A multi-operating condition identification method, system, device and medium for all-electric ships

Abstract

The present application relates to the technical field of ship operating condition identification, and specifically to a multi-operating condition identification method, system, equipment, and medium for all-electric ships, including: obtaining historical operating data of all-electric ships, performing feature extraction and noise reduction, and obtaining noise reduction data groups for each sample; preliminarily clustering the noise-reduced propulsion load and the first-order derivative of the propulsion load into multiple basic operating conditions, and adding a unique basic operating condition label to the noise reduction data group; judging whether the absolute value of the first-order derivative of the propulsion load after noise reduction of each sample is higher than a preset threshold, summarizing the noise reduction data groups corresponding to samples with all the results being yes into a high-fluctuation data set, and summarizing the propulsion load, the first-order derivative of the propulsion load, and the basic operating condition labels corresponding to other samples into a general data set; using the kernel principal component analysis algorithm to extract the main features of the data in the high-fluctuation data set and the general data set, and then inputting them into a pre-trained operating condition segmentation model to output specific operating condition categories. The present application can improve the accuracy and real-time performance of all-electric ship operating condition identification.

Description

Technical Field

[0001] The present application relates to the technical field of ship operating condition identification, and specifically to a multi-operating condition identification method, system, equipment and medium for an all-electric ship. Background Art

[0002] An all-electric ship is a ship that is completely powered by electricity, with its power system, propulsion system, and other auxiliary equipment all using electricity as their energy source. Compared to traditional diesel ships, the power system of an all-electric ship has low inertia and is more susceptible to interference. Sharp fluctuations in propulsion load can directly affect the stability of the power system, making high-precision propulsion load prediction a core component in ensuring efficient and stable ship operation. However, the actual operating conditions of ships are complex and changeable, affected by multiple factors such as the environment and the mission, resulting in severe challenges for load prediction models. Establishing an accurate multi-condition identification system that can dynamically distinguish between different navigation states is key to improving load prediction performance and ensuring stable ship operation.

[0003] Existing technologies address the problem of identifying ship operating conditions by constructing hybrid ship operating condition datasets and incorporating deep learning strategies to optimize generator fuel consumption, or by integrating operating condition models based on mathematical mechanisms, and by utilizing machine learning algorithms to achieve operating condition classification. For example, signal decomposition algorithms are used to address data noise, clustering methods are employed to classify basic operating conditions, and ensemble learning models are introduced to improve identification accuracy. These methods can achieve a certain degree of operating condition differentiation in specific scenarios, providing a technical foundation for ship operating status monitoring.

[0004] However, existing methods for identifying ship operating conditions still have significant limitations. The operating environment of all-electric ships is complex, and sensor data is susceptible to noise interference, resulting in high-dimensional, nonlinear, and non-stationary raw data. Traditional methods have difficulty effectively reducing noise and extracting key features. Furthermore, sudden operating conditions and random environmental changes require high real-time identification, and existing single clustering or classification models struggle to adapt their strategies, resulting in insufficient accuracy in identifying highly fluctuating operating conditions. Furthermore, feature redundancy in the operating condition identification process has not been effectively addressed, resulting in a heavy computational burden on the model, which limits its applicability to engineering applications in complex scenarios. Summary of the Invention

[0005] In response to the technical problems that existing ship operating condition identification methods are susceptible to noise interference, high-dimensional feature redundancy, low accuracy in high-fluctuation operating condition identification, and poor adaptability due to fixed model parameters, the present application provides a multi-operating condition identification method, system, equipment and medium for all-electric ships. Through data denoising and reconstruction, a two-stage dynamic identification framework, Bayesian optimization and weighted iterative strategy, it achieves high-dimensional feature simplification, accurate identification of sudden operating conditions and adaptive adjustment of models, which can improve the accuracy and real-time performance of all-electric ship operating condition identification in complex scenarios.

[0006] In a first aspect, the present application provides a multi-operating condition identification method for an all-electric ship, comprising the following steps:

[0007] S1. Obtain historical operating data for all-electric vessels. This historical operating data consists of samples at various timestamps. Each sample includes the propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters at the corresponding timestamp.

[0008] S2. Perform feature extraction and noise reduction on the historical operating data to obtain a noise-reduced data set for each sample, including the noise-reduced propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters.

[0009] S3. Input the denoised propulsion load and first-order derivative of the propulsion load of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic operating conditions and adding a unique basic operating condition label to each denoised data group;

[0010] S4. Determine whether the absolute value of the first-order derivative of the propulsion load after noise reduction for each sample is higher than a preset threshold. Aggregate the noise reduction data set corresponding to all samples with a positive result into a high-fluctuation dataset. Aggregate the propulsion load, first-order derivative of the propulsion load, and basic operating condition labels corresponding to all samples with a negative result into a general dataset.

[0011] S5. Conduct secondary subdivisions, including:

[0012] The kernel principal component analysis algorithm is used to uniformly extract the main features of the data in the high-volatility dataset and the general dataset, and then the pre-trained working condition segmentation model is input to output the specific working condition category. The working condition segmentation model is a Bayesian optimized weighted iterative LightGBM model.

[0013] It should be further explained that in step S1, the propulsion motor parameters include the propulsion motor speed and active power, the environmental parameters include wind speed, flow velocity and water surface wave height, and the power battery output parameters include lithium battery output power and fuel cell output power.

[0014] It should be further explained that in step S2, feature extraction and noise reduction of historical operation data include:

[0015] Decompose the original operating data into multiple eigenmodes through variational mode;

[0016] Calculate the maximum information coefficient of each eigenmode and the original data, and retain the eigenmode whose maximum information coefficient exceeds the preset threshold as the main signal;

[0017] The main signal is reconstructed and the interference noise is removed to obtain the noise reduction data set of each sample.

[0018] It should be further explained that the steps of variational mode decomposition include:

[0019] The variational model is iteratively solved by the alternating direction multiplier method and Fourier isometric transform to obtain the decomposed mode function set and the corresponding center frequency;

[0020] The steps to calculate the maximum information coefficient include:

[0021] The eigenmodes of the decomposed propulsion load data and the original data are discretized into a two-dimensional grid space, and the normalized maximum mutual information value is calculated as the maximum information coefficient;

[0022] The steps to remove interference noise include:

[0023] The eigenmodes whose maximum information coefficient is lower than a preset threshold are removed as noise signals, and the signal is reconstructed by linearly superposing the retained eigenmodes.

[0024] It should be further explained that the number of eigenmodes of the variational mode decomposition is set to 3-7, and the noise tolerance parameter is determined by the residual index method.

[0025] It should be further explained that the preset threshold of the maximum information coefficient is determined by the grid search method, and the maximum information coefficient threshold of the propulsion load data is 0.12.

[0026] It should be further explained that in step S3, there are four basic operating conditions, which are classified based on the propulsion load range and the distribution characteristics of the first-order derivative; the basic operating condition labels are basic operating condition 1 to basic operating condition 4.

[0027] It should be further explained that, in step S4, the preset threshold is 40-50.

[0028] It should be further explained that, in step S5, the kernel principal component analysis maps the original data into a high-dimensional space by introducing a radial basis kernel function, and the number of principal components extracted is 6-12.

[0029] It should be further explained that in step S5, the operating condition subdivision model includes the following hierarchical structure:

[0030] The input layer receives the main features extracted by kernel principal component analysis as input features;

[0031] The feature processing layer includes a mutually exclusive feature bundling module and a gradient unilateral sampling module, which respectively perform sparse feature compression and high-information sample screening on the input features to obtain optimized features;

[0032] The decision tree generation layer generates a binary tree structure according to the leaf growth strategy based on the optimized features and hyperparameters provided by the Bayesian optimization framework layer. The information gain of each split node is calculated through the weighted iterative objective function.

[0033] The Bayesian optimization framework layer dynamically adjusts the hyperparameter combination of the decision tree generation layer, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0034] The output layer maps the leaf node weights through a weighted iterative classifier to generate specific working condition category labels.

[0035] It should be further explained that the weighted iterative objective function is:

[0036]

[0037]

[0038]

[0039] Where, is the total number of samples;

[0040] is the sample weight adjustment coefficient;

[0041] is the cross entropy loss function;

[0042] For the The output of a decision tree;

[0043] For samples Vector representation of ;

[0044] is the decision tree complexity regularization term;

[0045] is a constant term;

[0046] For samples confidence level;

[0047] is the average distance between the feature point and other data of the same category;

[0048] is the average distance between the feature point and the feature data points of other categories;

[0049] for Type of working condition;

[0050] and is a constant, set according to the number and size of the training set.

[0051] It should be further explained that in step S5, the training steps of the working condition segmentation model include:

[0052] S501. Obtain historical operating data for all-electric ships with known specific operating conditions. Perform the same operations as steps S1-S4 to obtain a training high-volatility dataset and a training general-purpose dataset. Kernel principal component analysis is then used to uniformly extract key features from both the high-volatility and general-purpose datasets. The datasets are then divided into training, validation, and test sets in a 7:2:1 ratio. Stratified sampling is used to ensure a balanced distribution of samples for each specific operating condition.

[0053] S502. Initialize the hyperparameters of the working condition segmentation model using a Bayesian optimization parameter space search strategy, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0054] S503. Dynamically adjust hyperparameters using the Bayesian optimization framework: Calculate the weighted iterative objective function value on the validation set, generate the posterior probability distribution of the hyperparameters based on the function value changes, and iteratively update the hyperparameters of the working condition segmentation model.

[0055] S504. Input the training set into the working condition segmentation model for training, including:

[0056] The input features are compressed and filtered through the mutually exclusive feature bundling module and the gradient unilateral sampling module;

[0057] In the decision tree generation layer, a binary tree structure is generated according to the leaf growth strategy, and the information gain of each split node is calculated based on the correlation between the first-order derivative of the propulsion load and the environmental parameters;

[0058] Based on sample confidence Dynamically calculate weight coefficients , update the decision tree structure through the weighted iterative objective function and output the specific working condition category label;

[0059] S505. Input the test set into the working condition segmentation model. Calculate the classification accuracy for each specific working condition category based on the output of each sample and the actual specific working condition category. If the classification accuracy does not meet the preset conditions, return to step S503 to readjust the hyperparameters and iterate training. If the preset conditions are met, output the final model to obtain the pre-trained working condition segmentation model.

[0060] It should be further explained that in step S5, after the main features are extracted using kernel principal component analysis, source labels are added to the data from the high-volatility dataset and the general dataset respectively, and then input into the working condition segmentation model.

[0061] It should be further explained that in step S5, the specific operating condition categories include low-speed driving conditions, normal navigation conditions, high-speed cruising conditions, operating conditions and emergency or port conditions, among which low-speed driving conditions, normal navigation conditions and high-speed cruising conditions belong to general operating conditions, and operating conditions, emergency or port conditions belong to high-fluctuation conditions.

[0062] In a second aspect, the present application provides a multi-operating condition identification system for an all-electric ship, which is used to implement the multi-operating condition identification method for the all-electric ship with the integrated energy system, including:

[0063] Data acquisition module, used to obtain historical operating data of all-electric ships;

[0064] The data processing module is used to extract features and reduce noise on historical operation data to obtain a noise-reduced data group for each sample;

[0065] The preliminary clustering module is used to input the propulsion load and the first-order derivative of the propulsion load after noise reduction of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic working conditions, and adding a unique basic working condition label to each noise reduction data group;

[0066] The judgment and classification module is used to determine whether the absolute value of the first-order derivative of the propulsion load of each sample after noise reduction is higher than a preset threshold. The noise reduction data group corresponding to all samples with a positive result is aggregated into a high-fluctuation data set, and the propulsion load, first-order derivative of the propulsion load, and basic working condition label corresponding to all samples with a negative result are aggregated into a general data set.

[0067] The secondary segmentation module is used to uniformly extract the main features of the data in the high-volatility data set and the general data set using the kernel principal component analysis algorithm, and then input the pre-trained working condition segmentation model to output the specific working condition category.

[0068] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to implement the steps of the multi-operating condition identification method for the above-mentioned all-electric ship when executing the computer program.

[0069] In a fourth aspect, the present application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-operating condition identification method for the above-mentioned all-electric ship.

[0070] It can be seen from the above technical solutions that this application has the following advantages:

[0071] 1. This application uses a data reconstruction method based on variational modal decomposition and maximum information coefficient to perform modal decomposition and feature screening on multi-source heterogeneous data such as ship propulsion loads and environmental parameters, effectively eliminating noise interference and retaining key modal components that are strongly related to operating conditions. It solves the problem that traditional methods have difficulty in distinguishing effective signals from noise in nonlinear and non-stationary data, significantly improves the signal-to-noise ratio of the input data, and provides a highly reliable data foundation for subsequent operating condition identification models.

[0072] 2. This application adopts a two-stage recognition framework that collaborates with K-means clustering and Bayesian optimized weighted iterative LightGBM model. It divides the basic working conditions by preliminary clustering of propulsion load and its change rate, and then performs secondary subdivision of high-volatility data in combination with multi-dimensional operating characteristics. It overcomes the problem that a single clustering algorithm is sensitive to sudden working conditions and random environmental changes, and achieves accurate distinction between high-volatility scenarios such as low-speed navigation of ships, operational tasks, and extreme environments, thereby enhancing the robustness and continuity of recognition under complex working conditions.

[0073] 3. This application uses kernel principal component analysis to reduce the dimensionality of high-dimensional features, and combines it with the weighted iterative LightGBM objective function design to dynamically adjust sample weights and model parameters. This solves the problem of insufficient model generalization ability caused by feature redundancy and fixed hyperparameters in traditional methods, enabling the algorithm to adapt to the data distribution characteristics of different working conditions, reducing computational complexity while ensuring recognition accuracy, and improving the real-time and scalability of engineering deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0075] Figure 1 This is a flow chart of a multi-operating condition identification method for an all-electric ship in one embodiment of the present application.

[0076] Figure 2 This is a flowchart of the working condition segmentation model training steps in one embodiment of the present application.

[0077] Figure 3 It is a two-dimensional scatter plot of propulsion power and its change rate drawn using preliminary clustering results in one embodiment of the present application.

[0078] Figure 4 It is a two-dimensional scatter plot of propulsion power and its rate of change drawn using the secondary subdivision result in one embodiment of the present application.

[0079] Figure 5 This is a basic operating condition classification diagram of an all-electric ship under a period of continuous operation in one embodiment of the present application.

[0080] Figure 6 This is a classification diagram of specific working conditions of an all-electric ship under a period of continuous operation in one embodiment of the present application.

[0081] Figure 7 It is a schematic block diagram of a multi-operating condition identification system for an all-electric ship in one embodiment of the present application.

[0082] Figure 8 It is a schematic diagram of the hardware structure of an electronic device in one embodiment of the present application. DETAILED DESCRIPTION

[0083] In order to make the application objectives, features, and advantages of this application more obvious and easy to understand, the technical solutions protected by this application will be clearly and completely described below using specific embodiments and drawings. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0084] The following describes in detail the multi-operating-condition identification method for an all-electric ship involved in this application. Specific details, such as specific system structures and technologies, are provided for illustrative purposes, not for limitation, to facilitate a thorough understanding of the embodiments of this application. However, those skilled in the art will appreciate that this application may also be implemented in other embodiments without these specific details.

[0085] In the multi-operating-state identification method for an all-electric ship involved in this application, the term "comprising" is used to indicate the presence of the described features, entities, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, entities, steps, operations, elements, components, and / or their collections. The terms "include," "comprising," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0086] To facilitate the clear description of the technical solutions of this application, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or order of execution, and the words "first" and "second" do not necessarily mean different.

[0087] The phrases "one embodiment" or "some embodiments" described in this application mean that the specific features, structures, or characteristics described in the embodiment are included in one or more embodiments of the application. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in other embodiments," etc. that appear in different places in this application do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized.

[0088] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0089] The multi-operating condition identification method of the all-electric ship provided in the embodiment of the present application is executed by a computer device, and accordingly, the multi-operating condition identification system of the all-electric ship runs in the computer device.

[0090] The following are some explanations of terms in this plan to facilitate a better understanding of this plan:

[0091] K-means clustering algorithm: The K-means clustering algorithm is an unsupervised learning algorithm that iteratively divides samples into K clusters, minimizing the Euclidean distance of samples within the same cluster and maximizing the distance between different clusters. The number of clusters K must be specified in advance.

[0092] Kernel principal component analysis (KPCA) is a nonlinear extension of principal component analysis (PCA). It uses a kernel function (such as the radial basis kernel) to map data into a high-dimensional feature space and then performs linear dimensionality reduction. This algorithm can capture complex nonlinear structures in the data and is suitable for feature extraction and visualization.

[0093] Bayesian Optimization Weighted Iterative LightGBM Model: The LightGBM (Light Gradient Boosting Machine) model is an efficient machine learning framework based on the Gradient Boosting Decision Tree (GBDT). It accelerates training through a histogram algorithm, reduces memory usage through a leaf-based growth strategy, and supports parallel learning and categorical feature processing.

[0094] Weighted Iteration refers to dynamically adjusting the weights of samples or features during model training. For example, increasing the weight of samples with incorrect predictions to increase the model's attention to difficult examples.

[0095] Bayesian Optimization is a hyperparameter tuning method that intelligently selects parameter combinations by constructing a probabilistic proxy model of the objective function (such as a Gaussian process) and an acquisition function (such as expected improvement EI) to find the global optimal solution with fewer iterations. It is suitable for automated tuning of hyperparameters such as learning rate and tree depth in LightGBM.

[0096] Figure 1 This is a flow chart of a multi-operating-condition identification method for an all-electric ship according to an embodiment of the present application. Figure 1 The execution subject can be a multi-operating state identification system for an all-electric ship. According to different requirements, the order of the steps in the flowchart can be changed, and some can be omitted.

[0097] like Figure 1 As shown, the multi-operating condition identification method of the all-electric ship includes:

[0098] Step S1, obtaining the historical operating data of the all-electric ship. The historical operating data consists of samples at each time stamp. Each sample includes the propulsion load, the first-order derivative of the propulsion load, the propulsion motor parameters, the environmental parameters and the power battery output parameters at the corresponding time stamp.

[0099] By systematically collecting full-dimensional historical operating data such as propulsion load, motor speed, environmental parameters and battery output, the multi-physical field coupling characteristics of the ship's power system in the time series are fully covered, providing underlying data support with temporal and spatial correlation for subsequent working condition division, effectively avoiding the problem of misjudgment of working conditions caused by single parameter analysis, and at the same time realizing standardized preprocessing of massive heterogeneous data through structured storage.

[0100] In some specific embodiments, the propulsion motor parameters include propulsion motor speed and active power, the environmental parameters include wind speed, flow rate and water surface wave height, and the power battery output parameters include lithium battery output power and fuel cell output power.

[0101] By clearly defining the propulsion motor parameters including speed and active power, the environmental parameters covering wind speed, flow velocity and wave height, and the power battery parameters including the output power of lithium batteries and fuel cells, a multi-dimensional data acquisition system covering mechanical, electrical and environmental coupling effects is constructed. This not only ensures the complete coverage of the energy transfer chain of the ship power system by feature engineering, but also eliminates the magnitude differences of cross-sensor data through the standardized definition of physical dimensions, providing feature inputs with engineering interpretability for subsequent clustering and classification models, and enhancing the mapping consistency between the working condition identification results and the actual operation scenarios.

[0102] Step S2: extract features and reduce noise on the historical operation data to obtain a noise-reduced data set for each sample, including the noise-reduced propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters.

[0103] Through feature extraction and noise reduction processing, high-frequency noise and interference signals in the original data are effectively eliminated, the data quality is improved, and more accurate and stable input information is provided for subsequent clustering and segmentation models.

[0104] In some specific embodiments, feature extraction and noise reduction of historical operation data include:

[0105] Decompose the original operating data into multiple eigenmodes through variational mode;

[0106] Calculate the maximum information coefficient of each eigenmode and the original data, and retain the eigenmode whose maximum information coefficient exceeds the preset threshold as the main signal;

[0107] The main signal is reconstructed and the interference noise is removed to obtain the noise reduction data set of each sample.

[0108] Variational Mode Decomposition (VMD) is an adaptive, fully non-recursive signal processing algorithm. It decomposes a signal into multiple intrinsic modes (IMFs) with different center frequencies by constructing a constrained variational problem. Each eigenmode is closely distributed around its center frequency in the frequency domain. The optimal solution of the variational model is iteratively searched to determine the center frequency and bandwidth of each intrinsic mode function (IMF). It is suitable for non-stationary signal processing.

[0109] The Maximum Information Coefficient (MIC) is a statistic that measures the nonlinear correlation between variables. It is calculated by grid division and mutual information normalization. Its value range is [0,1]. Larger values ​​indicate stronger correlation. It is suitable for feature selection in high-dimensional data mining.

[0110] Through variational modal decomposition, the original signal is adaptively decomposed into multi-scale intrinsic modes, and the maximum information coefficient is combined to quantify the correlation between each mode and the original data, thereby achieving accurate separation of noise signals and effective features. It not only overcomes the empirical dependence of basis function selection in traditional wavelet decomposition, but also retains the time domain integrity of propulsion load mutation events through the linear superposition mechanism in the signal reconstruction process, thereby avoiding excessive smoothing of key operating condition characteristics in the denoising process, and providing subsequent algorithms with high-quality input data with both temporal continuity and feature significance.

[0111] In some specific embodiments, the variational mode decomposition step includes:

[0112] The variational model is iteratively solved by the alternating direction multiplier method and Fourier isometric transform to obtain the decomposed mode function set and the corresponding center frequency;

[0113] The steps to calculate the maximum information coefficient include:

[0114] The eigenmodes of the decomposed propulsion load data and the original data are discretized into a two-dimensional grid space, and the normalized maximum mutual information value is calculated as the maximum information coefficient;

[0115] The steps to remove interference noise include:

[0116] The eigenmodes whose maximum information coefficient is lower than a preset threshold are removed as noise signals, and the signal is reconstructed by linearly superposing the retained eigenmodes.

[0117] The Alternating Direction Method of Multipliers (ADMM) is a distributed optimization algorithm suitable for large-scale, decomposable convex optimization problems. It decomposes the original problem into multiple subproblems, solves them alternately, and introduces Lagrange multipliers to coordinate the constraints between the subproblems. It has the characteristics of fast convergence and parallel computing friendliness.

[0118] The variational model is a mathematical model built based on the variational principle. It transforms the problem into an optimization problem by minimizing the energy functional and is widely used in image processing, physical modeling and other fields.

[0119] In the process of calculating variational mode decomposition, we first need to construct the variational problem, introduce the Lagrange multiplication operator and Quadratic penalty factor Construct the augmented Lagrangian equation:

[0120]

[0121] In the formula, K is a positive integer, which indicates the number of modes to be decomposed. Represents the decomposed mode function set and the corresponding center frequency; is the Dirac function; is the convolution operator; f is the input signal;

[0122] Then, the alternating direction multiplier method and Fourier isometric transform are used to solve the equations to obtain the modal components and their center frequencies:

[0123]

[0124]

[0125]

[0126] Where, n represents the number of iterations, represents the noise tolerance, , , Represents the form after Fourier transformation.

[0127] The solution to the maximum mutual information coefficient is based on mutual information (MI). MI is an information measure that indicates the amount of information a random variable contains about another random variable and can be expressed as:

[0128]

[0129] in x, y Represents two different variables, p(x,y) is the joint probability density between variables, p(x)p(y) represents the boundary density;

[0130] The basic calculation principles of the maximum information coefficient include:

[0131] (1) Discretize the two variables into two-dimensional space, x, y The scatter plot is constructed i List, j Grid the rows and find the maximum mutual information value;

[0132] (2) Normalize the maximum mutual information value;

[0133] (3) The maximum mutual information value at different scales is selected as the MIC value, and the calculation formula is:

[0134]

[0135] Where, a, b Indicates x, y The number of grids divided in the direction, B It is a variable, and its numerical value is approximately 0.6 of the data volume. The larger the MIC value, the higher the correlation between the two variables.

[0136] The constrained optimization problem of variational modal decomposition is solved by the alternating direction multiplier method, and the frequency domain iterative calculation is accelerated by Fourier isometric transform, which significantly improves the convergence speed and stability of complex signal decomposition. At the same time, the maximum mutual information value is calculated by gridded spatial discretization, which solves the numerical instability problem of mutual information estimation of continuous variables. The optimal trade-off between signal fidelity and computational complexity is achieved through the directional elimination of noise modes and the linear reconstruction of retained modes, providing a theoretically rigorous and engineering feasible technical path for real-time data processing in industrial scenarios.

[0137] In some specific embodiments, the number of eigenmodes of variational mode decomposition is set to 3-7, and the noise tolerance parameter is determined by a residual index method.

[0138] Among them, the Residual Index Method evaluates model performance or detects outliers by analyzing the statistical characteristics (such as residual distribution, trend changes, etc.) of the model prediction residual (the difference between the predicted value and the true value). It is often used in time series analysis and quality control.

[0139] The number of modes and noise tolerance parameters of variational modal decomposition are dynamically determined through the residual index method, which achieves adaptive matching of decomposition granularity and data characteristics, avoids the problem of over-decomposition or under-decomposition caused by artificial experience-based preset parameters, and limits the number of modes to a reasonable range of 3-7. This not only ensures the completeness of signal feature extraction, but also prevents the surge in algorithm computational load due to modal redundancy, providing a highly universal and robust signal preprocessing solution for all-electric ship data under different navigation environments.

[0140] In some specific embodiments, the preset threshold of the maximum information coefficient is determined by a grid search method, and the maximum information coefficient threshold of the propulsion load data is 0.12.

[0141] Grid Search is a hyperparameter optimization method that selects the optimal parameters by traversing a predefined parameter combination space, training models one by one, and evaluating performance. It is computationally intensive but easy to parallelize and is suitable for scenarios with low parameter dimensions.

[0142] The grid search method was used to traverse the threshold candidate set with the maximum information coefficient. Combined with cross-validation, the signal reconstruction error and classification model accuracy under different thresholds were evaluated. The optimal threshold of 0.12 was determined in a data-driven manner. This not only avoided the subjective bias of manually setting the threshold, but also ensured the stability of the effective mode through quantitative screening criteria. The noise reduction process can maintain high robustness under different ship load fluctuation scenarios, providing a reproducible and verifiable parameter configuration benchmark for engineering deployment.

[0143] In step S3, the propulsion load and the first-order derivative of the propulsion load after noise reduction of each sample are input into the K-means clustering algorithm for preliminary clustering, classified into multiple basic working conditions, and a unique basic working condition label is added to each noise reduction data group.

[0144] The clustering steps of the K-means algorithm include:

[0145] Assume sample data Contains n objects, each of which has m dimensional attributes, and initializes k cluster centers , define the loss function:

[0146]

[0147] Let t=0,1,2...d be the number of iterations, and repeat the following process until the loss function converges:

[0148] For each sample xi, assign it to the nearest center Ci:

[0149]

[0150] For each cluster, recalculate the center point of the cluster:

[0151]

[0152] The K-means algorithm is used to perform unsupervised clustering of the propulsion load and its derivatives after noise reduction, and the initial partition boundary of the four-dimensional feature space is constructed through the Euclidean distance metric, realizing coarse-grained modeling of the typical operating mode of the ship. This process not only reduces the cost of manual labeling, but also enhances the physical interpretability of the basic operating condition category through iterative optimization of the cluster center, establishing a stable classification benchmark framework for secondary subdivision.

[0153] In some specific embodiments, the basic operating conditions are divided into 4 categories, and the classification is based on the distribution characteristics of the propulsion load range and the first-order derivative; the basic operating condition labels are basic operating condition 1 to basic operating condition 4.

[0154] A hierarchical operating condition classification system was constructed by dividing four basic operating conditions and defining unique labels. The propulsion load range reflects the steady-state energy demand characteristics, and the first-order derivative distribution represents the dynamic response characteristics. The combination of the two realizes the deep integration of the physical mechanism and data characteristics of the ship operation mode, which not only provides an explainable classification basis for unsupervised clustering, but also avoids the cross-contamination problem of multi-label data through label uniqueness constraints, providing a highly consistent pseudo-annotated dataset for secondary segmentation model training.

[0155] Step S4: Determine whether the absolute value of the first-order derivative of the propulsion load of each sample after denoising is higher than a preset threshold. Aggregate the denoised data groups corresponding to the samples with all yes results into a high-fluctuation data set, and aggregate the propulsion load, first-order derivative of the propulsion load, and basic operating condition labels corresponding to the samples with all no results into a general data set.

[0156] By using a preset threshold, the absolute value of the first-order derivative of the propulsion load is subjected to binary classification screening, and the high-fluctuation working condition is physically separated from the steady-state working condition, which significantly reduces the interference of dynamic mutation conditions on the conventional classification model. At the same time, through the differentiated processing strategy of the data set, the multi-dimensional original parameters of the high-fluctuation data are retained to capture transient characteristics, while only key parameters are retained for general data to improve computational efficiency, achieving a dynamic balance between computing resource allocation and model accuracy, and providing a structured input basis for subsequent multimodal data fusion and refined modeling.

[0157] In some specific embodiments, the preset threshold is 40-50.

[0158] By setting the absolute value threshold of the first-order derivative of the propulsion load to the engineering experience range of 40-50, the balance between the sensitivity and specificity of this threshold to high-fluctuation conditions was verified based on historical data statistics. It not only effectively captures the boundary characteristics of transient events such as sudden acceleration and emergency braking of ships, but also avoids the problem of false triggering caused by conventional wave disturbances. Through threshold quantization, the standardized operation of dynamic condition segmentation is achieved, providing clear and configurable screening rules for the subsequent construction of high-fluctuation data sets.

[0159] Step S5, performing secondary subdivision, including:

[0160] The kernel principal component analysis algorithm is used to uniformly extract the main features of the data in the high-volatility dataset and the general dataset, and then the pre-trained working condition segmentation model is input to output the specific working condition category. The working condition segmentation model is a Bayesian optimized weighted iterative LightGBM model.

[0161] Different from the traditional principal component analysis algorithm, kernel principal component analysis uses kernel techniques to simplify calculations, maps the low-dimensional linearly inseparable input space to a linearly separable high-dimensional feature space, and then uses PCA to reduce the dimensionality of the feature space, that is, first increase the dimensionality and then reduce the dimensionality.

[0162] The basis of kernel principal component analysis is the inner product transformation of vectors. Represents the original data, with a dimension of k. First, the original data is mapped to a high-dimensional feature space, using Represents the mapped data, whose dimension is d, and then PCA is used to reduce the dimension in the feature space to calculate The covariance matrix of :

[0163]

[0164] Solve for eigenvectors and calculate principal component directions p :

[0165]

[0166]

[0167] Where, represents the coefficient vector;

[0168] Finally, the kernel function is introduced Solve the formula and project the original data along the principal component direction to extract the main features.

[0169] The sampling technology of the LightGBM model uses unilateral sampling technology to filter data based on the gradient value of the data, retaining large gradient samples, discarding useless samples, and sampling small gradient samples, thereby reducing the amount of data for each round of training and improving the training speed. The specific calculation steps are:

[0170] (1) Calculate the absolute value of the gradient of the training data and sort it from large to small according to the absolute value;

[0171] (2) Select the first a% of samples as the large gradient sample set A, and select b% of the remaining samples to form the small gradient sample set B;

[0172] (3) Multiply the small gradient sample set B by the coefficient (1-a) / b to increase the attention;

[0173] (4) Recombining A and B to calculate information gain:

[0174]

[0175] in , , , , n is the total number of samples, G(.) represents the sample gradient, d is the node number, and j is the split eigenvalue;

[0176] The LightGBM model's mutually exclusive feature bundling technology can effectively merge multiple mutually exclusive features, thereby reducing the number of features. By introducing a deviation constant, the sparse matrix is ​​converted into a dense matrix, thereby reducing the number of features. This reduces the feature dimension without losing feature information, thereby reducing computational costs.

[0177] The LightGBM model uses a histogram algorithm instead of a pre-sorting algorithm to find split points, effectively utilizing the sparsity of high-dimensional data and converting continuous features into discrete values, which greatly reduces the amount of computation. Specifically, it discretizes continuous floating-point feature values ​​into h integers and constructs a histogram of width h; when traversing the data, it uses the discretized values ​​as indices to accumulate statistics in the histogram; and traverses to find the optimal split point based on the discrete values ​​of the histogram.

[0178] The LightGBM model adopts a leaf-based growth strategy. At each split, the leaf with the highest gain among all leaves at the same level is selected. This is generally the leaf with the largest amount of data. The cycle is then repeated until a tree is formed. In practical applications, in order to avoid overfitting caused by excessive depth, the LightGBM model sets a maximum depth.

[0179] The Bayesian optimization algorithm is a parameter optimization algorithm that consists of two key parts: a probabilistic proxy model and an acquisition function. First, the prior distribution of the proxy model is initialized, the acquisition function is maximized and the evaluation points are obtained. Then, the objective function value is calculated and the proxy function is updated. Finally, the optimal parameter distribution is obtained through iteration.

[0180] Through kernel principal component analysis, nonlinear dimensionality reduction is performed on high-dimensional heterogeneous data, which effectively extracts the potential correlation characteristics of propulsion load fluctuations, environmental parameters and battery output, and solves the problem of feature loss of traditional linear dimensionality reduction methods under high-fluctuation working conditions. At the same time, combined with the Bayesian optimization framework, the hyperparameter combination of the LightGBM model is dynamically adjusted. Through the global optimization strategy guided by the probability model, the discrimination accuracy of the classification boundary in the overlapping area of ​​complex working conditions is significantly improved, and finally the fine-grained division of five specific working conditions is achieved, providing high-reliability working condition label support for ship energy efficiency optimization and fault warning.

[0181] In some specific embodiments, the operating condition subdivision model includes the following hierarchical structure:

[0182] The input layer receives the main features extracted by kernel principal component analysis as input features;

[0183] The feature processing layer includes a mutually exclusive feature bundling module and a gradient unilateral sampling module, which respectively perform sparse feature compression and high-information sample screening on the input features to obtain optimized features;

[0184] The decision tree generation layer generates a binary tree structure according to the leaf growth strategy based on the hyperparameters provided by the Bayesian optimization framework layer. The information gain of each split node is calculated through the weighted iterative objective function.

[0185] The Bayesian optimization framework layer dynamically adjusts the hyperparameter combination of the decision tree generation layer, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0186] The output layer maps the leaf node weights through a weighted iterative classifier to generate specific working condition category labels.

[0187] By constructing a multi-level model architecture that includes feature bundling, gradient sampling, and Bayesian optimization, effective compression of high-dimensional sparse features and improvement of information density are achieved. The mutually exclusive feature bundling module reduces the dimensional redundancy of environmental parameters and battery output parameters, the gradient unilateral sampling module focuses on high-information samples to accelerate model convergence, and the Bayesian optimization framework dynamically adjusts the hyperparameter space through Gaussian process modeling. The three work together to significantly improve the model's generalization ability and computational efficiency under limited training data.

[0188] In some specific embodiments, the weighted iterative objective function is:

[0189]

[0190]

[0191]

[0192] Where, is the total number of samples;

[0193] is the sample weight adjustment coefficient;

[0194] is the cross entropy loss function;

[0195] For the The output of a decision tree;

[0196] For samples Vector representation of ;

[0197] is the decision tree complexity regularization term;

[0198] is a constant term;

[0199] For samples confidence level;

[0200] is the average distance between the feature point and other data of the same category;

[0201] is the average distance between the feature point and the feature data points of other categories;

[0202] for Type of working condition;

[0203] and is a constant, set according to the number and size of the training set.

[0204] By introducing a confidence-weighted objective function design, the sample classification confidence is associated with the feature space distribution, and the loss weights of different samples are dynamically adjusted. This allows the model to focus more on samples with fuzzy boundaries during training. At the same time, regularization terms and complexity constraints are used to suppress the risk of overfitting. This design effectively alleviates the interference of category imbalance and noise labels on model training, and improves the recognition recall rate of low-frequency but high-value categories such as emergency conditions and port conditions.

[0205] In some specific embodiments, the training step of the operating condition segmentation model includes:

[0206] S501. Obtain historical operating data for all-electric ships with known specific operating conditions. Perform the same operations as steps S1-S4 to obtain a training high-volatility dataset and a training general-purpose dataset. Kernel principal component analysis is then used to uniformly extract key features from both the high-volatility and general-purpose datasets. The datasets are then divided into training, validation, and test sets in a 7:2:1 ratio. Stratified sampling is used to ensure a balanced distribution of samples for each specific operating condition.

[0207] S502. Initialize the hyperparameters of the working condition segmentation model using a Bayesian optimization parameter space search strategy, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0208] S503. Dynamically adjust hyperparameters using the Bayesian optimization framework: Calculate the weighted iterative objective function value on the validation set, generate the posterior probability distribution of the hyperparameters based on the function value changes, and iteratively update the hyperparameters of the working condition segmentation model.

[0209] S504. Input the training set into the working condition segmentation model for training, including:

[0210] The input features are compressed and filtered through the mutually exclusive feature bundling module and the gradient unilateral sampling module;

[0211] In the decision tree generation layer, a binary tree structure is generated according to the leaf growth strategy, and the information gain of each split node is calculated based on the correlation between the first-order derivative of the propulsion load and the environmental parameters;

[0212] Based on sample confidence Dynamically calculate weight coefficients , update the decision tree structure through the weighted iterative objective function and output the specific working condition category label;

[0213] S505. Input the test set into the working condition segmentation model. Calculate the classification accuracy for each specific working condition category based on the output of each sample and the actual specific working condition category. If the classification accuracy does not meet the preset conditions, return to step S503 to readjust the hyperparameters and iterate training. If the preset conditions are met, output the final model to obtain the pre-trained working condition segmentation model.

[0214] A stratified sampling strategy is used to balance the working condition distribution of the training set, validation set, and test set, avoiding model preference problems caused by data skew. At the same time, Bayesian optimization is combined to perform probability-guided global search in the hyperparameter space. The contradiction between exploration and utilization is balanced by the acquisition function, achieving coordinated optimization of the model convergence speed and the quality of the optimal solution. Finally, iterative verification is used to ensure the stability of the model on unknown data, providing a highly reliable working condition segmentation model for actual deployment.

[0215] In some specific embodiments, after extracting the main features using kernel principal component analysis, source labels are added to the data from the high-volatility dataset and the general dataset respectively, and then input into the operating condition segmentation model.

[0216] By adding source labels to high-volatility and general data sets and embedding prior knowledge of data distribution in the feature vectors after kernel principal component analysis, the model can distinguish the feature space topological structure of samples with different volatility characteristics, enhancing the differentiated processing capability of transient and steady-state conditions in the classification decision-making process. At the same time, by guiding the feature weight distribution through label information, the model's sensitivity and discrimination accuracy for short-term high-load modes such as operating conditions are improved.

[0217] In some specific embodiments, kernel principal component analysis maps the original data to a high-dimensional space by introducing a radial basis kernel function, and the number of principal components extracted is 6-12.

[0218] The radial basis function (RBF) is a symmetric kernel function based on the distance between samples. It is widely used in machine learning models such as support vector machines (SVMs) and kernel methods to deal with nonlinear separable problems.

[0219] The original data is nonlinearly mapped to the high-dimensional reproducing kernel Hilbert space through the radial basis kernel function, which solves the dimensionality reduction failure problem of traditional principal component analysis in multi-condition ship data caused by nonlinear feature coupling. At the same time, the number of principal components is expanded to 6-12, which fully retains the interaction information between the dynamic characteristics of the propulsion system and environmental interference, provides a low-dimensional and highly discriminative feature expression for the subsequent classification model, and significantly reduces the computational complexity and memory usage of the model training.

[0220] In some specific embodiments, specific operating condition categories include low-speed driving conditions, normal navigation conditions, high-speed cruising conditions, operating conditions, and emergency or port conditions, among which low-speed driving conditions, normal navigation conditions, and high-speed cruising conditions belong to general operating conditions, and operating conditions, emergency or port conditions belong to high-fluctuation conditions.

[0221] By strictly matching five specific working conditions with the actual operating scenarios of ships, a direct mapping relationship between algorithm output and engineering decisions is established. The operating conditions cover typical high-power demand scenarios such as crane operation and cargo loading and unloading, while emergency or port conditions focus on sudden failures and berthing energy consumption management. This classification system not only meets the multi-objective needs of ship energy efficiency optimization, but also provides a unified and standardized working condition management interface for different departments.

[0222] In a specific embodiment, a multi-operating condition identification method for an all-electric ship includes:

[0223] Step S1, obtaining historical operating data of the all-electric ship, the historical operating data consisting of samples at various timestamps, each sample including propulsion load, first-order derivative of propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters at the corresponding timestamp;

[0224] Among them, propulsion motor parameters include propulsion motor speed and active power, environmental parameters include wind speed, flow velocity and water surface wave height, and power battery output parameters include lithium battery output power and fuel cell output power;

[0225] Step S2: extracting features and reducing noise from the historical operating data to obtain a noise-reduced data set for each sample, including the noise-reduced propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters;

[0226] Perform feature extraction and noise reduction on historical operation data, including:

[0227] The original operating data is decomposed into multiple eigenmodes through variational mode decomposition. The number of eigenmodes of variational mode decomposition is set to 3-7. The noise tolerance parameter is determined by the residual index method. The steps of variational mode decomposition include:

[0228] The variational model is iteratively solved by the alternating direction multiplier method and Fourier isometric transform to obtain the decomposed mode function set and the corresponding center frequency;

[0229] Calculate the maximum information coefficient of each eigenmode and the original data, and retain the eigenmode whose maximum information coefficient exceeds the preset threshold as the main signal. The preset threshold of the maximum information coefficient is determined by the grid search method. The maximum information coefficient threshold of the propulsion load data is 0.12. The steps for calculating the maximum information coefficient include:

[0230] The eigenmodes of the decomposed propulsion load data and the original data are discretized into a two-dimensional grid space, and the normalized maximum mutual information value is calculated as the maximum information coefficient;

[0231] Table 1 shows the maximum information coefficient values ​​of each eigenmode of the propulsion composite data and the original propulsion load data in this embodiment. Based on the data in the table, it can be concluded that the maximum information values ​​of different modes are quite different. The smaller the value, the weaker the correlation between the mode and the original propulsion load, and the lower the information content.

[0232] Table 1 Maximum information coefficient of eigenmode

[0233]

[0234] Reconstruct the main signal and remove the interference noise to obtain the noise reduction data set of each sample. The steps of removing the interference noise include:

[0235] The eigenmodes whose maximum information coefficient is lower than a preset threshold are removed as noise signals, and the signal is reconstructed by linearly superposing the retained eigenmodes;

[0236] The formula for reconstructing the main signal and removing the interference noise is:

[0237]

[0238] Where, For the i The preset threshold value of the maximum information coefficient corresponding to each eigenmode;

[0239] After reconstructing the main signal and removing the interference noise, the maximum information coefficient of each input feature is improved as shown in Table 2:

[0240] Table 2. Maximum information coefficient improvement of input features

[0241]

[0242] It can be seen that the maximum information coefficients of the processed ship characteristic data have been greatly improved. Among them, the daily load and wind speed have increased the most, reaching 28.5% and 26.8% respectively, followed by flow velocity, wave height, fuel cell output, etc. The propulsion motor-related parameters have a relatively large maximum information coefficient, so the improvement is relatively small, about 10%. The smallest improvement is in the power lithium battery output. This shows that the data processing method of this scheme can retain the data information to a large extent and eliminate irrelevant or interfering components.

[0243] Step S3: Input the propulsion load and the first-order derivative of the propulsion load after noise reduction of each sample into the K-means clustering algorithm for preliminary clustering, classify them into multiple basic working conditions, and add a unique basic working condition label to each noise reduction data group;

[0244] Among them, there are 4 basic working conditions, which are classified based on the distribution characteristics of propulsion load range and first-order derivative; the basic working condition labels are basic working condition 1 to basic working condition 4

[0245] Step S4, determining whether the absolute value of the first-order derivative of the propulsion load of each sample after noise reduction is higher than a preset threshold, the preset threshold being 48;

[0246] The noise reduction data group corresponding to the samples with all the results being yes is summarized as the high-fluctuation data set, and the propulsion load, the first-order derivative of the propulsion load, and the basic working condition label corresponding to the samples with all the results being no are summarized as the general data set;

[0247] Step S5, performing secondary subdivision, including:

[0248] The kernel principal component analysis algorithm is used to uniformly extract the main features of the data in the high-volatility dataset and the general dataset. Then, source labels are added to the data from the high-volatility dataset and the general dataset respectively. The data are then input into the pre-trained working condition segmentation model to output specific working condition categories. The specific working condition categories include low-speed driving conditions, normal sailing conditions, high-speed cruising conditions, operating conditions, and emergency or port conditions. Among them, low-speed driving conditions, normal sailing conditions, and high-speed cruising conditions are general operating conditions, while operating conditions, emergency or port conditions are high-volatility conditions. The kernel principal component analysis maps the original data into a high-dimensional space by introducing the radial basis kernel function, and the number of principal components extracted is 6-12.

[0249] The working condition segmentation model is a Bayesian optimized weighted iterative LightGBM model, which contains the following hierarchical structure:

[0250] The input layer receives the main features extracted by kernel principal component analysis as input features;

[0251] The feature processing layer includes a mutually exclusive feature bundling module and a gradient unilateral sampling module, which respectively perform sparse feature compression and high-information sample screening on the input features to obtain optimized features;

[0252] The decision tree generation layer generates a binary tree structure according to the leaf growth strategy based on the hyperparameters provided by the Bayesian optimization framework layer. Each split node calculates the information gain through a weighted iterative objective function. The weighted iterative objective function is:

[0253]

[0254]

[0255]

[0256] Where, is the total number of samples;

[0257] is the sample weight adjustment coefficient;

[0258] is the cross entropy loss function;

[0259] For the The output of a decision tree;

[0260] For samples Vector representation of ;

[0261] is the decision tree complexity regularization term;

[0262] is a constant term;

[0263] For samples confidence level;

[0264] is the average distance between the feature point and other data of the same category;

[0265] is the average distance between the feature point and the feature data points of other categories;

[0266] for Type of working condition;

[0267] and is a constant, set according to the number and scale of the training set;

[0268] The Bayesian optimization framework layer dynamically adjusts the hyperparameter combination of the decision tree generation layer, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0269] The output layer maps the leaf node weights through a weighted iterative classifier to generate specific working condition category labels;

[0270] Figure 2 This is a flow chart of the working condition segmentation model training steps in this embodiment. Figure 2 As shown in Figure 2, the training steps of the working condition segmentation model include:

[0271] S501. Obtain historical operating data for all-electric ships with known specific operating conditions. Perform the same operations as steps S1-S4 to obtain a training high-volatility dataset and a training general-purpose dataset. Kernel principal component analysis is then used to uniformly extract key features from both the high-volatility and general-purpose datasets. The datasets are then divided into training, validation, and test sets in a 7:2:1 ratio. Stratified sampling is used to ensure a balanced distribution of samples for each specific operating condition.

[0272] S502. Initialize the hyperparameters of the working condition segmentation model using a Bayesian optimization parameter space search strategy, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio;

[0273] S503. Dynamically adjust hyperparameters using the Bayesian optimization framework: Calculate the weighted iterative objective function value on the validation set, generate the posterior probability distribution of the hyperparameters based on the function value changes, and iteratively update the hyperparameters of the working condition segmentation model.

[0274] In this example, the final hyperparameter values ​​obtained after iterative update are shown in Table 3:

[0275] Table 3 Hyperparameter values ​​and optimal values

[0276]

[0277] S504. Input the training set into the working condition segmentation model for training, including:

[0278] The input features are compressed and filtered through the mutually exclusive feature bundling module and the gradient unilateral sampling module;

[0279] In the decision tree generation layer, a binary tree structure is generated according to the leaf growth strategy, and the information gain of each split node is calculated based on the correlation between the first-order derivative of the propulsion load and the environmental parameters;

[0280] Based on sample confidence Dynamically calculate weight coefficients , update the decision tree structure through the weighted iterative objective function and output the specific working condition category label;

[0281] S505. Input the test set into the working condition segmentation model. Calculate the classification accuracy for each specific working condition category based on the output of each sample and the actual specific working condition category. If the classification accuracy does not meet the preset conditions, return to step S503 to readjust the hyperparameters and iterate training. If the preset conditions are met, output the final model to obtain the pre-trained working condition segmentation model.

[0282] The multi-operating condition identification method of this embodiment is used to build an operating condition identification framework. The specific parameters of the operating condition identification framework are set as follows:

[0283] The variational mode decomposition is constructed by the subpross module, in which the optimal variational mode decomposition is used to determine the number of decomposed eigenmodes KVMD=7*(3~7), that is, each input feature is decomposed into 3 to 7 eigenmodes, and then the residual index method REI is used to determine the noise tolerance tau;

[0284] The penalty parameter α is set in the range of 1500-2500, and a grid search is performed with a step size of 100;

[0285] Kernel PCA was constructed using the scikit-learn library. The number of principal components was determined to be 12 based on the actual prediction effect, variance, and selection parameters.

[0286] K-means determines the number of categories K based on the silhouette coefficient. The closer the silhouette coefficient is to 1, the better the classification effect. Then the number of categories K is selected through an iterative method, as shown in Table 4:

[0287] Table 4 Working condition identification framework parameter setting table

[0288]

[0289] Simulation experiments were conducted using the working condition identification method of this embodiment and a commonly used model classification method. The accuracy of ship working condition identification was evaluated using internal indicators. That is, the quality of the algorithms was compared using the attribute characteristics of the data set. The clustering quality was determined by comparing the average similarity between clusters or the average similarity within clusters.

[0290] Commonly used model classification methods include K-means clustering, K-nearest neighbor algorithm, random forest method, and LightGBM;

[0291] The specific evaluation criteria are the Calinski-Harabaz Index (CH Index), the Davies-Bouldin Index (DB Index), and the stability score.

[0292] Among them, the CH index calculates the ratio of the intra-cluster distance to the inter-cluster distance in a form similar to variance:

[0293]

[0294] Where x is the data set, N is the data capacity, K is the number of categories; Bk and Wk are the covariance matrices between and within the class respectively, and the calculation formula is:

[0295] ;

[0296] The DB index is an algorithm for evaluating clustering metrics. This method calculates the average sum of the intra-class distances of any two categories divided by the center distance of the two categories and finds the maximum value:

[0297]

[0298]

[0299] in s i represents the sample dispersion in category i, M ij represents the distance between the i-th class and the j-th class;

[0300] The smaller the DB, the smaller the intra-class distance and the larger the inter-class distance, which means the better the clustering effect;

[0301] The specific calculation process of the stability score is as follows: first, all models are tested using the same dataset, and each model is run 10 times; then a certain amount of noise interference is added to the dataset, and the model is run 10 times on the perturbed dataset. The average similarity between all running results is calculated using the formula:

[0302]

[0303] Where N is the number of runs, C i is the result of the i-th run, C ref is the reference result (the first result is used as a reference);

[0304] The stability coefficient ranges from [0,1]. A value of 1 indicates that the clustering result is completely consistent with the first one, and the model stability is high; a value of 0 indicates that the result is irrelevant to the first result, and the model stability is low.

[0305] The simulation experiment was run in Python 3.7, using libraries including TensorFlow 1.13.1, Keras 2.24, and Sklearn 0.23.2. The prediction model was run on a 12th-generation Intel(R) Core(TM) i5-12400F 2.50 GHz processor with 16GB of RAM and a 64-bit operating system. Validation was performed using a test set of 15,000 data points covering a wide range of ship navigation conditions, encompassing most scenarios encountered during ship operation.

[0306] In this embodiment, the two-dimensional scatter plot of propulsion power and its change rate drawn using the preliminary clustering results is as follows: Figure 3 As shown, Figure 3 The horizontal axis is the propulsion power and the vertical axis is the corresponding derivative, that is, the rate of change. Points of different colors represent different categories. Figure 3 The characteristics of the K-means preliminary classification can be summarized as follows: the data set is divided into four basic operating condition categories based on propulsion power, and the change rate has little effect on category differentiation;

[0307] The basic parameters of the four basic operating condition categories formed by preliminary clustering are shown in Table 5:

[0308] Table 5 Basic operating parameters

[0309]

[0310] Summarizing Table 5, we can see that from basic operating condition 1 to basic operating condition 4, the average propulsion load of the four basic operating conditions gradually increases. The basic operating condition intervals are adjacent to each other, and the differences in average and maximum change rates are not significant. Among them, basic operating condition 3 accounts for the largest amount of data, reaching 40%. At the same time, from the data proportion and fluctuation data proportion in Table 5, it can be seen that although basic operating condition 4 has a small amount of data, it has a high proportion of fluctuation data;

[0311] The two-dimensional scatter plot of propulsion power and its rate of change drawn using the secondary subdivision results is as follows: Figure 4 As shown, by comparison Figure 4 and Figure 3 It can be seen that in Figure 4 In the initial clustering, the basic operating condition categories of the horizontal distribution were simplified from 4 to 3; the data points with high change rates and the data with low change rates were reclassified; the number of categories increased, forming two new operating conditions with high volatility;

[0312] The basic parameters of the five specific working condition categories formed by the secondary subdivision are shown in Table 6:

[0313] Table 6 Secondary subdivision stage operating parameters

[0314]

[0315] It can be seen that under low-speed driving conditions, the propulsion load remains within a low range, the output power of the fuel cell and lithium battery is low, and the ship's behavior at this time is characterized by traveling at a lower speed; under normal sailing conditions, the propulsion load and battery output power increase compared to the low-speed driving condition, and the average rate of change of the propulsion load is similar. At this time, the ship behaves as if it is sailing in open waters; under high-speed cruising conditions, the propulsion load can reach the rated output, and the average output power of the battery increases. At this time, the ship performs the cruising mission at a higher speed; the above three operating conditions have the same characteristics, including a low proportion of fluctuation data (up to 10%), low average and maximum rate of change of propulsion load, and fuel cells as the main power output. Combined with the ship's behavior, they are classified as general operating conditions;

[0316] Different from normal operating conditions, the average and maximum values ​​of propulsion load change rates in other operating conditions are higher, the propulsion load range is larger, and the proportion of fluctuation data is high. The power battery is the main power output. Among them, under operating conditions, the fluctuation data accounts for 60.7%, and the average propulsion load change rate increases by 61.2% compared with high-speed cruising conditions. At this time, the ship is carrying cargo or passengers. Under emergency or port conditions, the fluctuation data accounts for as high as 75.4%, and the average propulsion load change rate increases by 108% compared with high-speed cruising conditions. At this time, the ship is in extreme weather or entering or leaving the port.

[0317] Based on the above feature analysis, the five detailed operating conditions are divided into two categories, namely general operating conditions and high-fluctuation conditions. It can be seen that after the second subdivision, some large-fluctuation data points are distinguished from low-fluctuation data points, and the high-fluctuation conditions are successfully distinguished from the general operating conditions.

[0318] The basic working condition classification diagram of the all-electric ship under a period of continuous operation in this embodiment is as follows Figure 5 As shown in the figure, the classification diagram of specific working conditions of all-electric ships under a period of continuous working time is as follows Figure 6 As shown, Figure 5 and Figure 6 The figures include 2,000 data points of the same continuous working time of the all-electric ship. The lines of different colors represent the different working condition categories identified by the model, and the cross marks represent the connection points between different working conditions for a clearer observation of the distribution of working conditions.

[0319] from Figure 5 It can be seen from the figure that the initial identification feature is the parallel distribution formed by the propulsion power, which is consistent with the Figure 3Correspondingly, in the initial recognition stage of K-means, due to the isotropy of the method, the load change rate cannot be effectively paid attention to, which leads to the repeated appearance of multiple different categories in a continuous curve; Figure 6 It can be seen that the secondary subdivision has a good guarantee for the continuity of working conditions.

[0320] The simulation results are shown in Table 7. It should be noted that in Table 7, the multi-condition identification method of this embodiment is abbreviated as K-Bo-LightGBM:

[0321] Table 7 Comparison of simulation results

[0322]

[0323] It can be seen from Table 7 that the multi-condition identification method of this embodiment has the highest CH index and the lowest DB index. In particular, compared with the K-means method and LightGBM, the multi-condition identification method of this embodiment has an improvement of 14% and 27% in the CH index, and a reduction of 6% and 17% in the DB index, respectively. In terms of stability score, the multi-condition identification method of this embodiment has the highest stability score of 0.93.

[0324] The above simulation experiment results demonstrate that the multi-condition identification method of the present application has strong reliability and high identification stability when dealing with environmental interference and abnormal points.

[0325] The following is an embodiment of a multi-operating condition identification system for an all-electric ship provided in an embodiment of the present disclosure. The multi-operating condition identification system for an all-electric ship and the multi-operating condition identification method for an all-electric ship in the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the multi-operating condition identification system for an all-electric ship, reference can be made to the embodiment of the multi-operating condition identification method for an all-electric ship in the above-mentioned embodiments.

[0326] A mobile terminal implementing various embodiments of the present application will now be described with reference to the accompanying drawings. In the subsequent description, suffixes such as "module," "component," or "unit" used to represent components are used solely to facilitate description of the embodiments of the present application and do not inherently have specific meanings. Therefore, "module" and "component" may be used interchangeably.

[0327] like Figure 7 As shown in the figure, the multi-operating condition identification system of the all-electric ship includes:

[0328] Data acquisition module, used to obtain historical operating data of all-electric ships;

[0329] The data processing module is used to extract features and reduce noise on historical operation data to obtain a noise-reduced data group for each sample;

[0330] The preliminary clustering module is used to input the propulsion load and the first-order derivative of the propulsion load after noise reduction of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic working conditions, and adding a unique basic working condition label to each noise reduction data group;

[0331] The judgment and classification module is used to determine whether the absolute value of the first-order derivative of the propulsion load of each sample after noise reduction is higher than a preset threshold. The noise reduction data group corresponding to all samples with a positive result is aggregated into a high-fluctuation data set, and the propulsion load, first-order derivative of the propulsion load, and basic working condition label corresponding to all samples with a negative result are aggregated into a general data set.

[0332] The secondary segmentation module is used to uniformly extract the main features of the data in the high-volatility data set and the general data set using the kernel principal component analysis algorithm, and then input the pre-trained working condition segmentation model to output the specific working condition category.

[0333] The multi-operating condition identification system of the integrated energy system all-electric ship of this embodiment is used to implement a multi-operating condition identification method of the all-electric ship, and the steps include:

[0334] S1. Obtain historical operating data for all-electric vessels. This historical operating data consists of samples at various timestamps. Each sample includes the propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters at the corresponding timestamp.

[0335] S2. Perform feature extraction and noise reduction on the historical operating data to obtain a noise-reduced data set for each sample, including the noise-reduced propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters.

[0336] S3. Input the denoised propulsion load and first-order derivative of the propulsion load of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic operating conditions and adding a unique basic operating condition label to each denoised data group;

[0337] S4. Determine whether the absolute value of the first-order derivative of the propulsion load after noise reduction for each sample is higher than a preset threshold. Aggregate the noise reduction data set corresponding to all samples with a positive result into a high-fluctuation dataset. Aggregate the propulsion load, first-order derivative of the propulsion load, and basic operating condition labels corresponding to all samples with a negative result into a general dataset.

[0338] S5. Conduct secondary subdivisions, including:

[0339] The kernel principal component analysis algorithm is used to uniformly extract the main features of the data in the high-volatility dataset and the general dataset, and then the pre-trained working condition segmentation model is input to output the specific working condition category. The working condition segmentation model is a Bayesian optimized weighted iterative LightGBM model.

[0340] The present application also provides an electronic device for implementing various embodiments of the present application. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0341] Figure 8 A schematic diagram of the hardware structure of an electronic device for implementing various embodiments of the present application.

[0342] The electronic device includes, but is not limited to, components such as a processor and a memory. Those skilled in the art will appreciate that the electronic device structures described in the embodiments of the present application do not limit the electronic device, and the electronic device may include more or fewer components than shown, or may combine certain components or arrange the components differently.

[0343] In the embodiments of the present application, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or claimed herein.

[0344] In the embodiment of the present application, the processor can be implemented by using at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a processor, a controller, a microcontroller, a microprocessor, and an electronic unit designed to perform the functions described herein. In some cases, such an embodiment can be implemented in a controller. For software implementation, an embodiment such as a process or function can be implemented with a separate software module that allows the execution of at least one function or operation. The software code can be implemented by a software application (or program) written in any appropriate programming language, and the software code can be stored in a memory and executed by a controller.

[0345] In addition, the electronic device includes some functional modules not shown, which will not be described here.

[0346] Those skilled in the art will appreciate that various aspects of the electronic device provided herein may be implemented as a system, method, or program product. Therefore, various aspects of the present disclosure may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0347] This application also provides a storage medium storing a program product capable of implementing a multi-operating-condition identification method for an all-electric ship. In some possible implementations, various aspects of this disclosure may also be implemented in the form of a program product comprising program code. When the program product is executed on a terminal device, the program code is configured to cause the terminal device to execute the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of the disclosure.

[0348] The storage medium can be any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0349] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-operating condition identification method for an all-electric ship, characterized in that: include: S1. Obtain historical operating data for all-electric vessels. This historical operating data consists of samples at various timestamps. Each sample includes the propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters at the corresponding timestamp. S2. Perform feature extraction and noise reduction on the historical operating data to obtain a noise-reduced data set for each sample, including the noise-reduced propulsion load, the first-order derivative of the propulsion load, propulsion motor parameters, environmental parameters, and power battery output parameters. Feature extraction and noise reduction of historical operation data include: Decompose the original operating data into multiple eigenmodes through variational mode; Calculate the maximum information coefficient of each eigenmode and the original data, and retain the eigenmode whose maximum information coefficient exceeds the preset threshold as the main signal; Reconstruct the main signal and remove the interference noise to obtain the noise reduction data set of each sample; S3. Input the denoised propulsion load and first-order derivative of the propulsion load of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic operating conditions and adding a unique basic operating condition label to each denoised data group; S4. Determine whether the absolute value of the first-order derivative of the propulsion load after noise reduction for each sample is higher than a preset threshold. Aggregate the noise reduction data set corresponding to all samples with a positive result into a high-fluctuation dataset. Aggregate the propulsion load, first-order derivative of the propulsion load, and basic operating condition labels corresponding to all samples with a negative result into a general dataset. S5. Conduct secondary subdivisions, including: The kernel principal component analysis algorithm is used to uniformly extract the main features of the data in the high-volatility dataset and the general dataset, and then the pre-trained working condition segmentation model is input to output the specific working condition category. The working condition segmentation model is a Bayesian optimized weighted iterative LightGBM model.

2. The multi-operating condition identification method according to claim 1, characterized in that: The steps of variational mode decomposition include: The variational model is iteratively solved by the alternating direction multiplier method and Fourier isometric transform to obtain the decomposed mode function set and the corresponding center frequency; The steps to calculate the maximum information coefficient include: The eigenmodes of the decomposed propulsion load data and the original data are discretized into a two-dimensional grid space, and the normalized maximum mutual information value is calculated as the maximum information coefficient; The steps to remove interference noise include: The eigenmodes whose maximum information coefficient is lower than a preset threshold are removed as noise signals, and the signal is reconstructed by linearly superposing the retained eigenmodes.

3. The multi-operating condition identification method according to claim 1, characterized in that: In step S5, the operating condition subdivision model includes the following hierarchical structure: The input layer receives the main features extracted by kernel principal component analysis as input features; The feature processing layer includes a mutually exclusive feature bundling module and a gradient unilateral sampling module, which respectively perform sparse feature compression and high-information sample screening on the input features to obtain optimized features; The decision tree generation layer generates a binary tree structure according to the leaf growth strategy based on the hyperparameters provided by the Bayesian optimization framework layer. The information gain of each split node is calculated through the weighted iterative objective function. The Bayesian optimization framework layer dynamically adjusts the hyperparameter combination of the decision tree generation layer, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio; The output layer maps the leaf node weights through a weighted iterative classifier to generate specific working condition category labels.

4. The multi-operating condition identification method according to claim 3, characterized in that: The weighted iterative objective function is: Where, is the total number of samples; is the sample weight adjustment coefficient; is the cross entropy loss function; For the The output of a decision tree; For samples Vector representation of ; is the decision tree complexity regularization term; is a constant term; For samples confidence level; is the average distance between the feature point and other data of the same category; is the average distance between the feature point and the feature data points of other categories; for Type of working condition; and is a constant, set according to the number and size of the training set.

5. The multi-operating condition identification method according to claim 3, characterized in that: In step S5, the training steps of the working condition segmentation model include: S501. Obtain historical operating data for all-electric ships with known specific operating conditions. Perform the same operations as steps S1-S4 to obtain a training high-volatility dataset and a training general-purpose dataset. Kernel principal component analysis is then used to uniformly extract key features from both the high-volatility and general-purpose datasets. The datasets are then divided into training, validation, and test sets in a 7:2:1 ratio. Stratified sampling is used to ensure a balanced distribution of samples for each specific operating condition. S502. Initialize the hyperparameters of the working condition segmentation model using a Bayesian optimization parameter space search strategy, including the number of leaf nodes, maximum depth, learning rate, and feature sampling ratio; S503. Dynamically adjust hyperparameters using the Bayesian optimization framework: Calculate the weighted iterative objective function value on the validation set, generate the posterior probability distribution of the hyperparameters based on the function value changes, and iteratively update the hyperparameters of the working condition segmentation model. S504. Input the training set into the working condition segmentation model for training, including: The input features are compressed and filtered through the mutually exclusive feature bundling module and the gradient unilateral sampling module; In the decision tree generation layer, a binary tree structure is generated according to the leaf growth strategy, and the information gain of each split node is calculated based on the correlation between the first-order derivative of the propulsion load and the environmental parameters; Based on sample confidence Dynamically calculate weight coefficients , update the decision tree structure through the weighted iterative objective function and output the specific working condition category label; S505. Input the test set into the working condition segmentation model. Calculate the classification accuracy for each specific working condition category based on the output of each sample and the actual specific working condition category. If the classification accuracy does not meet the preset conditions, return to step S503 to readjust the hyperparameters and iterate training. If the preset conditions are met, output the final model to obtain the pre-trained working condition segmentation model.

6. The multi-operating condition identification method according to claim 1, characterized in that: Specific operating condition categories include low-speed driving conditions, normal navigation conditions, high-speed cruising conditions, operating conditions, and emergency or port conditions. Among them, low-speed driving conditions, normal navigation conditions, and high-speed cruising conditions belong to general operating conditions, while operating conditions, emergency or port conditions belong to high-fluctuation conditions.

7. A multi-operating condition identification system for an all-electric ship, characterized in that: A method for implementing the multi-operating condition identification method according to any one of claims 1 to 6, comprising: Data acquisition module, used to obtain historical operating data of all-electric ships; The data processing module is used to extract features and reduce noise on historical operation data to obtain a noise-reduced data group for each sample; The preliminary clustering module is used to input the propulsion load and the first-order derivative of the propulsion load after noise reduction of each sample into the K-means clustering algorithm for preliminary clustering, classifying them into multiple basic working conditions, and adding a unique basic working condition label to each noise reduction data group; The judgment and classification module is used to determine whether the absolute value of the first-order derivative of the propulsion load of each sample after noise reduction is higher than a preset threshold. The noise reduction data group corresponding to all samples with a positive result is aggregated into a high-fluctuation data set, and the propulsion load, first-order derivative of the propulsion load, and basic working condition label corresponding to all samples with a negative result are aggregated into a general data set. The secondary segmentation module is used to uniformly extract the main features of the data in the high-volatility data set and the general data set using the kernel principal component analysis algorithm, and then input the pre-trained working condition segmentation model to output the specific working condition category.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor is configured to implement the steps of the multi-operating condition identification method as claimed in any one of claims 1 to 6 when executing the computer program.

9. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-operating condition identification method according to any one of claims 1 to 6.

Images