Ship shaft power generation energy management and efficiency optimization method based on multi-working-condition adaptation

Abstract

The invention relates to the technical field of ship power system control, in particular to a ship shaft power generation energy management and efficiency optimization method based on multi-working-condition adaptation, and the method comprises the steps: a data collection and processing step: collecting and preprocessing multi-source heterogeneous operation parameters capable of comprehensively representing a ship propulsion system, a power system, a motion state and an external environment in real time; a working condition dynamic identification step; an optimization model dynamic construction step; an online optimization and control step; and an adaptive learning step: based on new data generated in a system operation process, performing periodic incremental training on the time sequence deep learning model to update network parameters of the time sequence deep learning model. The self-adaptive learning enables the system to continuously optimize the working condition identification precision along with the time lapse of ship operation and the change of the external environment, and improves the long-term stability and adaptability of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship power system control, and particularly relates to a ship shaft generator energy management and efficiency optimization method based on multi-working condition adaptation. BACKGROUND

[0002] The ship shaft generator (SG) is an important energy-saving device that generates power using the excess power of the main engine. However, the existing energy management method of the shaft generator system has significant shortcomings: Coarse working condition recognition and poor adaptability: The existing technology usually only relies on a single or a few parameters such as the main engine speed or power to roughly judge the ship working condition (such as "constant speed navigation" and "maneuvering navigation"), and lacks fine and multi-dimensional recognition of the real and complex running state of the ship. For example, at the same rated speed of the main engine, the ship may be in completely different working conditions such as head wind, following wind, light load or heavy load, and the main engine excess power and power grid demand fluctuation characteristics differ greatly. Coarse recognition leads to inaccurate adaptation of the energy management strategy, either causing energy waste due to excessive conservation or affecting the safety of the main engine due to excessive aggressiveness.

[0003] Static energy management, lack of foresight and collaboration: Most existing schemes use a fixed rule or lookup table-based energy distribution strategy, and the control parameters (such as power distribution ratio and bus voltage threshold) are usually preset and static. This method cannot respond to the dynamic changes in excess power and random fluctuations in power consumption during ship navigation, nor can it respond to upcoming working condition changes (such as upcoming turning or avoidance operations). There is a lack of collaborative optimization between the energy management unit, the shaft generator, the auxiliary generator and the energy storage device, often resulting in "treating the headache and treating the foot pain", which leads to low overall system operation efficiency.

[0004] Localized efficiency optimization, lack of systematization: Existing optimization often only focuses on the instantaneous efficiency of the shaft generator itself or a certain point of the main engine, ignoring the fact that the shaft generator system is a coupled system involving mechanical, electrical and thermal energy conversion links. For example, blindly pursuing high power output of the shaft generator may cause the main engine to operate in an inefficient region, thereby increasing the fuel consumption rate; or frequently starting and stopping the auxiliary machine to stabilize the power grid frequency, resulting in a decrease in the average load rate of the auxiliary machine, which worsens its own efficiency. This local optimization rather than global optimization approach makes it difficult to truly improve the overall energy efficiency of the ship.

[0005] Therefore, there is an urgent need for a method that can accurately perceive the comprehensive working condition of the ship and dynamically, prospectively and collaboratively perform global energy management and efficiency optimization. SUMMARY

[0006] In order to achieve the above purpose, the application provides a ship shaft power generation energy management and efficiency optimization method based on multi-working condition adaptation, characterized by comprising the following steps: A data acquisition and processing step: real-time acquisition and preprocessing of multi-source heterogeneous operating parameters, including a main engine parameter group, a ship motion and environment parameter group, a power system parameter group, and a navigation instruction parameter group; A working condition dynamic identification step: inputting the processed main engine parameter group, ship motion and environment parameter group, power system parameter group, and navigation instruction parameter group into a pre-trained time series deep learning model to calculate a working condition identification result, which is used to represent the confidence degree of the ship being in a plurality of predefined refined working conditions at the current time, the working condition identification result includes a dominant working condition, and the refined working condition is a typical operating state distinguished from the traditional two-way method according to multi-dimensional operating parameter combination clustering; An optimization model dynamic construction step: selecting and constructing a global efficiency optimization model according to the dominant working condition in the working condition identification result, the optimization objective of the global efficiency optimization model being to make the comprehensive efficiency of the whole path from main engine fuel consumption to power grid power output highest, the decision variables of the global efficiency optimization model including power instructions of the shaft generator, auxiliary generator, and energy storage system, and the constraint condition coefficients of the global efficiency optimization model being adaptively adjusted according to the dominant working condition; An online optimization and control step: using an online optimization algorithm to solve the constructed global efficiency optimization model, and using the optimal decision variable values obtained by solving as control instructions, which are simultaneously issued to the execution mechanisms of the shaft generator, auxiliary generator, and energy storage system; An adaptive learning step: based on new data generated during system operation, periodically incrementally training the time series deep learning model to update its network parameters.

[0007] Preferably, the time series deep learning model is a classification model combining a long short-term memory network and an attention mechanism; The pre-training includes: using a historical data set containing time series samples of the multi-source heterogeneous operating parameters and their corresponding artificial labeled working condition labels to train the model, and minimizing the cross-entropy loss between the model prediction results and the true labels in the training process through a back propagation algorithm, so as to learn the nonlinear mapping relationship from the complex multi-dimensional input sequence to the refined working condition classification; The working condition identification result is a multi-dimensional vector, the value of each element in the vector representing the confidence probability of the model determining that the current state belongs to a predefined refined working condition, and the dominant working condition being the working condition category with the maximum confidence probability value.

[0008] Preferably, the predefined refined working condition includes at least the following categories: Rated speed smooth sailing in head sea condition: its characteristics are that the main engine speed is stable near the rated value, the ship speed is high and the fluctuation is small, the longitudinal and lateral angles are low, the wind and waves come from the stern direction, and the power grid load is smooth; Rated speed heavy load sailing in head sea condition: its characteristics are that the main engine speed is stable at the rated value, the ship speed is lower than expected, the longitudinal angle amplitude is large and the period is fixed, the wind and waves come from the bow direction, the main engine output torque and fuel injection are close to the upper limit; Speed reduction maneuvering and large rudder angle turning condition: its characteristics are that the main engine speed is in the process of falling or lower than the rated value, the rudder angle command set value deviates from zero continuously, the ship speed and heading continuously change, and the power grid load may fluctuate greatly; The category set of the refined working conditions is inducted and defined by applying an unsupervised clustering algorithm on a large amount of historical data and combining domain expert knowledge.

[0009] Preferably, in the dynamic construction step of the optimization model, the objective function of the global efficiency optimization model is a mathematical representation of the overall path comprehensive efficiency, which needs to consider the fuel consumption rate of the main engine at the current load point, the power generation efficiency of the shaft generator, the power generation efficiency of each auxiliary generator at its specific load point, and the charge and discharge efficiency of the energy storage system; The specific form of the objective function is adjusted according to the optimization emphasis implied by the dominant working condition.

[0010] Preferably, the adaptive adjustment of the constraint condition coefficient refers to: The upper limit value of the main engine total load safe operation interval in the main engine load constraint is set higher in the working condition of abundant main engine surplus power to fully utilize energy, and is set lower in the working condition of tight main engine surplus power to reserve more power margin for the propulsion system; The high-efficiency operation interval range of a single auxiliary generator in the auxiliary generator operation constraint is dynamically calculated according to the number of currently running auxiliary generators and the total power grid load, to ensure that the load rate of each running auxiliary generator falls within the high-efficiency section of its efficiency characteristic curve as much as possible; The expected value of the target state of charge in the energy storage system operation constraint is set according to the prediction of the working condition trend in the future period of time, and if it is predicted to enter a high-load working condition, the expected value is adjusted higher to store energy, and if it is predicted to enter a low-load working condition, the expected value is adjusted lower to prepare to absorb excess energy.

[0011] Preferably, in the online optimization and control step, the online optimization algorithm used is one of linear programming algorithm, quadratic programming algorithm or genetic algorithm; the algorithm selection strategy is: when the global efficiency optimization model is judged to be a convex optimization problem, the linear programming or quadratic programming algorithm with fast calculation speed is preferentially selected for accurate solution; when the model contains non-convex or discrete variables, the genetic algorithm is selected for heuristic search to obtain a suboptimal solution meeting the engineering application requirements within an acceptable time.

[0012] Preferably, the control instruction synchronization issuing refers to: the shaft generator target active power instruction, the auxiliary generator switching number and target power distribution instruction, and the energy storage system target charge and discharge power instruction calculated in one control period are respectively sent to the respective execution controllers through the ship control network in the same communication period, and each controller is required to start executing the new instruction at the starting point of the next unified control time, so as to ensure the coordinated action of multiple energy subsystems and avoid power shock or system instability due to asynchronous instructions.

[0013] Preferably, in the adaptive learning step, the specific process of periodic incremental training is: the system continuously stores the actual data in the running, including the input multi-source heterogeneous operating parameters, the model output working condition recognition results, the issued control instructions and the system state changes after execution, as new training samples in the buffer area; every fixed time period or when the number of new samples reaches a preset size, a part of the samples are extracted from the buffer area to perform one round of incremental training on the time series deep learning model; a small learning rate is used to fine-tune the network parameters during training, so that the model can slowly adapt to the running characteristic drift of the ship due to the ship hull fouling and the performance degradation of the main engine.

[0014] Preferably, in the data acquisition and processing step, The acquired multi-source heterogeneous operating parameters specifically include: The main engine parameter group: the main engine fuel engine shaft speed and the main engine output torque are acquired through the sensors installed on the main engine, and the main engine fuel injection amount, the main engine scavenging pressure and the main engine exhaust temperature are read through the main engine control system; The ship motion and environmental parameter group: the ship speed over ground is obtained through the global satellite positioning system receiver, the pitch angle and the roll angle are obtained through the attitude sensor installed near the center of gravity of the ship, the relative wind speed and the relative wind direction are obtained through the wind speed and direction sensor installed at the top of the mast, and the wave height is estimated by analyzing the ship motion response and combining the model; Power system parameter group: the active power and the reactive power of the shaft generator are collected through the current transformer and the voltage transformer installed on the output loop of the shaft generator, the total active load and the bus voltage of the power grid are collected through the current transformer and the voltage transformer installed on the bus of the main distribution board, and the current state of charge of the energy storage system is read through the battery management system communication interface of the energy storage system; Navigation instruction parameter group: the current main engine speed instruction set value and the rudder angle instruction set value are read from the bridge control system and the rudder control system through the ship local area network; The preprocessing comprises the following steps: firstly, all parameters except the rudder angle instruction are smoothed by using a sliding average filtering method to suppress high-frequency noise, and the time length of the sliding average window is pre-set according to the physical characteristics and the change speed of the parameters; secondly, all parameters are processed by using an amplitude limiting filtering method to eliminate abnormal values that are physically impossible due to instantaneous interference of the sensor, and the threshold of the amplitude limiting is determined according to the historical statistical maximum and minimum values of each parameter with a certain margin; finally, all parameters are marked with a unified time scale and aligned to ensure that the data at the same time correspond to the same sampling period.

[0015] Preferably, the method forms a closed-loop control system, and the data acquisition and processing step, the working condition dynamic identification step, the optimized model dynamic construction step and the online optimization and control step are periodically and cyclically executed, and the cycle period is determined according to the speed of the dynamic response of the ship power system.

[0016] Advantages of the present application: 1. The present application proposes a ship working condition recognition method based on multi-dimensional perception and dynamic adaptation, which accurately recognizes the specific working condition of the ship by comprehensively considering the main engine speed, power, speed, sea conditions, ship load and other real-time monitoring parameters. This fine working condition recognition method can ensure the accurate adaptation of the energy management strategy under complex working conditions, avoid energy waste and main engine safety risks in the traditional method, and thus improve the adaptability and accuracy of energy management.

[0017] 2. The present application introduces a dynamic energy distribution strategy based on a prediction model, which can respond to the changes in surplus power and load fluctuations during ship navigation in real time. Especially for the upcoming working condition conversion, the forward-looking energy management method of the present application can adjust the energy distribution according to the prediction information, thereby optimizing the energy efficiency of the whole system. In addition, the present application adopts a collaborative scheduling mechanism based on a global optimization algorithm, which significantly improves the coordination between the energy management unit, the shaft generator, the auxiliary generator and the energy storage device, avoids the local optimization problem of a single device, and thus improves the efficiency of the whole system.

[0018] 3、The application proposes a global optimization method, which combines the coupling relationship between mechanical, electronic and thermal systems, and optimizes the efficiency of the ship shaft generator system in all directions. Through multi-link comprehensive coordination optimization, the traditional method of local optimization of the shaft generator or the main engine is avoided. The method can improve the power output of the shaft generator while avoiding the main engine in the low efficiency operation area, reducing fuel consumption, and maintaining the best load rate of the auxiliary generator through optimizing the start-stop frequency, thereby improving the overall energy utilization efficiency of the system and achieving overall improvement of the ship energy efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0020] Fig. 1 The step flow chart of the method of the application; Fig. 2 The step flow chart of the adaptive adjustment of the constraint condition coefficient in the method of the application. DETAILED DESCRIPTION

[0021] The application will be described in detail below in combination with the drawings and specific embodiments. It should be noted here that, in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement some known technologies; and the drawings are only used to more specifically describe the embodiments, and are not intended to specifically limit the application.

[0022] Please refer to Figs. 1-2 The embodiment of the application provides a ship shaft generator energy management and efficiency optimization method based on multi-working condition adaptation, which collects the running data of the ship in real time, and pre-processes the data. The collected parameters include the propulsion system data of the ship (such as the speed of the main engine, the power output), the power system data (such as the grid voltage, the current, the power demand), the motion state of the ship (such as the speed, the heading, the ship attitude, etc.), and the external environment data (such as the sea state, the wind speed, the temperature, etc.). These parameters come from various sensors of the ship, such as speed sensors, temperature sensors, pressure sensors, etc. In order to ensure the accuracy and real-time performance of the data, the collected data need to be pre-processed, such as filtering, denoising and standardization, to ensure that the subsequent analysis process can effectively reflect the actual running condition of the ship.

[0023] The data acquisition and processing steps can ensure the acquisition of multi-dimensional and real-time running data of the ship, and provide a basis for accurate monitoring of the working condition of the ship. This data processing procedure provides reliable and accurate input for subsequent working condition identification and optimization models.

[0024] The pre-processed multi-source heterogeneous data is input into a pre-trained time series deep learning model. The model can capture the running characteristics of the ship under different working conditions by learning from the historical running data of the ship. Through the time series deep learning model, the system can calculate the confidence of the ship in multiple predefined working conditions, and identify the working condition according to the results. The working condition is not only dependent on the traditional speed or rotating speed, but is finely divided by combining multiple dimensional parameters, so as to identify complex working conditions such as "sailing against the wind" and "heavy load sailing" which are different from traditional two-classification.

[0025] Through the introduction of deep learning technology, the working condition identification becomes more accurate and can reflect the complex running state of the ship in real time. Compared with the rough judgment of traditional methods, the dynamic identification method provided by the present application can fully adapt to the variable running condition and avoid the failure of energy management strategy caused by inaccurate working condition identification.

[0026] According to the dominant working condition in the working condition identification result, the system dynamically selects and constructs a global efficiency optimization model. The goal of the model is to adjust the power instructions of each energy system of the ship (such as shaft generator, auxiliary generator and energy storage system) to make the whole path efficiency from main engine fuel consumption to power grid power output reach the highest. The constraint conditions of the optimization model are dynamically adjusted according to the current identified dominant working condition, for example, when the ship is in a light load state, it may be more inclined to optimize fuel efficiency, and when it is in a heavy load state, it needs to consider the stability of power supply more.

[0027] The dynamically constructed optimization model can flexibly adjust the energy distribution strategy according to the real-time working condition changes, so that the whole system runs more in line with the actual needs of the ship. Through the global optimization method, the present application can comprehensively consider the coordinated operation of different devices, improve the overall energy efficiency of the system, and avoid energy waste or unstable power output caused by static strategy.

[0028] An online optimization algorithm is used to solve the global efficiency optimization model in real time, and the optimal decision variables (such as the power of each generator, the power of the energy storage system, etc.) obtained are sent to the corresponding execution mechanism. Through real-time optimization, the system can dynamically respond to various changes in the ship's sailing process, such as load fluctuation and power grid frequency change, to ensure the coordinated work of the ship's energy devices.

[0029] The real-time and flexibility of online optimization ensure that the ship can obtain the best energy distribution under different sailing states, avoiding the lag and inadaptation problems caused by traditional static methods, and enhancing the response ability and stability of the system.

[0030] During the operation of the ship, new data is continuously generated, and the system will periodically incrementally train the time series deep learning model based on these new data, updating the network parameters, so that the model can adapt to the dynamic changes in long-term operation. This step can continuously optimize the working condition recognition ability, ensuring the accuracy and reliability of the model after a long time of operation.

[0031] Adaptive learning can continuously optimize the accuracy of working condition recognition as the ship runs over time and the external environment changes, improving the long-term stability and adaptability of the system. In this way, the model can more accurately identify complex and changing working conditions, continuously improving overall energy efficiency.

[0032] Through accurate data collection and processing, dynamic working condition recognition, construction of a global optimization model, real-time online optimization and control, and adaptive learning mechanism, the energy efficiency and stability of the ship's shaft generator system are comprehensively improved. Compared with existing technologies, this method has higher adaptability, flexibility and efficiency optimization capability, and can greatly improve the overall energy management level of the ship.

[0033] In one possible implementation, in this method, the data collection stage comprehensively obtains multiple dimensions of operating parameters through various sensors installed at different positions on the ship. These parameters are divided into four categories: Through sensors installed on the main engine, real-time collection of core data such as main engine shaft speed, main engine output torque, etc. is performed, while through the main engine control system, reading of operating states such as fuel injection amount, scavenging pressure and exhaust temperature, etc. is performed. These data help to comprehensively understand the working state of the main engine, ensure that the power output of the main engine is within a reasonable range, and avoid excessive fuel consumption or excessive mechanical load.

[0034] The ship's ground speed is obtained through the Global Positioning System (GPS), and the pitch and roll angles are measured using attitude sensors near the ship's center of gravity to understand the attitude changes of the ship. The wind speed and direction sensors at the top of the mast provide wind speed and direction information to help assess the wind impact during sailing. In addition, combined with the ship's motion response and model, the wave height can be estimated. These data are crucial for analyzing the power demand of the ship under different sailing states.

[0035] In the power system, the active power and reactive power of the shaft generator, as well as the active load and bus voltage of the power grid, are collected through current transformers and voltage transformers. These power data can reflect the load state of the system, helping to adjust the output of the generator and the energy storage system. In addition, the state of charge of the energy storage device is read through the battery management system communication interface of the energy storage system, so as to evaluate the energy storage capacity of the system and optimize energy distribution.

[0036] Navigation instructions are mainly obtained through the ship local area network, including main engine speed instruction and rudder angle instruction. By adjusting these instruction values in real time, the system can effectively control the speed and direction of the ship, ensuring stable operation of the ship under the given working condition.

[0037] The collected multi-source heterogeneous data often has noise or interference, which needs to be preprocessed to ensure the accuracy and reliability of the data. The specific preprocessing steps are as follows: For all parameters except the rudder angle instruction, sliding average filtering method is used for smoothing processing to suppress high-frequency noise. The length of the sliding average window is preset according to the physical characteristics of each parameter and its change rate. For example, the speed of the main engine changes slowly, while the current and voltage change quickly, so their window sizes are different.

[0038] All parameters are subjected to amplitude limiting processing to eliminate abnormal values caused by sensor transient interference or failure. The amplitude limiting value is determined according to the historical statistical maximum and minimum values of each parameter, and a certain margin is reserved on this basis to ensure the physical reasonableness of the data. This process effectively avoids the influence of extreme or unrealistic values on the accuracy of the data.

[0039] All collected data will be marked with a uniform time stamp to ensure that data from the same time comes from the same sampling period. Through alignment operation, the consistency between each data in time is ensured, preventing errors caused by asynchronous data collection.

[0040] Through efficient data collection, processing and preprocessing process, reliable support can be provided for subsequent working condition recognition and optimization decision, and finally the energy efficient management of the whole system is realized.

[0041] In one possible implementation, in the method, the time series deep learning model used is a classification model combining long short-term memory network (LSTM) and attention mechanism. LSTM, as a special recurrent neural network (RNN), has strong time series data processing capability and can effectively capture long-term dependencies in data. Combined with the attention mechanism, the model not only pays attention to important time segments in the input sequence, but also dynamically adjusts the area it pays attention to according to the context, so as to more accurately process multi-source heterogeneous time series data.

[0042] LSTM is capable of learning the dynamic changes of time series data, and is particularly suitable for capturing the complex time sequence characteristics of ships under different working conditions, such as the changing trends of parameters such as engine speed, ship speed, wind speed, etc. in different time periods.

[0043] Attention mechanism plays a crucial role in sequence data processing, as it allows the model to autonomously determine which historical information should be focused on when processing each time step. For the problem of ship multi-condition adaptation, the changes in certain parameters at certain times may be crucial for the final condition judgment, and the attention mechanism can help the model automatically identify these key moments, improving the recognition ability and accuracy of the model.

[0044] The pre-training of the model is carried out through the training of multi-source heterogeneous running parameter time series samples in the historical data set and the corresponding artificial labeled condition labels. The training data includes the running parameters of the ship under different working conditions, such as engine speed, fuel injection amount, power load, etc., and each time series sample has been manually labeled as the corresponding working condition category.

[0045] Through the use of the backpropagation algorithm, the cross-entropy loss between the model's prediction results and the true labels is minimized during the training process. Cross-entropy loss is a standard for measuring the difference between the model's prediction results and the actual labels, and backpropagation optimizes the model parameters through gradient descent, allowing the model to gradually adjust its internal parameters to reduce prediction errors and better learn the relationship between input data and output labels.

[0046] The process of model optimization through backpropagation can effectively learn the non-linear mapping relationship between complex multi-dimensional input sequences (including multiple different parameters such as speed, wind speed, ship attitude, etc.) and refined working condition classification (such as normal working condition, overload working condition, etc.). This relationship reflects the relationship between different ship operating states and specific working condition categories, and can help the system accurately determine the current operating state of the ship.

[0047] After the model is trained, input real-time multi-source heterogeneous running parameter data, and the model will output a multi-dimensional vector. Each element in this multi-dimensional vector represents the confidence probability of the model determining that the current state belongs to a certain predefined refined working condition. The confidence probability of each working condition category represents the degree of confidence of the model for that working condition. The higher the confidence, the more likely it is that the current state belongs to that working condition.

[0048] Finally, by comparing the confidence probabilities of each working condition category in the multi-dimensional vector, the working condition category with the highest confidence is selected as the "dominant working condition" of the current state. This dominant working condition represents the best prediction result of the model for the current working condition of the ship, and can provide a basis for subsequent energy management and optimization decisions.

[0049] The time series deep learning model combining LSTM and attention mechanism can greatly improve the energy management capability of the system through efficient working condition recognition in the energy management and efficiency optimization of the ship shaft power generation. The training and optimization process of the model ensures efficient mapping from complex multi-dimensional input data to refined working condition classification, laying the foundation for intelligent management of the ship.

[0050] In one possible implementation, the refined working condition categories defined in the method include a plurality of different navigation states, which can be used to refine the energy management and efficiency optimization strategies of the ship in different navigation environments. Specifically, the following working conditions are included: Rated speed smooth sailing with wind and wave working condition: In this working condition, the speed of the main engine is stable near the rated value, the speed of the ship is high and the fluctuation is small, which indicates that the ship is running in an ideal wind and wave sailing state. The amplitude of the pitch and roll angles of the ship is low, indicating that the sailing process is smooth, the wind and wave come from the stern, and the power grid load is relatively stable. This working condition is suitable for normal and efficient navigation of the ship.

[0051] Rated speed heavy load sailing with wind and wave working condition: This working condition is characterized by the main engine speed remaining stable at the rated value, and the ship speed is lower than expected, usually indicating that the ship encounters large headwinds and waves, affecting the ship speed. The pitch angle is large and fixed, indicating that the ship needs more stability when facing adverse weather conditions. At this time, the torque output of the main engine is close to the upper limit, and the fuel injection amount is also at a high level, indicating that the ship is in a heavy load state.

[0052] Speed reduction maneuvering and large rudder angle turning working condition: In this working condition, the speed of the main engine is in the process of decreasing, or lower than the rated value, which means that the ship is doing maneuvering or deceleration operation. The rudder angle command continuously deviates from zero, indicating that the ship is making a large angle turn, and the speed and heading are continuously changing. The power grid load may fluctuate greatly, indicating that the ship is in complex operating conditions, and the load of the system will also change more severely.

[0053] The definition of refined working conditions is not fixed, but is derived through analysis and processing of a large amount of historical data, combined with unsupervised clustering algorithms and expert knowledge in the field. Unsupervised clustering algorithms are used to automatically mine different working condition categories from ship operation data, especially the similarity between complex patterns or groups hidden in the data. Expert knowledge helps to further verify and adjust these categories, ensuring that the definition of each working condition category is consistent with the actual performance in operation.

[0054] Unsupervised clustering algorithms can automatically identify patterns in the data, clustering similar conditions based on changes in ship operating parameters such as speed, steering angle, wind and waves. By classifying these conditions, the ship's operating state can be effectively categorized, providing strong support for subsequent energy management and efficiency optimization.

[0055] Unsupervised clustering results can only provide potential group structures in the data, while domain expert knowledge is used to understand these structures and further adjust them. Experts can assess whether each condition is reasonable based on their practical experience and make corrections to ensure that the definition of refined conditions meets the actual sailing needs.

[0056] Through the combination of unsupervised clustering algorithms and expert knowledge, a set of refined ship operating conditions can be defined, providing a more intelligent and efficient framework for ship energy management and efficiency optimization, significantly improving the adaptability and energy efficiency of ships in complex environments.

[0057] In one possible implementation, to calculate the comprehensive efficiency, the following aspects need to be considered: Main engine fuel consumption rate: The efficiency of the main engine directly affects the overall fuel consumption. The fuel consumption rate of the main engine varies greatly under different load conditions, so the efficiency needs to be estimated based on the specific consumption of the main engine at the current load point.

[0058] Shaft generator efficiency: The shaft generator converts the mechanical energy of the ship into electrical energy, and its efficiency is affected by factors such as load, operating speed and working conditions, which need to be considered in the model.

[0059] Auxiliary generator efficiency: The efficiency of the auxiliary generator under different load points of the ship should also be considered in the comprehensive consideration. The load state of different auxiliary machines will affect their power generation efficiency, affecting the supply and distribution of energy.

[0060] Energy storage system charging and discharging efficiency: The charging and discharging efficiency of the energy storage system is crucial for the use of ship energy, especially in dealing with energy surplus or deficiency, how to efficiently use the energy storage system can optimize the overall operation of the ship.

[0061] The specific form of the objective function will be adjusted according to the current operating conditions of the ship. For example: Excess power abundant condition: When the ship is in an excess power abundant state, the optimization goal will be biased towards maximizing energy capture. At this time, the system will focus on storing excess energy for use when power is tight, or by improving power generation efficiency to increase energy collection.

[0062] Power surplus condition: In the power surplus condition, the optimization focus of the system shifts to ensuring operational stability and main engine safety. In this case, the system prioritizes the safe operation of the main engine, avoiding overload or damage, while reasonably allocating limited energy resources to ensure normal navigation of the ship.

[0063] The optimization process needs to dynamically adapt to changes in different conditions. During navigation, the load and operating state of the ship will change, and the system needs to dynamically adjust the parameters and objective function of the optimization model through real-time monitoring and data feedback to respond to the changing navigation environment and conditions.

[0064] In one possible implementation, the main engine is the core power source of the ship, and the reasonable allocation of its load directly affects the efficient use of energy. Under different conditions, the upper limit of the main engine load will be dynamically adjusted: When the ship is in a state of abundant power, the upper limit of the main engine load will be set to a higher value to fully utilize the available energy. This allows the ship to use as much power as possible from the main engine without overloading.

[0065] In the power shortage condition, the upper limit of the main engine load will be appropriately reduced to ensure that the main engine does not overload and to leave more power margin for the propulsion system. This adjustment ensures that the ship can still operate safely and stably under high load conditions.

[0066] Auxiliary generator plays a crucial role in the energy supply of the ship. Each auxiliary generator has an efficiency characteristic curve that presents a certain high-efficiency operating interval, so it must be ensured that the auxiliary generator operates in this high-efficiency section: According to the number of currently operating auxiliary generators and the load condition of the entire power grid, the system dynamically calculates the operating load of each auxiliary generator to ensure that it works in the highest efficiency range as much as possible. This can reduce energy loss and improve overall power generation efficiency.

[0067] The energy storage system plays a crucial role in regulating the flow of energy on the ship. The state of charge of the energy storage system must be adjusted according to the prediction of future conditions: If it is predicted that the ship will enter a high-load condition in the future, the energy storage system will adjust the target state of charge to a higher value to reserve more energy to cope with the upcoming demand peak.

[0068] When it is predicted that the ship will be in a low-load state in the future, the target state of charge of the energy storage system will be appropriately lowered to ensure that it can absorb and store excess energy. This process helps to balance the supply and demand of energy on the ship and avoids energy waste.

[0069] The adaptive adjustment of constraint coefficients helps to achieve optimal operation of the ship system under varying conditions, improve overall energy efficiency, and effectively ensure the safety and stability of ship operation, ultimately reducing energy consumption and enhancing the economy and sustainability of the ship.

[0070] In one possible implementation, first, according to the structure of the global efficiency optimization model, it is judged whether it belongs to a convex optimization problem. Convex optimization problems have a unique optimal solution and usually have good mathematical properties, which can be quickly solved by precise mathematical solving methods. Specifically, when the energy management model of the ship has characteristics such as continuity and monotonicity, it can be judged as a convex optimization problem.

[0071] When the model is determined to be a convex optimization problem, the algorithm selection strategy will preferentially select linear programming or quadratic programming algorithms. These two algorithms have the advantages of fast calculation speed and high solution accuracy, and are suitable for solving optimization problems with deterministic constraints. Linear programming algorithm is suitable for systems with linear objective function and constraint conditions, while quadratic programming can handle problems with quadratic objective function. These algorithms can quickly provide accurate optimal solutions, thereby improving the efficiency of ship energy management.

[0072] If the global efficiency optimization model contains non-convex parts or discrete variables (such as the selection of certain operating states, switch control, etc.), the algorithm selection strategy switches to genetic algorithm. Genetic algorithm is a heuristic search method that can effectively avoid falling into local optimal solution when solving complex nonlinear and non-convex optimization problems. Genetic algorithm simulates the process of natural selection, adapts to environmental changes, and quickly finds suboptimal solutions to problems, and is suitable for handling large-scale or discrete problems.

[0073] The online optimization process monitors the changes in the operating conditions of the ship in real time and adjusts the optimization strategy according to real-time feedback. Regardless of which algorithm is used, the optimization process is completed within a certain time to meet the needs of engineering applications. Through online optimization, the ship can flexibly adjust the energy management strategy according to the current operating conditions to maximize system operating efficiency.

[0074] By dynamically selecting appropriate optimization algorithms based on different operating conditions and model characteristics, the ship shaft generator energy management system can flexibly respond to varying operating requirements, improve efficiency while ensuring system stability and safety, and provide strong support for efficient energy management of the ship.

[0075] In one possible implementation, in each control period, a plurality of control instructions are calculated according to the operating conditions and energy demand of the ship. These instructions include: The target active power instruction of the shaft generator indicates the power that the shaft generator should generate.

[0076] The number of auxiliary generator switching stations and the target power distribution instruction determines the number of auxiliary generators that need to be started or stopped and the power distribution of each unit.

[0077] The target charge and discharge power instruction of the energy storage system specifies whether the energy storage device should be charged or discharged and the power size of the charge and discharge.

[0078] In one communication cycle, the above instructions are sent to the corresponding execution controllers through the ship's control network (such as the ship's control bus or other real-time communication systems). Each controller is responsible for the execution of the corresponding energy subsystem, such as the generator controller for power regulation of the generator, and the energy storage system controller for battery charging and discharging.

[0079] Through the coordination of the ship's control network, all the calculated instructions are sent and the controllers are required to start executing the new instructions at the beginning of the next unified control time. In this way, all the instructions will be executed at the same time, ensuring that each subsystem is started or adjusted at the same time node to ensure the coordinated operation of the system.

[0080] Since the control instructions of multiple energy subsystems are executed at the same time, the coordinated action between these subsystems can be ensured, and power shock or system instability caused by asynchronous instructions can be avoided. Each controller receives the instructions and operates at the unified control time node, so that the power output of different energy sources can be balanced to ensure the smooth operation of the ship system.

[0081] The technical features of synchronous control instruction delivery significantly improve the efficiency and stability of the ship's energy management system, providing more reliable protection for the ship's energy use, while avoiding the problem of system instability caused by asynchronous control instructions.

[0082] In one possible implementation, the system continuously monitors and collects real-time data from multiple sources during the operation of the ship, including: Input multi-source heterogeneous operating parameters: for example, ship speed, load, sailing environment (sea conditions, weather, etc.), etc.

[0083] Model output working condition recognition results: the results of the current working condition analysis are used to infer the running state of the ship.

[0084] The control instructions issued and the system state changes after execution: the sending of instructions and the changes in system state (such as engine output power, generator load, etc.). All these data are stored in the buffer as new training samples.

[0085] Every certain time period, or when the number of new samples in the buffer reaches a preset size, the system extracts a portion of samples from the buffer for training. At this time, the model performs an incremental training based on these new data, i.e., updates and optimizes the model parameters based on the original model, rather than starting from scratch. Specifically, this process includes: Small step learning rate: A small learning rate is used for training to prevent drastic adjustments to model parameters. The purpose of this is to fine-tune existing network weights so that the model can adapt to slow changes during ship operation, such as fouling of the hull, performance degradation of the main engine, etc.

[0086] Adaptive adjustment: As training progresses, the model gradually adapts to subtle changes in ship operation, adjusts control strategies, and ensures continuous and effective energy management and efficiency optimization.

[0087] After each incremental training, the updated model continues to function in the ship's operation, adjusting control instructions in real time and monitoring actual execution effects. New training data will continue to be collected and stored to provide a basis for the next round of incremental training, forming a continuous optimization closed loop.

[0088] Periodic incremental training not only enhances the system's ability to respond to performance changes during long-term operation, but also ensures efficient and stable operation of the ship under different operating conditions through continuous adaptive optimization.

[0089] In one possible implementation, the system continuously collects ship operation data, including input parameters of the power system, output results, equipment status, and operating condition changes. After preprocessing, these data are transmitted to the control part of the system. The data processing process includes denoising, standardization, and real-time updating to provide an accurate basis for subsequent dynamic identification.

[0090] Through analysis of real-time data, the system identifies the current operating condition based on the existing model. This process quickly determines the current operating state of the ship through real-time matching and adjustment of input parameters. The dynamic operating condition identification result is the basis for online optimization, which can help the system predict future power demand and load changes.

[0091] Based on the results of operating condition identification, the system dynamically constructs an optimization model to develop the optimal power generation strategy and energy distribution scheme. This model not only adjusts based on historical data and operating condition changes, but also predicts and prepares for possible future operating condition changes.

[0092] On the basis of the optimization model construction, the system performs real-time energy management and optimization of control instructions. This process is realized through a feedback mechanism and a controller to achieve real-time adjustment, ensuring that each subsystem of the ship power system (such as generators, energy storage devices, etc.) operates efficiently according to the current working conditions, maintaining system stability and optimizing energy utilization.

[0093] The above four steps (data acquisition and processing, working condition dynamic identification, dynamic construction of optimization model, online optimization and control) will be periodically executed. This cycle period will be determined according to the dynamic response speed of the ship power system. That is, when the system response is fast, the cycle period is short, so as to respond to the working condition changes more timely; when the system response is slow, the cycle period can be appropriately extended to avoid instability or oscillation caused by too frequent control operations.

[0094] The periodic cycle execution method of the closed-loop control system can ensure that the ship power system responds to the working condition changes in time, while avoiding instability caused by frequent control, thereby realizing efficient and stable energy management.

[0095] The present application encompasses any substitutions, modifications, equivalent methods and solutions made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0096] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A method for ship shaft power generation energy management and efficiency optimization based on multi-working condition adaptation, characterized in that, The method comprises the following steps: Data acquisition and processing step: real-time acquisition and preprocessing of multi-source heterogeneous operating parameters, including host parameter group, ship motion and environmental parameter group, power system parameter group, and navigation instruction parameter group; Working condition dynamic identification step: input the processed host parameter group, ship motion and environmental parameter group, power system parameter group, and navigation instruction parameter group into a pre-trained time series deep learning model to calculate a working condition identification result, which is a confidence degree for representing that the ship is in a plurality of pre-defined refined working conditions at the current time, the working condition identification result includes a dominant working condition, and the refined working condition is a typical operating state distinguished from the traditional two-way method according to multi-dimensional operating parameter combination clustering; Optimization model dynamic construction step: according to the dominant working condition in the working condition identification result, a global efficiency optimization model is selected and constructed, the optimization objective of the global efficiency optimization model is to make the comprehensive efficiency of the whole path from host fuel consumption to power grid power output highest, the decision variables of the global efficiency optimization model include power instructions of the shaft generator, auxiliary generator and energy storage system, and the constraint condition coefficients of the global efficiency optimization model are adaptively adjusted according to the dominant working condition; Online optimization and control step: an online optimization algorithm is used to solve the constructed global efficiency optimization model, and the optimal decision variable value obtained by solving is used as a control instruction and is simultaneously issued to the execution mechanism of the shaft generator, auxiliary generator and energy storage system; Adaptive learning step: based on new data generated during system operation, the time series deep learning model is periodically incrementally trained to update its network parameters.

2. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method according to claim 1, characterized in that, The time series deep learning model is a classification model combining long short-term memory network and attention mechanism; The pre-training includes: using a historical data set containing time series samples of the multi-source heterogeneous operating parameters and corresponding artificial labeled working condition labels to train the model, and minimizing the cross-entropy loss between the model prediction result and the true label in the training process by using a back propagation algorithm, so as to learn the non-linear mapping relationship from the complex multi-dimensional input sequence to the refined working condition classification; The working condition identification result is a multi-dimensional vector, and the value of each element in the vector represents the confidence probability that the model determines that the current state belongs to a certain pre-defined refined working condition, and the dominant working condition is the working condition category with the maximum confidence probability value.

3. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method according to claim 2, characterized in that, The pre-defined refined working conditions at least include the following categories: Rated speed smooth sailing in head wind working condition: Its characteristics are that the host speed is stable near the rated value, the ship speed is high and has small fluctuations, the longitudinal and lateral shaking angle amplitudes are low, the wind and waves come from the stern direction, and the power grid load is stable; Rated speed heavy load sailing in head wind working condition: Its characteristics are that the host speed is stable at the rated value, the ship speed is lower than expected, the longitudinal shaking angle amplitude is large and fixed, the wind and waves come from the bow direction, and the host output torque and fuel injection amount are close to the upper limit; Speed reduction maneuvering and large rudder angle turning working condition: Its characteristics are that the main engine speed is in the process of falling or lower than the rated value, the rudder angle command set value deviates from zero continuously, the ship speed and course change continuously, and the power grid load may have large fluctuations; The category set of the fine working conditions is obtained by applying an unsupervised clustering algorithm to a large amount of historical data and combining field expert knowledge for induction and definition.

4. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, In the step of dynamically constructing the optimization model, the objective function of the global efficiency optimization model is a mathematical representation of the overall path synthesis efficiency, and the calculation of the efficiency needs to comprehensively consider the fuel consumption rate of the main engine at the current load point, the power generation efficiency of the shaft generator, the power generation efficiency of each auxiliary generator at the specific load point, and the charge and discharge efficiency of the energy storage system. The specific form of the objective function is adjusted according to the optimization emphasis implied by the dominant working condition.

5. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method according to any of claims 1 or 4, characterized in that, The adaptive adjustment of the constraint condition coefficient refers to: The upper limit value of the main engine total load safe operation interval in the main engine load constraint is set to be higher in the working condition of sufficient main engine surplus power to fully utilize the energy, and is set to be lower in the working condition of insufficient main engine surplus power to reserve more power margin for the propulsion system; The high-efficiency operation interval range of a single auxiliary generator in the auxiliary generator operation constraint is dynamically calculated according to the number of currently running auxiliary generators and the total power grid load, so as to ensure that the load rate of each running auxiliary generator falls within the high-efficiency section of its efficiency characteristic curve. The expected value of the target state of charge of the energy storage system in the energy storage system operation constraint is set according to the prediction of the working condition trend in the future period of time, and if it is predicted to enter a high-load working condition, the expected value is adjusted to be higher to store energy, and if it is predicted to enter a low-load working condition, the expected value is adjusted to be lower to prepare to absorb excess energy.

6. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, In the step of online optimization and control, the online optimization algorithm used is one of a linear programming algorithm, a quadratic programming algorithm or a genetic algorithm; the algorithm selection strategy is: when the global efficiency optimization model is judged to be a convex optimization problem, the linear programming or quadratic programming algorithm with fast calculation speed is preferentially selected for accurate solution; when the model contains non-convex or discrete variables, the genetic algorithm is selected for heuristic search to obtain a suboptimal solution that meets the engineering application requirements within an acceptable time.

7. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, The synchronous issuance of the control instructions refers to: the shaft generator target active power instruction, the auxiliary generator switching number and target power distribution instruction, and the energy storage system target charge and discharge power instruction calculated in one control period are sent to the respective execution controllers through the ship control network in the same communication period, and each controller is required to start executing the new instructions at the start of the next unified control time, so as to ensure the coordinated action of multiple energy subsystems and avoid power shock or system instability caused by asynchronous instructions.

8. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, In the adaptive learning step, the specific process of periodic incremental training is as follows: the system continuously stores the actual data in the running buffer zone, including the input multi-source heterogeneous operating parameters, the model output working condition recognition results, the issued control instructions and the system state changes after execution; every fixed time period or when the number of new samples reaches a preset size, a part of the samples are extracted from the buffer zone to perform a round of incremental training on the time series deep learning model. During training, a small learning rate is used to fine-tune the network parameters, so that the model can slowly adapt to the drift of the running characteristics of the ship due to the fouling of the hull and the performance degradation of the main engine.

9. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, In the data acquisition and processing step, The acquired multi-source heterogeneous operating parameters specifically include: The main engine parameter group: the shaft speed and output torque of the main engine fuel engine are collected through the sensors installed on the main engine, and the fuel injection amount, scavenging pressure and exhaust temperature of the main engine are read through the main engine control system; The ship motion and environmental parameter group: the ship ground speed is obtained through the global satellite positioning system receiver, the pitch angle and roll angle are obtained through the attitude sensor installed near the center of gravity of the ship, the relative wind speed and direction are obtained through the wind speed and direction sensor installed at the top of the mast, and the wave height is estimated by analyzing the ship motion response and combining the model; The power system parameter group: the active and reactive power of the shaft generator is collected through the current and voltage transformers installed on the output circuit of the shaft generator, the total active load and bus voltage of the power grid are collected through the current and voltage transformers installed on the bus of the main distribution board, and the current state of charge of the energy storage system is read through the battery management system communication interface of the energy storage system; The navigation instruction parameter group: the current main engine speed instruction set value and rudder angle instruction set value are read from the bridge control system and rudder control system through the ship local area network; The preprocessing includes: first, using the moving average filtering method to smooth all parameters except the rudder angle instruction to suppress high-frequency noise, the time length of the moving average window is pre-set according to the physical characteristics and change speed of the parameters; second, using the amplitude limiting filtering method to process all parameters to eliminate abnormal values that are physically impossible due to sensor instantaneous interference, the threshold of the amplitude limiting is determined according to the historical statistical maximum and minimum values of each parameter with a certain margin; finally, all parameters are marked with a unified time scale and aligned to ensure that the data at the same time corresponds to the same sampling period.

10. The multi-operating condition adapted ship shaft power generation energy management and efficiency optimization method of claim 1, wherein, The method forms a closed-loop control system, and the data acquisition and processing step, the working condition dynamic identification step, the optimization model dynamic construction step and the online optimization and control step are periodically executed in a cycle, and the cycle period is determined according to the speed of the dynamic response of the ship power system.

Images