection parameter switching accuracy.

[0104] Simulation results under different operating conditions and switching directions show that the method can smoothly complete the mode switching in three typical switching conditions: pure diesel to dual fuel, dual fuel to pure clean energy, and dual fuel to pure diesel. The speed fluctuation, engine vibration, fuel consumption and exhaust emission indicators are always kept at a low level. The control effect is stable in the face of changes in operating conditions and external sea state disturbances, demonstrating excellent adaptive capability.

[0105] On the one hand, the intelligent connected system can sense dynamic information such as speed, load, and weather in real time, automatically identify the current navigation conditions, and complete the condition adaptation without manual intervention. On the other hand, the NSDP dynamic programming algorithm can dynamically update and optimize the strategy according to real-time constraints, adapting to the switching requirements of three fuel modes; the LSTM deep learning network learns patterns based on historical samples and can autonomously adapt to equipment characteristic drift and random disturbances. Simultaneously, the system adopts a full-state closed-loop feedback structure, which can correct control quantities in real time, further enhancing dynamic adaptability. Compared with traditional controllers with fixed parameters, this method can autonomously cope with complex and ever-changing ship operation scenarios, exhibiting stronger overall adaptability.

[0106] To more clearly demonstrate the control effect during mode switching, the vibration amplitude during fuel mode switching is used as the research object, and the comparison results are as follows: Figure 9 As shown. Figure 9In the simulation, the first 1-4 minutes are classified as Condition 1; the next 5-7 minutes as Condition 2; and the last 8-10 minutes as Condition 3. The simulation results show that the vibration amplitude of the aircraft corresponding to the strategy proposed in this embodiment is controlled within the range of 0.15-0.22 m / s². 2 Within the specified range, the overall vibration level is superior to the comparative scheme. Under various fuel mode switching conditions, the vibration amplitude of the proposed strategy remains at a low level, the switching response speed is faster, and the control effect is stable under different operating conditions.

[0107] First, intelligent connected vehicle technology constructs a multi-dimensional perception system that can simultaneously monitor engine operating parameters, ship navigation status, and external environmental parameters, predicting fuel mode switching nodes and changes in operating conditions in advance. This solves the action lag problem inherent in traditional control, reducing sudden power changes at the source. Second, the NSDP non-dominated sorting dynamic programming algorithm takes switching smoothness as its core optimization objective, combining it with speed constraints to solve for the optimal control scheme, rationally allocating speed differences to avoid drastic speed fluctuations and reduce vibration-induced factors. Finally, the LSTM deep learning network, relying on its temporal learning capabilities, fully fits the engine's dynamic response characteristics, performing fine-grained real-time adjustments to fuel injection and jet fuel volume, dynamically offsetting external interference and system errors. Furthermore, adaptive signal decomposition and blind source separation algorithms ensure the reliability of the speed signal, preventing control commands from deviating due to signal distortion.

[0108] The aforementioned technologies work together to ensure a continuous and smooth fuel mode switching process, effectively mitigating power shocks and ultimately significantly reducing ship vibration amplitude. Because the control system combining intelligent connectivity, NSDP, and LSTM can autonomously identify operating conditions, dynamically optimize strategies, and compensate for disturbances in real time, the strategy proposed in this embodiment possesses excellent adaptive performance and can adapt to complex navigation and switching scenarios.

[0109] According to the second aspect of this application, such as Figure 10 As shown in the figure, this embodiment provides a dual-fuel hybrid power ship fuel mode switching system based on NSDP and LSTM, including: The first acquisition module is used to collect real-time navigation information of dual-fuel hybrid ships, extract several initial feature signals from the real-time navigation information, and fuse all the initial feature signals to obtain spatiotemporal fusion features. The second acquisition module is used to input the spatiotemporal fusion features into a pre-trained LSTM deep learning network and output adjustment coefficients. The pre-trained LSTM deep learning network is trained through a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP. The switching module is used to switch the fuel mode of the dual-fuel hybrid ship based on the adjustment coefficient.

[0110] Specifically, this embodiment corresponds one-to-one with the above method embodiments. The functions of each module have been described in detail in the corresponding method embodiments, so they will not be repeated here.

[0111] According to a third aspect of this application, this embodiment provides a computer-readable storage medium having a computer program stored thereon, the computer program including executable instructions that, when executed by a processor, implement the method described above.

[0112] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or system capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0113] According to the fourth aspect of this application, such as Figure 11 As shown, an electronic device is provided, comprising: One or more processors; Memory is used to store executable instructions for the processor, which, when executed by one or more processors, cause one or more processors to implement the methods described above.

[0114] Electronic devices are manifested in the form of general-purpose computing devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including memory and processor).

[0115] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of a computer system, connecting all parts of the computer system through various interfaces and lines.

[0116] Memory can be used to store computer programs and / or modules. The processor implements various functions of the computer system by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and memory) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for switching fuel modes in a dual-fuel hybrid ship based on NSDP and LSTM, characterized in that, include: Real-time navigation information of dual-fuel hybrid ships is collected, several initial feature signals are extracted from the real-time navigation information, and all initial feature signals are fused to obtain spatiotemporal fusion features. The spatiotemporal fusion features are input into a pre-trained LSTM deep learning network, and the output adjustment coefficients are obtained. The pre-trained LSTM deep learning network is trained by a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP. Based on the aforementioned adjustment coefficient, the fuel mode of the dual-fuel hybrid ship is switched.

2. The method according to claim 1, characterized in that, The step of denoising the initial feature signals using independent component analysis before fusing all initial feature signals includes: The initial feature signal is decomposed into several intrinsic mode components and a residual component by using ensemble empirical modes; Calculate the correlation coefficient between each intrinsic mode component and the initial characteristic signal one by one, and determine whether the intrinsic mode component is a noise component or a useful component based on the correlation coefficient; The noise components are purified using an independent component analysis algorithm; The initial characteristic signal of the denoised signal is obtained by reconstructing the signal from all purified noise components and all useful components.

3. The method according to claim 1, characterized in that, The purification process for noise components using the independent component analysis algorithm includes: The noise component is represented as a combination of a noiseless speed signal and an unknown hybrid matrix. The combination form is preprocessed by centering and whitening to obtain one or more independent components; The signal separation vector is obtained by iteratively calculating for each independent component. After orthogonalizing and normalizing the signal separation vector, it is multiplied with the noise component to obtain the purified noise component.

4. The method according to claim 1, characterized in that: The real-time navigation information includes sea state information, traffic flow information, power operation information, and navigation information; The sea state information includes real-time wind speed, wind direction, significant wave height, water current speed and direction, air temperature, and water temperature. The traffic flow information includes the vessel's position, speed, course, distance to surrounding vessels, speed, course, nearest encounter distance, and channel congestion. The power operation information includes engine speed, output torque, coolant temperature, exhaust temperature, battery state of charge, diesel injection quantity, gas injection quantity, and mode switching command. The navigation information includes route type, speed limit, route gradient, estimated arrival time, and information on the division of idling zone / acceleration zone / cruising zone.

5. The method according to claim 1, characterized in that: The spatiotemporal fusion features are input into a pre-trained LSTM deep learning network. The spatiotemporal fusion features include engine speed, battery state of charge, ship speed, real-time wind speed, significant wave height, distance to surrounding ships, and course gradient. Output adjustment coefficients, which include diesel injection adjustment coefficients and clean energy injection adjustment coefficients.

6. The method according to claim 1, characterized in that: The LSTM deep learning network includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer, and an output layer connected in sequence. The input layer is used to output spatiotemporal fusion features. The first LSTM layer includes 128 neurons, the second LSTM layer includes 64 neurons, and the fully connected layer includes 32 neurons. The output layer is used to output adjustment coefficients.

7. The method according to claim 1, characterized in that, The method for generating the Pareto optimal switching strategy set includes: Construct a ternary minimization objective function: In the formula, The stability index for the switching process is the speed overshoot rate; The economic indicators for the voyage are fuel consumption per unit distance. The environmental performance indicators for the navigation process are nitrogen oxide emission concentrations. The formula for calculating the stability index is: in, Let the stability objective function be... For speed overshoot; This represents the engine's maximum speed during the switching process. The target rotational speed; The formula for calculating the economic indicators is as follows: In the formula, Total fuel consumption per unit distance; This refers to diesel fuel consumption. For clean energy consumption, This is the calorific value conversion factor. This refers to the total distance traveled by the ship. The formula for calculating the environmental protection index is as follows: In the formula, NO for mode switching process x Total emissions; For real-time NO x Emission concentration, To switch the start time, The time when the switch is completed; Establish constraints: Speed ​​constraint: In the formula, This represents the change in engine speed; This is the upper limit of the rate of change of rotational speed, meaning that the increase or decrease in rotational speed per second must not exceed 30 revolutions per minute; Injection volume constraint: , In the formula, This refers to the real-time injection volume of the diesel injection actuator; This is the maximum permissible injection quantity for the diesel injection system; As a clean energy source, it represents the real-time injection volume of the injection actuator; This represents the maximum permissible injection volume for a clean energy injection system. SOC security constraints: In the formula, State of charge of the ship's energy storage batteries; Switching time constraints: In the formula, This represents the total duration of the fuel mode switch. Preset time; By time step Discretize the switching process into N decision stages: Define state variables: In the formula, Let k be the system state vector at stage k. Let be the engine speed in the k-th stage; This represents the battery state of charge in the k-th stage. Let be the ship's speed in stage k; Let be the real-time wind speed at stage k; Define control variables: In the formula, This is the control input vector for the k-th stage; This refers to the diesel injection coefficient; The gas (clean energy) injection coefficient; State transition equations are established based on engine dynamics and battery models: In the formula, Let be the system state vector for the (k+1)th discrete decision stage; Let be the system state vector for the k-th discrete decision stage; This is the control input vector for the k-th discrete decision-making stage; Starting from the final stage, we work backward to calculate the non-dominated solutions for each stage, and retain the uniformly distributed Pareto optimal switching strategy set by sorting by congestion.

8. A fuel mode switching system for dual-fuel hybrid power ships based on NSDP and LSTM, characterized in that, include: The first acquisition module is used to collect real-time navigation information of dual-fuel hybrid ships, extract several initial feature signals from the real-time navigation information, and fuse all the initial feature signals to obtain spatiotemporal fusion features. The second acquisition module is used to input the spatiotemporal fusion features into a pre-trained LSTM deep learning network and output adjustment coefficients. The pre-trained LSTM deep learning network is trained through a Pareto optimal switching strategy set and a real ship navigation strategy set. The Pareto optimal switching strategy set is generated offline by a fuel mode switching multi-objective optimization model based on NSDP. The switching module is used to switch the fuel mode of the dual-fuel hybrid ship based on the adjustment coefficient.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program includes executable instructions that, when executed by a processor, implement the method of any one of claims 1-7.

10. An electronic device, characterized in that, include: One or more processors; A memory for storing executable instructions of the processor, which, when executed by the one or more processors, cause the one or more processors to perform the method according to any one of claims 1-7.