Ship control auxiliary method based on machine learning

Abstract

The present invention discloses a ship control auxiliary method based on machine learning, which relates to the field of ship intelligent control technology. The method includes ship data collection, fuel consumption prediction, dynamic speed planning and energy-saving feedback control. The present invention adopts a prediction method combining multimodality and disturbance adaptation to predict fuel consumption, uniformly models multidimensional data, strengthens the perception of environmental disturbances, achieves accurate prediction of fuel consumption under complex navigation conditions, and effectively improves the reliability of subsequent dynamic speed planning; adopts a speed planning method combining multi-objective and segmented constraints to perform dynamic speed planning, comprehensively considers the three core factors of fuel consumption, navigation time and safety risks, constructs a multi-objective cost function, and introduces segmented constraints to ensure the physical feasibility and risk control of speed adjustment, optimizes speed through a reinforcement learning mechanism, realizes intelligent adaptive speed adjustment, and improves the energy saving capability, navigation efficiency and operational safety of ships under complex navigation conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship intelligent control, and specifically refers to a ship control auxiliary method based on machine learning. Background Art

[0002] The ship control assistance method based on machine learning uses machine learning technology to model, analyze and learn various data during the ship's navigation process, thereby realizing an intelligent method of speed adjustment. It aims to dynamically adapt to the complex and changing marine environment, effectively reduce fuel consumption and operating costs, and promote the development of ship control towards intelligence and greenness while ensuring navigation safety.

[0003] However, in the existing ship control process, the marine environment is complex and changeable, and the traditional fuel energy consumption prediction has a weak response ability to environmental disturbances and often relies on empirical data, resulting in large prediction deviations and difficulty in supporting dynamic energy-saving control of speed. There are technical problems; the traditional speed control optimization goal is single, and it is not adaptable enough to complex sea conditions, and it is difficult to take into account fuel economy, navigation efficiency and safety risks. As a result, when facing a complex marine environment, it may not only bring unnecessary fuel waste, but also easily cause safety risks. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a ship control auxiliary method based on machine learning. In view of the technical problems that the marine environment is complex and changeable in the existing ship control process, the traditional fuel energy consumption prediction has a weak response to environmental disturbances, and often relies on empirical data, resulting in large prediction deviations, making it difficult to support dynamic energy-saving control of speed, this solution creatively adopts a prediction method that combines multi-modal and disturbance adaptive methods to predict fuel energy consumption, unifies the modeling of multi-dimensional data, strengthens the perception of environmental disturbances, and realizes accurate prediction of fuel energy consumption under complex navigation conditions, thereby effectively improving the reliability of subsequent dynamic speed planning; in view of the technical problems that the traditional speed control optimization objectives are poor in the existing ship control process, The speed planning method is single and lacks adaptability to complex sea conditions. It is difficult to balance fuel economy, navigation efficiency and safety risks, which may not only bring unnecessary fuel waste but also easily cause technical problems of safety risks when facing complex marine environments. This solution creatively adopts a speed planning method that combines multiple objectives and segmented constraints for dynamic speed planning. It comprehensively considers the three core factors of fuel consumption, navigation time and safety risks, constructs a multi-objective cost function, and introduces segmented constraints to ensure the physical feasibility and risk control of speed adjustment. At the same time, it optimizes the speed through the reinforcement learning mechanism to achieve intelligent adaptive speed adjustment in a changing environment, thereby significantly improving the energy-saving ability, navigation efficiency and operational safety of the ship under complex navigation conditions.

[0005] The technical solution adopted by the present invention is as follows: The ship control auxiliary method based on machine learning provided by the present invention comprises the following steps:

[0006] Step S1: ship data collection;

[0007] Step S2: fuel consumption prediction;

[0008] Step S3: dynamic speed planning;

[0009] Step S4: Energy-saving feedback control.

[0010] Furthermore, in step S1, the ship data collection is specifically to obtain ship control perception data by collecting ship operation data, ship structure data and external environment data;

[0011] The ship operation data includes fuel consumption, engine speed, propeller load, speed, heading and draft; the ship structure data includes ship type, load and ship length-to-width ratio; the external environment data includes wind speed, wind direction, wave height, current speed, current direction and sea water temperature.

[0012] Furthermore, in step S2, the fuel energy consumption prediction is used to predict the fuel energy consumption of the ship, specifically using a prediction method combining multimodal and disturbance adaptive to perform fuel energy consumption prediction to obtain ship fuel energy consumption prediction data, including the following steps:

[0013] Step S21: Multimodal feature encoding, used to fuse multimodal data, specifically, encoding the ship operation data, ship structure data, and external environment data respectively through three parallel multi-layer perceptrons, and splicing the encoding results to generate the ship navigation state vector;

[0014] The encoding result includes a ship operation vector, a ship structure vector and an external environment vector;

[0015] Step S22: Disturbance perception prediction, which is used to perceive disturbances and predict fuel consumption simultaneously. Specifically, a disturbance modeling mechanism is introduced into the standard transformer time series model to construct a transformer time series model with enhanced disturbance perception. Time series modeling is performed on the ship's navigation state vector to obtain an initial prediction value and confidence interval for fuel consumption.

[0016] The perturbation modeling mechanism is introduced into the standard transformer timing model to improve the stability of the model in the face of extreme environmental changes. Specifically, the Monte Carlo random dropout operation is applied in the self-attention layer and feedforward network layer of the standard transformer timing model to simulate the model's predicted response to changes in external environmental perturbations.

[0017] Step S23: elastic response analysis is used to quantify the marginal impact of changes in input variables on fuel consumption. Specifically, the input variables are extracted from ship operation data, ship structure data, and external environment data. The partial derivatives of the initial predicted fuel consumption values ​​with respect to the input variables are calculated and normalized to generate an energy consumption elastic response matrix.

[0018] Step S24: Scenario adaptive adjustment, which is used to adjust the fuel consumption prediction output according to the external environment. Specifically, a conditional attention mechanism is established to generate a scenario adjustment factor based on the external environment vector and the initial fuel consumption prediction value. The initial fuel consumption prediction value is then adjusted using the scenario adjustment factor to obtain the model-predicted fuel consumption.

[0019] Step S25: Generate ship fuel energy consumption prediction data, specifically by executing steps S21 to S24 to generate ship fuel energy consumption prediction data, the ship fuel energy consumption prediction data includes the ship navigation state vector, energy consumption elasticity response matrix, model predicted fuel energy consumption and confidence interval.

[0020] Furthermore, in step S3, the dynamic speed planning is used to dynamically plan the ship's speed. Specifically, a speed planning method combining multiple objectives and segmented constraints is used to perform dynamic speed planning to obtain an optimal speed solution, including the following steps:

[0021] Step S31: constructing a multi-objective cost function, which is used to construct a multi-objective cost function that simultaneously considers fuel energy consumption, flight time, and safety risks. Specifically, the fuel energy consumption cost function is constructed by combining the energy consumption elasticity response matrix and the model to predict fuel energy consumption. Then, the flight time cost and risk cost are defined to construct the multi-objective cost function.

[0022] Step S32: Segment constraint design, which is used to establish physical restrictions and risk control constraints for each time period in the navigation path. Specifically, speed constraint modeling, acceleration constraint modeling, and risk constraint modeling are used to obtain multi-segment navigation constraint conditions.

[0023] Step S33: Optimizing the speed plan, specifically, optimizing the speed plan by combining the multi-objective cost function and the multi-segment navigation constraints, and constructing a reinforcement learning model to obtain the optimal speed plan;

[0024] The construction of the reinforcement learning model specifically involves constructing the reinforcement learning model by setting a state space, setting an action space, setting a reward function, and setting a training strategy;

[0025] The setting of the state space specifically involves setting the ship's navigation state vector as the state space;

[0026] The setting of the action space specifically includes setting the value range of the speed increment as the action space, and setting the speed increment as the action;

[0027] The setting of the reward function is specifically to construct a reward function based on the multi-objective cost function and the navigation multi-segment constraint conditions;

[0028] The setting of the training strategy specifically adopts a proximal policy optimization algorithm to train the reinforcement learning model to maximize the reward function.

[0029] Furthermore, in step S4, the energy-saving feedback control is specifically to collect ship control perception data in real time. When the deviation between the actual fuel energy consumption and the model-predicted fuel energy consumption is greater than the error threshold, the speed plan is re-planned by executing steps S2 and S3 to achieve closed-loop optimization control of the ship speed.

[0030] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0031] (1) In view of the technical problems in the existing ship control process, the marine environment is complex and changeable, and the traditional fuel energy consumption prediction has a weak response ability to environmental disturbances. It often relies on empirical data, resulting in large prediction deviations and difficulty in supporting dynamic speed energy-saving control. This solution creatively adopts a prediction method that combines multimodal and disturbance adaptive methods to predict fuel energy consumption, unifies the modeling of multi-dimensional data, strengthens the perception of environmental disturbances, and realizes accurate prediction of fuel energy consumption under complex navigation conditions, thereby effectively improving the reliability of subsequent dynamic speed planning;

[0032] (2) In view of the technical problems in the existing ship control process, the traditional speed control optimization target is single, the adaptability to complex sea conditions is insufficient, and it is difficult to take into account fuel economy, navigation efficiency and safety risks. As a result, when facing a complex marine environment, it may not only cause unnecessary fuel waste, but also easily cause safety risks. This scheme creatively adopts a speed planning method that combines multiple objectives and segmented constraints to perform dynamic speed planning, comprehensively considers the three core factors of fuel energy consumption, navigation time and safety risks, constructs a multi-objective cost function, and introduces segmented constraints to ensure the physical feasibility and risk control of speed adjustment. At the same time, the speed is optimized through the reinforcement learning mechanism, and intelligent adaptive speed adjustment is achieved in a changing environment, thereby significantly improving the energy saving ability, navigation efficiency and operation safety of the ship under complex navigation conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flow chart of the ship control assistance method based on machine learning provided by the present invention;

[0034] Figure 2 Schematic diagram of the process of step S2;

[0035] Figure 3 Schematic diagram of the process of step S3.

[0036] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0039] Example 1, see Figure 1 The present invention provides a ship control auxiliary method based on machine learning, which includes the following steps:

[0040] Step S1: ship data collection;

[0041] Step S2: fuel consumption prediction;

[0042] Step S3: dynamic speed planning;

[0043] Step S4: Energy-saving feedback control.

[0044] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the ship data collection is specifically to obtain ship control perception data by collecting ship operation data, ship structure data and external environment data;

[0045] The ship operation data includes fuel consumption, engine speed, propeller load, speed, heading and draft; the ship structure data includes ship type, load and ship length-to-width ratio; the external environment data includes wind speed, wind direction, wave height, current speed, current direction and sea water temperature.

[0046] Example 3, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S2, the fuel energy consumption prediction is used to predict the fuel energy consumption of the ship. Specifically, a prediction method combining multimodal and disturbance adaptive is used to perform fuel energy consumption prediction to obtain ship fuel energy consumption prediction data, including the following steps:

[0047] Step S21: Multimodal feature encoding is used to fuse multimodal data. Specifically, three parallel multi-layer perceptrons are used to encode the ship operation data, ship structure data, and external environment data respectively, and the encoding results are spliced ​​to generate the ship navigation state vector. The calculation formula is:

[0048] ;

[0049] Where h t is the ship navigation state vector at the t-th time step, t is the time step index, Concat(·) is the concatenation operation, It is the first multi-layer perceptron used to encode ship operation data. is the ship operation data at the t-th time step, It is the second multi-layer perceptron, used to encode ship structure data. is the ship structure data at the t-th time step, It is the third multi-layer perceptron, used to encode external environment data. is the external environment data at the tth time step;

[0050] The encoding result includes a ship operation vector, a ship structure vector and an external environment vector;

[0051] Step S22: Disturbance perception prediction, which is used to perceive disturbances and predict fuel consumption simultaneously. Specifically, a disturbance modeling mechanism is introduced into the standard transformer time series model to construct a transformer time series model with enhanced disturbance perception. Time series modeling is performed on the ship's navigation state vector to obtain an initial prediction value and confidence interval for fuel consumption.

[0052] The perturbation modeling mechanism is introduced into the standard transformer timing model to improve the stability of the model in the face of extreme environmental changes. Specifically, the Monte Carlo random dropout operation is applied in the self-attention layer and feedforward network layer of the standard transformer timing model to simulate the model's predicted response to changes in external environmental perturbations.

[0053] Step S23: elastic response analysis is used to quantify the marginal impact of changes in input variables on fuel consumption. Specifically, the input variables are extracted from ship operation data, ship structure data, and external environment data. The partial derivatives of the initial predicted fuel consumption values ​​with respect to the input variables are calculated and normalized to generate an energy consumption elastic response matrix.

[0054] The calculation formula for calculating the partial derivative of the initial predicted value of fuel energy consumption with respect to the input variable and performing normalization is:

[0055] ;

[0056] Where, is the relative elasticity coefficient of the i-th input variable at the t-th time step, is the initial predicted value of fuel consumption at time step t, i is the input variable index, x i is the i-th input variable, is the partial derivative symbol, is the normalization factor;

[0057] The input variables include engine speed, propeller load, ship speed, heading, draft, ship type, load, ship aspect ratio, wind speed, wind direction, wave height, current speed, current direction and seawater temperature;

[0058] The calculation formula of the energy consumption elastic response matrix is:

[0059] ;

[0060] Where, is the energy elastic response matrix at the t-th time step, is the relative elasticity coefficient of the first input variable at the tth time step, is the relative elasticity coefficient of the second input variable at the tth time step, is the relative elasticity coefficient of the mth input variable at the tth time step, where m is the number of input variables;

[0061] Step S24: Scenario adaptive adjustment, which is used to adjust the fuel consumption prediction output according to the external environment. Specifically, a conditional attention mechanism is established to generate a scenario adjustment factor based on the external environment vector and the initial fuel consumption prediction value. The initial fuel consumption prediction value is then adjusted using the scenario adjustment factor to obtain the model-predicted fuel consumption.

[0062] The conditional attention mechanism is constructed by mapping the external environment vector into a key vector and a value vector through a fully connected layer, and mapping the initial predicted fuel consumption value into a query vector. Then, the scene adjustment factor is calculated by the attention weight. The calculation formula is:

[0063] ;

[0064] ;

[0065] Where, is the attention weight, softmax(·) is the softmax normalization function, Q is the query vector, K is the key vector, T is the transpose operation, and d a is the dimension of the attention projection space, is the scene adjustment factor at time step t, and V is the value vector;

[0066] The calculation formula for adjusting the initial predicted value of fuel energy consumption by the scenario adjustment factor is:

[0067] ;

[0068] Where, is the model-predicted fuel consumption at time step t;

[0069] Step S25: Generate ship fuel energy consumption prediction data, specifically by executing steps S21 to S24 to generate ship fuel energy consumption prediction data, the ship fuel energy consumption prediction data including the ship navigation state vector, the energy consumption elastic response matrix, the model predicted fuel energy consumption and the confidence interval;

[0070] By performing the above operations, in order to address the technical problems in the existing ship control process, such as the complex and changeable maritime environment, the weak response ability of traditional fuel energy consumption prediction to environmental disturbances, and the reliance on empirical data, which leads to large prediction deviations and makes it difficult to support dynamic energy-saving control of speed, this solution creatively adopts a prediction method that combines multimodal and disturbance adaptive methods to predict fuel energy consumption, unifies the modeling of multi-dimensional data, strengthens the perception of environmental disturbances, and realizes accurate prediction of fuel energy consumption under complex navigation conditions, thereby effectively improving the reliability of subsequent dynamic speed planning.

[0071] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the dynamic speed planning is used to dynamically plan the ship's speed. Specifically, a speed planning method combining multiple objectives and segmented constraints is used to perform dynamic speed planning to obtain the optimal speed solution, including the following steps:

[0072] Step S31: constructing a multi-objective cost function, which is used to construct a multi-objective cost function that simultaneously considers fuel energy consumption, flight time, and safety risks. Specifically, the fuel energy consumption cost function is constructed by combining the energy consumption elasticity response matrix and the model to predict fuel energy consumption. Then, the flight time cost and risk cost are defined to construct the multi-objective cost function.

[0073] The calculation formula of the fuel energy consumption cost function is:

[0074] ;

[0075] Where, Jfuel is the fuel energy consumption cost function, t max is the maximum time step, is the energy consumption elasticity penalty coefficient, ||·||1 is the L1 norm symbol, is the element-wise multiplication symbol, is the deviation of the ship speed at the tth time step relative to the reference speed;

[0076] The calculation formula for defining the flight time cost and risk cost is:

[0077] ;

[0078] ;

[0079] Where, is the voyage time cost, specifically the voyage time at the tth time step, is the sailing distance at time step t, v t is the speed at time step t, R t is the risk cost, specifically the safety risk value at time step t, is the security risk assessment function, is the environmental parameter vector at the tth time step, including wind speed, wave height and current velocity, w1 is the energy consumption sensitivity weight, is the energy consumption sensitivity term, ||·||2 is the L2 norm symbol, is the energy consumption elastic response normalization factor, specifically the maximum energy consumption elastic response value under the historical speed, w2 is the environmental disturbance weight, is the environmental disturbance term, F max is the environmental disturbance normalization factor, specifically the maximum environmental disturbance intensity at the historical speed;

[0080] The calculation formula of the environmental disturbance term is:

[0081] ;

[0082] Where, is the wind speed weight, wind t is the wind speed at time step t, is the wave height weight, wave t is the wave height at time step t, is the flow rate weight, water t is the ocean current velocity at time step t;

[0083] The calculation formula of the multi-objective cost function is:

[0084] ;

[0085] Where J is the multi-objective cost function, is the fuel energy consumption cost weight, is the flight time cost weight, is the risk cost weight;

[0086] Step S32: Segment constraint design, which is used to establish physical restrictions and risk control constraints for each time period in the navigation path. Specifically, speed constraint modeling, acceleration constraint modeling, and risk constraint modeling are used to obtain multi-segment navigation constraint conditions.

[0087] The speed constraint model is used to limit the speed, and the calculation formula is:

[0088] ;

[0089] Where, v min is the physical minimum value of the speed, v max It is the physical maximum value of the speed;

[0090] The acceleration constraint modeling is used to smooth speed changes, and the calculation formula is:

[0091] ;

[0092] Where, v t-1 is the speed at time step t-1, a max is the acceleration threshold, is the time step;

[0093] The risk constraint modeling is used to control risk costs, and the calculation formula is:

[0094] ;

[0095] Where R max is the risk threshold;

[0096] Step S33: Optimizing the speed plan, specifically, optimizing the speed plan by combining the multi-objective cost function and the multi-segment navigation constraints, and constructing a reinforcement learning model to obtain the optimal speed plan;

[0097] The construction of the reinforcement learning model specifically involves constructing the reinforcement learning model by setting a state space, setting an action space, setting a reward function, and setting a training strategy;

[0098] The setting of the state space specifically involves setting the ship's navigation state vector as the state space;

[0099] The setting of the action space specifically includes setting the value range of the speed increment as the action space, and setting the speed increment as the action;

[0100] The reward function is specifically set up based on the multi-objective cost function and the navigation multi-segment constraint conditions to construct a reward function, and the calculation formula is:

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] In the formula, reward t is the reward function, is the speed penalty weight, Pen1 is the penalty value for violating the speed constraint, is the acceleration penalty weight, Pen2 is the penalty value for violating the acceleration constraint, is the risk penalty weight, Pen3 is the penalty value for violating the risk constraint, and max(·) is the maximum value function;

[0106] The setting of the training strategy is specifically to use a proximal policy optimization algorithm to train the reinforcement learning model to maximize the reward function;

[0107] By performing the above operations, in order to address the technical problems in the existing ship control process, such as the single optimization target of traditional speed control, insufficient adaptability to complex sea conditions, and difficulty in balancing fuel economy, navigation efficiency and safety risks, which may not only lead to unnecessary fuel waste but also easily cause safety risks when facing a complex maritime environment, this solution creatively adopts a speed planning method that combines multiple objectives and segmented constraints to perform dynamic speed planning, comprehensively considering the three core factors of fuel energy consumption, navigation time and safety risks, constructing a multi-objective cost function, and introducing segmented constraints to ensure the physical feasibility and risk control of speed adjustment. At the same time, the speed is optimized through a reinforcement learning mechanism to achieve intelligent adaptive speed adjustment in a changing environment, thereby significantly improving the ship's energy saving ability, navigation efficiency and operational safety under complex navigation conditions.

[0108] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the energy-saving feedback control is specifically to collect ship control perception data in real time. When the deviation between the actual fuel energy consumption and the model-predicted fuel energy consumption is greater than the error threshold, the speed plan is re-planned by executing steps S2 and S3 to achieve closed-loop optimization control of the ship speed.

[0109] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0110] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0111] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A ship control assistance method based on machine learning, characterized by: The method comprises the following steps: Step S1: ship data collection to obtain ship control perception data; Step S2: Fuel consumption prediction, used to predict the fuel consumption of the ship, specifically using a prediction method combining multimodal and disturbance adaptive to perform fuel consumption prediction and obtain ship fuel consumption prediction data, including the following steps: Step S21: multimodal feature encoding; Step S22: disturbance perception prediction; Step S23: elastic response analysis; Step S24: scenario adaptive adjustment; Step S25: generation of ship fuel consumption prediction data; In step S21, the multimodal feature encoding is used to fuse multimodal data. Specifically, three parallel multi-layer perceptrons are used to encode the ship operation data, ship structure data, and external environment data, respectively, and the encoding results are spliced ​​to generate a ship navigation state vector; the encoding result includes the ship operation vector, the ship structure vector, and the external environment vector; In step S22, the disturbance perception prediction is used to perceive the disturbance and predict the fuel consumption at the same time. Specifically, a disturbance perception-enhanced transformer timing model is constructed by introducing a disturbance modeling mechanism into the standard transformer timing model, and the ship's navigation state vector is subjected to timing modeling to obtain the initial predicted value of the fuel consumption and the confidence interval. The perturbation modeling mechanism is introduced into the standard transformer timing model to improve the stability of the model in the face of extreme environmental changes. Specifically, the Monte Carlo random dropout operation is applied in the self-attention layer and feedforward network layer of the standard transformer timing model to simulate the model's predicted response to changes in external environmental perturbations. In step S23, the elastic response analysis is used to quantify the marginal impact of changes in input variables on fuel consumption. Specifically, the input variables are extracted from the ship operation data, ship structure data, and external environment data. The partial derivatives of the initial predicted fuel consumption values ​​with respect to the input variables are calculated and normalized to generate an energy consumption elastic response matrix. In step S24, the scenario adaptive adjustment is used to adjust the fuel consumption prediction output according to the external environment. Specifically, a conditional attention mechanism is established to generate a scenario adjustment factor based on the external environment vector and the initial fuel consumption prediction value. The initial fuel consumption prediction value is then adjusted by the scenario adjustment factor to obtain the model-predicted fuel consumption. In step S25, the ship fuel energy consumption prediction data is generated, specifically by executing steps S21 to S24 to generate the ship fuel energy consumption prediction data, the ship fuel energy consumption prediction data including the ship navigation state vector, the energy consumption elastic response matrix, the model predicted fuel energy consumption and the confidence interval; Step S3: Dynamic speed planning, which is used to dynamically plan the ship's speed. Specifically, a speed planning method combining multiple objectives and segmented constraints is used to perform dynamic speed planning and obtain the optimal speed solution. The method includes the following steps: Step S31: Multi-objective cost function construction; Step S32: Segmented constraint design; Step S33: Speed ​​solution optimization; In step S31, the multi-objective cost function is constructed to simultaneously consider fuel energy consumption, flight time, and safety risks. Specifically, the fuel energy consumption cost function is constructed by combining the energy consumption elasticity response matrix and the model to predict fuel energy consumption. Then, the flight time cost and risk cost are defined to construct the multi-objective cost function. In step S32, the segmented constraint design is used to establish physical restrictions and risk control constraints for each time period in the navigation path. Specifically, speed constraint modeling, acceleration constraint modeling, and risk constraint modeling are used to obtain multi-segmented navigation constraint conditions. Step S4: Energy-saving feedback control.

2. The ship control assistance method based on machine learning according to claim 1, characterized in that: In step S33, the speed plan is optimized by combining a multi-objective cost function and multi-segment navigation constraints, and optimizing the speed plan by building a reinforcement learning model to obtain the optimal speed plan. The construction of the reinforcement learning model specifically involves constructing the reinforcement learning model by setting a state space, setting an action space, setting a reward function, and setting a training strategy; The setting of the state space is specifically to set the ship's navigation state vector as the state space; the setting of the action space is specifically to set the value range of the speed increment as the action space, and set the speed increment as the action; the setting of the reward function is specifically to construct a reward function based on the multi-objective cost function and the navigation multi-segment constraint conditions; the setting of the training strategy is specifically to use the proximal strategy optimization algorithm to perform reinforcement learning model training to maximize the reward function.

3. The ship control assistance method based on machine learning according to claim 2, characterized in that: In step S4, the energy-saving feedback control is specifically to collect ship control perception data in real time. When the deviation between the actual fuel energy consumption and the model-predicted fuel energy consumption is greater than the error threshold, the speed plan is re-planned by executing steps S2 and S3 to achieve closed-loop optimization control of the ship speed.

4. The ship control assistance method based on machine learning according to claim 3 is characterized in that: In step S1, the ship data collection is specifically to obtain ship control perception data by collecting ship operation data, ship structure data and external environment data; The ship operation data includes fuel consumption, engine speed, propeller load, speed, heading and draft; the ship structure data includes ship type, load and ship length-to-width ratio; the external environment data includes wind speed, wind direction, wave height, current speed, current direction and sea water temperature.

Images