A method for constructing a ship fuel consumption prediction model and application thereof

Abstract

The embodiment of the application provides a kind of ship oil consumption prediction model construction method and related equipment, belong to intelligent shipping technical field.The method includes: based on the voyage data set of different independent voyages and its corresponding physical element feature extraction result carries out model construction, to obtain the candidate double-layer gray-box oil consumption model of fusion physical and data double drive, and the candidate improved double-layer gray-box oil consumption model obtained after doing physical information supplement;Further, construct covariant deviation scene and utilize covariant deviation scene to the candidate model is screened, obtains several base models;Finally, based on physical element feature enhancement mechanism to base model is fused, to obtain target oil consumption prediction model;The embodiment method takes different independent voyages as modeling unit, based on the strategy of constructing covariant offset scene, candidate model screening and physical element feature enhancement fusion, constructs and obtains the ship oil consumption prediction model with the generalization performance meeting the demand.

Description

Technical Field

[0001] This application relates to the field of intelligent shipping technology, and in particular to a method for constructing a ship fuel consumption prediction model and its application. Background Technology

[0002] Operational energy efficiency management is crucial for maximizing the economic and environmental benefits of ocean-going vessels, and accurate fuel consumption models are the foundation for optimizing and controlling operational energy efficiency. Ship fuel consumption models can be used for route optimization, speed optimization, trim optimization, and comprehensive energy efficiency assessment; therefore, their predictive accuracy directly impacts the results of ship operational energy efficiency management.

[0003] For ocean-going vessels, fuel consumption is affected not only by traditional factors such as speed, draft, and wind and waves, but also by the thrust of the propeller. The thrust of the propeller is closely related to the relative wind direction, the angle of attack of the propeller, and the operating state of the propeller. There are also mutual influences between multiple propellers, making the energy transfer relationship and fuel consumption mechanism of ocean-going vessels more complex. In the face of this situation, existing fuel consumption models have insufficient generalization performance.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a method for constructing a ship fuel consumption prediction model and its application. This method uses different independent voyages as modeling units and, based on strategies such as constructing covariate offset scenarios, candidate model screening, and physical element feature enhancement fusion, constructs a ship fuel consumption prediction model with generalization performance that meets the requirements.

[0006] To achieve the above objectives, one aspect of this application proposes a method for constructing a ship fuel consumption prediction model, the method comprising: Obtain a sample voyage dataset; the sample voyage dataset includes sample voyage data from several different independent voyages; extract physical element features from the sample voyage dataset based on the ship's energy transfer relationship to obtain the sample physical element feature extraction results; the physical element features are used to characterize the coupling relationship between the wind vane boost, hull drag, wind and wave disturbance, and main engine load; Based on the sample voyage dataset, the sample physical element feature extraction results, the preset fuel consumption physical model, and the preset data-driven model, a model is constructed to obtain several candidate two-layer gray box fuel consumption models; the physical element feature extraction results serve as constraint inputs for the model construction process; physical information is supplemented for each candidate two-layer gray box fuel consumption model to construct a candidate improved two-layer gray box fuel consumption model corresponding to each candidate two-layer gray box fuel consumption model; A covariate shift scenario is formed based on the sample cruise dataset; the covariate shift scenario consists of a training dataset and a test dataset, and the data distribution of the training dataset and the test dataset are different; based on the covariate shift scenario, several candidate two-layer gray box fuel consumption models and several candidate improved two-layer gray box fuel consumption models are tested, and several base models are selected according to the test results; Based on the physical element feature enhancement mechanism and the physical element feature extraction results of the samples, several base models are fused, and a target fuel consumption prediction model is constructed based on the fusion results.

[0007] In some embodiments, the candidate dual-layer gray box fuel consumption model includes a serial dual-layer gray box fuel consumption model; the serial dual-layer gray box fuel consumption model is obtained in the following manner: The sample voyage dataset is input into the preset fuel consumption physical model for calculation, and the physical model output results are obtained. The physical output results, the sample physical element feature extraction results, and the sample voyage dataset are input into the preset data-driven model, allowing the model to learn the relationship between the input data and fuel consumption prediction, so as to obtain the serial double-layer gray box fuel consumption model.

[0008] In some embodiments, the candidate dual-layer gray box fuel consumption model includes a parallel dual-layer gray box fuel consumption model; the parallel dual-layer gray box fuel consumption model is obtained in the following manner: The sample cruise dataset is input into the preset fuel consumption physical model for calculation, and the physical model output result is obtained; the sample cruise dataset is input into the preset data-driven model for calculation, and the data-driven model output result is obtained. The physical model output, the data-driven model output, and the sample physical element feature extraction results are fused together. The fused result is used as the input data for the pre-model, which learns the relationship between the input data and fuel consumption prediction to obtain the parallel two-layer gray box fuel consumption model.

[0009] In some embodiments, the process of forming a covariate offset scenario based on the sample cruise dataset includes: Select sample cruise data from several independent cruises from the sample cruise dataset, use the sample cruise data from one of the independent cruises as the training dataset, and use the sample cruise data from the remaining one or more independent cruises as the test dataset, so as to form a covariate shift scenario where the distribution of training data differs from that of test data.

[0010] In some embodiments, the fusion of several base models based on the physical element feature enhancement mechanism, and the construction of a target fuel consumption prediction model based on the fusion result, includes: The sample voyage dataset is input into the preset fuel consumption physical model for calculation, and the physical model output results are obtained. The sample cruise dataset is input into several base models for calculation, and the base model output results are obtained. Based on the output of the physical model, a first feature extraction is performed to obtain a first feature extraction result; based on the output of the base model, a second feature extraction is performed to obtain a second feature extraction result; the purpose of the first feature extraction or the second feature extraction is to obtain physical information that is directly related to changes in fuel consumption; The first feature extraction result and the second feature extraction result are used as additional input data or model structure embedding for the preset data-driven model, so that the model learns the relationship between the input data and fuel consumption prediction in order to construct the target fuel consumption prediction model.

[0011] Furthermore, another aspect of this application proposes a method for predicting ship fuel consumption, which is applied to a target fuel consumption prediction model constructed by the above-described construction method; the method includes the following steps: Obtain the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset; The voyage dataset is input into several trained base models for calculation, and prediction results of several base models are obtained; a meta-feature set is constructed based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the voyage dataset. The meta-feature set is input into the target fuel consumption prediction model for calculation to obtain the fuel consumption prediction result for the target voyage.

[0012] In some embodiments, the target fuel consumption prediction model is trained in the following manner: Obtain the sample dataset; The sample dataset is input into several base models for calculation to obtain the fuel consumption prediction results of each base model; the sample dataset is input into a preset fuel consumption physical model for calculation to obtain the fuel consumption prediction results of the physical model. The target value dataset is obtained by calculating the fuel consumption prediction results of each base model and the fuel consumption prediction results of the physical model. The fuel consumption prediction pre-model is trained based on the sample dataset and the target value dataset. The fuel consumption prediction pre-model is adjusted according to the training results until the preset requirements are met. The trained target fuel consumption prediction model is obtained based on the adjustment results.

[0013] Furthermore, another aspect of this application proposes a ship fuel consumption prediction system, which is applied to the above-described ship fuel consumption prediction method; the system includes: The data acquisition module is used to acquire the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset; The data processing module is used to input the voyage dataset into several trained base models for calculation, and obtain the prediction results of several base models; and to construct a meta-feature set based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the voyage dataset. The fuel consumption prediction module is used to input the meta-feature set into the target fuel consumption prediction model for calculation, and obtain the fuel consumption prediction result for the target voyage.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described ship fuel consumption prediction method.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described ship fuel consumption prediction method.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a method for constructing a ship fuel consumption prediction model, as well as a corresponding ship fuel consumption prediction method, system, electronic device, and storage medium; the model construction method provided in the embodiments of this application obtains a sample voyage dataset containing sample voyage data from different independent voyages and its corresponding physical element feature extraction results; then, based on the obtained dataset, physical element feature extraction results, and fuel consumption physical model (white box model) and data-driven model (black box model), a model is constructed to obtain a candidate two-layer gray box fuel consumption model that integrates physical and data-driven approaches, and a candidate improved two-layer gray box fuel consumption model obtained after supplementing physical information; further, a covariate bias scenario is constructed and the candidate model is screened using the covariate bias scenario to obtain several base models; finally, based on the physical element feature enhancement mechanism, several base models are fused to construct the target model. The proposed fuel consumption prediction model incorporates physical element feature information into the model building process, providing a physical prior foundation and physical support for subsequent dual-drive modeling. Furthermore, it utilizes sample cruise data from different independent cruises to construct covariate bias scenarios. By designing the distribution differences between training and test data, the model testing process considers the impact of covariate bias, thereby selecting base models that maintain good generalization performance across cruises. This effectively improves the generalization performance of the target fuel consumption prediction model obtained by fusing base models and reduces model prediction accuracy errors when facing covariate bias, thus significantly improving model accuracy and generalization performance. Additionally, by enhancing the fusion of different dual-drive fuel consumption models through physical element features to construct the final target fuel consumption model, the model's adaptability to changes in wind field, sea state, loading status, and wind turbine operating status is enhanced, and the physical interpretability of model predictions is improved. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for constructing a ship fuel consumption prediction model provided in an embodiment of this application; Figure 2 This is an overall flowchart of the target fuel consumption prediction model construction process provided in the embodiments of this application; Figure 3 This is a flowchart of the process for constructing the candidate dual-layer gray box fuel consumption model TTGM provided in the embodiments of this application; Figure 4 This is a flowchart illustrating the construction process of the candidate improved dual-layer gray box fuel consumption model ITGM provided in the embodiments of this application; Figure 5 This is a logic block diagram of an experimental design and evaluation method for covariate shift provided in an embodiment of this application; Figure 6 This is a flowchart of the prediction and training process of the target fuel consumption prediction model provided in the embodiments of this application; Figure 7 This is a structural block diagram of a ship fuel consumption prediction system provided in an embodiment of this application; Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application.

[0019] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0020] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0023] 1) Physical model: An abstract and simplified representation of an objective physical system, outlining its core evolution mechanism through mathematical equations, logical frameworks or numerical simulations, revealing its inherent laws and achieving verifiable predictions while ignoring secondary disturbances.

[0024] 2) Data-driven model: Based on massive observation data, machine learning or statistical algorithms are used to automatically mine the implicit statistical mapping relationships between variables, thereby realizing a modeling paradigm for prediction and decision inference of unknown states.

[0025] 3) Covariate shift: This refers to the difference in the distribution of features (independent variables) between the data used during model training (training set) and the data encountered during actual model application (test set), but the relationship between features and labels (conditional distribution) remains unchanged.

[0026] Existing ship fuel consumption modeling methods mainly include physical models and data-driven models. Among them, data-driven models have strong nonlinear representation capabilities and can usually achieve high-precision prediction of main engine fuel consumption for a target voyage under complex sea conditions. However, their high prediction accuracy depends on the assumption that the training data and test data satisfy approximately independent and identically distributed (i.i.d.). In reality, the marine environment is highly variable. Different voyages differ in weather conditions, loading status, maneuvering methods, and other operating conditions, resulting in significant differences in the distribution of input features across voyage datasets and exhibiting covariate shift.

[0027] In the initial stages of data collection, due to practical limitations, training data typically comes from only a few voyages, making it difficult to comprehensively cover complex ocean conditions and variable ship operating states. This leads to data-driven ship fuel consumption prediction models being affected by covariate bias when predicting across voyage conditions, significantly reducing their generalization performance. Covariate bias is one of the important reasons for the failure of data-driven model predictions. Even if the model performs well on training data, its predictive performance will significantly decrease once the distribution of test data is inconsistent with that of training data.

[0028] For ocean-going vessels, fuel consumption is affected not only by traditional factors such as speed, draft, and wind and waves, but also by the thrust generated by wind turbines. Wind turbine thrust is closely related to relative wind direction, wind turbine angle of attack, and wind turbine operating status. Furthermore, multiple wind turbines interact with each other, making the energy transfer relationships and fuel consumption mechanisms of ocean-going vessels more complex. While existing physical models possess good physical interpretability and a certain degree of generalization ability, they often lack sufficient accuracy. Data-driven models have higher accuracy but weaker generalization ability under covariate shift scenarios. Dual-driven models, combining physics and data, are considered an important direction for balancing accuracy and generalization ability; however, current research still falls short in improving their generalization performance under covariate shift scenarios.

[0029] In view of this, this application provides a method for constructing a ship fuel consumption prediction model, as well as a corresponding ship fuel consumption prediction method, system, electronic device, and storage medium. The model construction method provided in this application involves acquiring a sample voyage dataset containing sample voyage data from different independent voyages and its corresponding physical element feature extraction results. Then, based on the acquired dataset, the physical element feature extraction results, and the fuel consumption physical model (white-box model) and the data-driven model (black-box model), a model is constructed to obtain a candidate two-layer gray-box fuel consumption model that integrates physical and data-driven approaches, and a candidate improved two-layer gray-box fuel consumption model obtained after supplementing with physical information. Further, a covariate bias scenario is constructed and used to screen candidate models, thereby obtaining several base models. Finally, based on the physical element feature enhancement mechanism, the several base models are fused to construct the target fuel consumption prediction model. On the one hand, the method of this application introduces physical element feature information into the model building process, providing a physical prior foundation and physical support for subsequent dual-drive modeling. On the other hand, the method of this application can construct covariate bias scenarios using sample cruise data from different independent cruises. By designing the distribution differences between training and test data, the model testing process takes into account the influence of covariate bias, thereby selecting base models that can still guarantee good generalization performance across cruises. This effectively improves the generalization performance of the target fuel consumption prediction model obtained by fusing base models and reduces the model prediction accuracy error when facing covariate bias, effectively improving the model's accuracy and generalization performance. In addition, by enhancing the fusion of different dual-drive fuel consumption models through physical element features to construct the final target fuel consumption model, the model's adaptability to changes in wind field, sea state, loading status, and wind turbine operating status can be enhanced, and the physical interpretability of the model prediction can be improved.

[0030] Figure 1 This is an optional flowchart of a method for constructing a ship fuel consumption prediction model provided in an embodiment of this application. Figure 2 This is an overall flowchart of the target fuel consumption prediction model construction process provided in the embodiments of this application; Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0031] Step S101: Obtain the sample voyage dataset; the sample voyage dataset includes sample voyage data from several different independent voyages; extract physical element features from the sample voyage dataset based on the ship's energy transfer relationship to obtain the sample physical element feature extraction results; the physical element features are used to characterize the coupling relationship between the wind vane boost, hull resistance, wind and wave disturbance, and main engine load.

[0032] The method obtains sample voyage data containing different independent voyages and the corresponding physical element feature extraction results as the data basis for subsequent calculations. In this regard, the physical element features obtained by the method of this application are not directly used as a separate theoretical fuel consumption model for correction, but are used as prediction inputs in the target voyage adaptation prediction process. In addition, physical element feature extraction can be performed through a physical model, by extracting physical prior knowledge and ship parameters from the voyage data as the data basis for subsequent calculations.

[0033] Step S102: Based on the sample voyage dataset, the sample physical element feature extraction results, the preset fuel consumption physical model, and the preset data-driven model, a model is constructed to obtain several candidate two-layer gray box fuel consumption models; the physical element feature extraction results are used as constraint inputs for the model construction process; physical information is supplemented for each candidate two-layer gray box fuel consumption model to construct a candidate improved two-layer gray box fuel consumption model corresponding to each candidate two-layer gray box fuel consumption model.

[0034] The physical element feature extraction results are used as the constraint input for model construction. Then, the model is constructed based on the acquired data, traditional fuel consumption physics models, and traditional data-driven models. The input of the traditional fuel consumption physics model is the operating condition parameter with clear physical dimensions, and the output is the theoretical fuel consumption rate, which belongs to the white box model. The traditional data-driven model belongs to the black box model. It does not require physical formulas, but directly learns the numerical mapping relationship between data and predicts the fuel consumption estimate by inputting massive historical sensor data. The method of this embodiment is based on these two traditional models as references to construct the model, thereby obtaining the dual-drive dual-layer gray box fuel consumption model (TTGM). Among them, the physical element feature extraction results are used as the constraint input for model construction, and as the physical prior support in model construction. In addition, physical information is supplemented for each candidate dual-layer gray box fuel consumption model, that is, more physical prior knowledge is further incorporated into the model, thereby constructing the corresponding candidate improved dual-layer gray box fuel consumption model (ITGM).

[0035] Furthermore, the physical information may include one or more physical characteristics related to wing boost, drag variation, propulsion power, energy transfer efficiency, and main engine fuel consumption. The candidate improved dual-layer gray box fuel consumption model (ITGM) not only retains the ability of data-driven models to learn complex nonlinear relationships, but also enhances the physical interpretability and cross-flight generalization ability of the model.

[0036] Step S103: A covariate shift scenario is formed based on the sample cruise dataset. The covariate shift scenario consists of a training dataset and a test dataset, and the data distribution of the training dataset and the test dataset are different. Based on the covariate shift scenario, several candidate two-layer gray box fuel consumption models and several candidate improved two-layer gray box fuel consumption models are tested, and several base models are selected based on the test results.

[0037] By constructing a covariate bias scenario, i.e. a scenario where the data distribution of the training dataset and the test dataset differs, the impact of covariate bias under cross-voyage conditions is simulated. Then, the impact of covariate bias is considered in the model testing, and the model that meets the requirements is selected as the base model.

[0038] Step S104: Based on the physical element feature enhancement mechanism, several base models are fused, and a target fuel consumption prediction model is constructed based on the fusion result.

[0039] Finally, the base models are fused based on the physical element feature enhancement mechanism. Due to the previous model testing and screening, the target fuel consumption prediction model obtained by fusing the base models can still maintain good generalization performance across voyages.

[0040] Therefore, the model training method in this application can be carried out through various machine learning algorithms and hyperparameter optimization, and there is no limitation on it.

[0041] Optionally, the cruise dataset in step S101 needs to be preprocessed to provide a unified and reliable data foundation. Data preprocessing methods include time synchronization, outlier removal, missing value handling, smoothing, unit unification, and feature standardization.

[0042] In some embodiments, combined with Figure 3 , Figure 3 This is a flowchart illustrating the process of constructing a candidate dual-layer gray box fuel consumption model (TTGM) according to an embodiment of this application. The candidate dual-layer gray box fuel consumption model (TTGM) includes a serial dual-layer gray box fuel consumption model (serial TTGM). The serial dual-layer gray box fuel consumption model is obtained through the following methods: Step S201: Input the sample voyage dataset into the preset fuel consumption physical model for calculation and obtain the physical model output results.

[0043] Obtain the output of the physical model and use it as a constraint enhancement for the candidate prediction model.

[0044] Step S202: Input the physical output results, the sample physical element feature extraction results, and the sample voyage dataset into the preset data-driven model, so that the model learns the relationship between the input data and fuel consumption prediction, in order to obtain the serial double-layer gray box fuel consumption model.

[0045] Among them, it can directly input the output of the physical model into the data-driven model for calculation, or it can calculate the difference between the target data of the original dataset and the output data of the physical model to obtain new target data, and then input the newly obtained data and the original data together into the data-driven model for calculation.

[0046] In a serial manner, the physical prediction values ​​or physical representation values ​​output by the fuel consumption physical model are used as new inputs to the data-driven model. These, along with the original input features, are input into the data-driven model, allowing the model to learn the numerical relationships related to fuel consumption prediction, thereby constructing a serial two-layer gray box fuel consumption model.

[0047] In some embodiments, the candidate two-layer gray box fuel consumption model (TTGM) includes a parallel two-layer gray box fuel consumption model (parallel TTGM); the parallel two-layer gray box fuel consumption model is obtained in the following manner: Step S301: Input the sample voyage dataset into the preset fuel consumption physical model for calculation and obtain the physical model output result; input the sample voyage dataset into the preset data-driven model for calculation and obtain the data-driven model output result.

[0048] Fuel consumption is predicted using both a physical fuel consumption model and a data-driven model. The two prediction results are then fused together and used as the data input for subsequent models.

[0049] Step S302: The physical model output, the data-driven model output, and the sample physical element feature extraction results are fused together. The fused result is used as the input data of the pre-model, and the pre-model learns the relationship between the input data and fuel consumption prediction to obtain a parallel two-layer gray box fuel consumption model.

[0050] The prediction results of the two models are fused, and the fused result is used as the model input to construct a parallel dual-drive fuel consumption model.

[0051] For the serial two-layer gray box fuel consumption model (serial TTGM) and the parallel two-layer gray box fuel consumption model (parallel TTGM), by introducing the mechanistic information of the physical model into the data-driven modeling process, the model can have a certain physical constraint capability while fitting nonlinear laws, thereby alleviating the problem of generalization degradation of the pure data-driven model under the condition of distribution variation.

[0052] In some embodiments, such as Figure 4 As shown, Figure 4 This is a flowchart illustrating the construction process of the candidate improved dual-layer gray box fuel consumption model (ITGM) provided in this application embodiment. By introducing more physical information into the candidate dual-layer gray box fuel consumption model TTGM, the candidate improved dual-layer gray box fuel consumption model is obtained. Specifically, the physical information includes wind wing drag, friction drag, appendage drag, air drag, residual drag, and wave drag, etc., extracted from the flight dataset.

[0053] In some embodiments, a covariate offset scenario is formed based on a sample cruise dataset, including: Select sample cruise data from several independent cruises from the sample cruise dataset. Use the sample cruise data from one of the independent cruises as the training dataset and the sample cruise data from the remaining one or more independent cruises as the test dataset to form a covariate shift scenario where the distribution of the training data differs from that of the test data.

[0054] The method in this embodiment no longer adopts the conventional approach of randomly dividing the training set and test set within the same voyage. Instead, it constructs the training set and test set using different independent voyages to truly reflect the covariate shift caused by differences in environment and operational strategies between different voyages.

[0055] In some embodiments, a target fuel consumption prediction model is constructed based on a physical element feature enhancement mechanism and several base models, and the fusion results are used to construct the model. Step S401: Input the sample voyage dataset into the preset fuel consumption physical model for calculation and obtain the physical model output results; input the sample voyage dataset into several base models for calculation and obtain the base model output results.

[0056] Obtain the model outputs of several selected base models and physical models as the data basis for subsequent analysis.

[0057] Step S402: Based on the output of the physical model, perform first feature extraction to obtain first feature extraction result; based on the output of the base model, perform second feature extraction to obtain second feature extraction result; the purpose of first feature extraction or second feature extraction is to obtain physical information that is directly related to changes in fuel consumption.

[0058] Based on the outputs of several base models and the physical model, features are extracted to obtain physical information directly related to changes in fuel consumption, which serves as the learning basis for subsequent model construction.

[0059] Step S403: Use the first feature extraction result and the second feature extraction result as additional input data or model structure embedding for the preset data-driven model, so that the model learns the relationship between the input data and fuel consumption prediction, in order to construct the target fuel consumption prediction model.

[0060] Using the aforementioned feature extraction results as model input, the model learns the relationship between the input data and fuel consumption prediction, thereby constructing the target fuel consumption prediction model.

[0061] In some embodiments, such as Figure 5 As shown, Figure 5 This is a logic block diagram of an experimental design and evaluation method for covariate shifting provided in an embodiment of this application; to verify the generalization ability of each model in cross-flight scenarios, the model generalization ability can be evaluated in the following ways: Step S501: Select at least three cruise datasets from multiple independent cruises; Step S502: Use one of the voyages as the training set and the remaining one or more voyages as the test set; Step S503: A covariate shift scenario is formed where the distribution of training data and test data are significantly inconsistent. Step S504: Test the predictive performance of the model in this scenario. Step S505: Use one or more of the following metrics to quantitatively evaluate the generalization performance of the model: mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE).

[0062] In one embodiment, the ship prediction model of the present invention can achieve a MAPE of 3.95% in the covariate offset scenario, which is 66.75%, 10.84%, 54.86%, and 61.39% lower than the best single model, TGM, physical model, and traditional data-driven model, respectively. This shows that introducing physical element features and screening base models in cross-voyage scenarios can effectively improve the prediction reliability of the model when facing the target voyage distribution offset.

[0063] In summary, the method for constructing the ship fuel consumption prediction model provided in this embodiment has at least the following beneficial effects: (1) The method in this embodiment constructs training voyages and target voyages using independent voyages as the basic modeling unit, which avoids the problem of data distribution being too similar due to random division of training and test sets within the same voyage. It can more realistically reflect the cross-voyage covariate shift scenario faced by ocean-going vessels in actual operation.

[0064] (2) The method in this embodiment constructs a double-layer gray box ship energy consumption model that improves the cross-voyage extrapolation performance by integrating physical element feature enhancement mechanism. The input of the model is no longer composed of prediction results of several base models, but further integrates physical element features generated by physical models, thereby enhancing the model's adaptability to changes in wind field, sea state, loading status and wind turbine operating status.

[0065] Another aspect of this application proposes a method for predicting ship fuel consumption, which is applied to the target fuel consumption prediction model constructed by the above-described construction method; such as Figure 6 As shown, Figure 6 The prediction and training processes of the above-mentioned target fuel consumption prediction model are demonstrated; the prediction method includes the following steps: Step S601: Obtain the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset.

[0066] Step S602: Input the voyage dataset into several trained base models for calculation to obtain the prediction results of several base models; construct a meta-feature set based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the voyage dataset.

[0067] Step S603: Input the meta-feature set into the target fuel consumption prediction model for calculation to obtain the fuel consumption prediction result for the target voyage.

[0068] In response, the input to the target fuel consumption prediction model is no longer solely composed of the prediction results from the base model, but also includes the results of physical element feature extraction; thus enabling the model to utilize physical mechanisms.

[0069] In some embodiments, such as Figure 6 As shown, the target fuel consumption prediction model is trained in the following way: Step S701: Obtain the sample dataset; Step S702: Input the sample dataset into several base models for calculation to obtain the fuel consumption prediction results of each base model; input the sample dataset into a preset fuel consumption physical model for calculation to obtain the fuel consumption prediction results of the physical model. Step S703: Calculate the target value dataset based on the fuel consumption prediction results of each base model and the fuel consumption prediction results of the physical model. Step S704: Train the fuel consumption prediction pre-model based on the sample dataset and the target value dataset, adjust the fuel consumption prediction pre-model according to the training results until it meets the preset requirements, and obtain the trained target fuel consumption prediction model according to the adjustment results.

[0070] In this regard, the target fuel consumption prediction model can be trained using various algorithms or hyperparameter optimization methods, and there are no restrictions on which one to use.

[0071] In some embodiments, the selected base models may include a serial improved dual-layer gray box fuel consumption model, a parallel improved dual-layer gray box fuel consumption model, and a parallel dual-layer gray box fuel consumption model.

[0072] In some embodiments, the training process for the target ship fuel consumption prediction model may include: The first phase comprises four model framework structures: a serial traditional double-layer gray box fuel consumption model, a serial improved double-layer gray box fuel consumption model, a parallel traditional double-layer gray box fuel consumption model, and a parallel improved double-layer gray box fuel consumption model. ITGM does not directly use the initial input data as input to the data-driven model; instead, the physical model generates a set of physical meta-features as input data. These physical meta-features include the propulsion provided by the wind turbine, the drag generated by hull surface friction, residual drag, air drag, appendage drag, and drag amplification under wave action. In the second stage, a physical model is added and a parallel structure is adopted. The obtained meta-feature set and the target value are input into the meta-model for training. Please see Figure 7 This application also provides a ship fuel consumption prediction system that can implement the above-mentioned ship fuel consumption prediction method. The system includes: The data acquisition module is used to acquire the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset; The data processing module is used to input the cruise dataset into several trained base models for calculation, and obtain the prediction results of several base models; based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the cruise dataset, a meta-feature set is constructed. The fuel consumption prediction module is used to input the meta-feature set into the target fuel consumption prediction model for calculation, and obtain the fuel consumption prediction result for the target voyage.

[0073] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0074] Please see Figure 8 This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0075] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0076] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0077] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0078] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0079] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0080] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0081] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0082] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0083] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0084] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0085] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0087] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0088] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for constructing a ship fuel consumption prediction model, characterized in that, The method includes the following steps: Obtain a sample voyage dataset; the sample voyage dataset includes sample voyage data from several different independent voyages; extract physical element features from the sample voyage dataset based on the ship's energy transfer relationship to obtain the sample physical element feature extraction results; the physical element features are used to characterize the coupling relationship between the wind vane boost, hull drag, wind and wave disturbance, and main engine load; Based on the sample voyage dataset, the sample physical element feature extraction results, the preset fuel consumption physical model, and the preset data-driven model, a model is constructed to obtain several candidate two-layer gray box fuel consumption models; the physical element feature extraction results serve as constraint inputs for the model construction process; physical information is supplemented for each candidate two-layer gray box fuel consumption model to construct a candidate improved two-layer gray box fuel consumption model corresponding to each candidate two-layer gray box fuel consumption model; A covariate shift scenario is formed based on the sample cruise dataset; the covariate shift scenario consists of a training dataset and a test dataset, and the data distribution of the training dataset and the test dataset are different; based on the covariate shift scenario, several candidate two-layer gray box fuel consumption models and several candidate improved two-layer gray box fuel consumption models are tested, and several base models are selected according to the test results; Based on the physical element feature enhancement mechanism, several base models are fused, and a target fuel consumption prediction model is constructed based on the fusion result.

2. The construction method according to claim 1, characterized in that, The candidate dual-layer gray box fuel consumption model includes a serial dual-layer gray box fuel consumption model; the serial dual-layer gray box fuel consumption model is obtained in the following way: The sample voyage dataset is input into the preset fuel consumption physical model for calculation, and the physical model output results are obtained. The physical output results, the sample physical element feature extraction results, and the sample voyage dataset are input into the preset data-driven model, allowing the model to learn the relationship between the input data and fuel consumption prediction, so as to obtain the serial double-layer gray box fuel consumption model.

3. The construction method according to claim 1, characterized in that, The candidate dual-layer gray box fuel consumption model includes a parallel dual-layer gray box fuel consumption model; the parallel dual-layer gray box fuel consumption model is obtained in the following way: The sample voyage dataset is input into the preset fuel consumption physical model for calculation, and the physical model output results are obtained. The sample cruise dataset is input into the preset data-driven model for calculation, and the output result of the data-driven model is obtained. The physical model output, the data-driven model output, and the sample physical element feature extraction results are fused together. The fused result is used as the input data for the pre-model, which learns the relationship between the input data and fuel consumption prediction to obtain the parallel two-layer gray box fuel consumption model.

4. The construction method according to claim 1, characterized in that, The scenario for forming covariate offsets based on the sample cruise dataset includes: Select sample cruise data from several independent cruises from the sample cruise dataset, use the sample cruise data from one of the independent cruises as the training dataset, and use the sample cruise data from the remaining one or more independent cruises as the test dataset, so as to form a covariate shift scenario where the distribution of training data differs from that of test data.

5. The construction method according to claim 1, characterized in that, The method of fusing several base models based on physical element feature enhancement, and constructing a target fuel consumption prediction model based on the fusion result, includes: The sample voyage dataset is input into the preset fuel consumption physical model for calculation, and the physical model output results are obtained. The sample cruise dataset is input into several base models for calculation, and the base model output results are obtained. Based on the output of the physical model, a first feature extraction is performed to obtain a first feature extraction result; based on the output of the base model, a second feature extraction is performed to obtain a second feature extraction result; the purpose of the first feature extraction or the second feature extraction is to obtain physical information that is directly related to changes in fuel consumption; The first feature extraction result and the second feature extraction result are used as additional input data or model structure embedding for the preset data-driven model, so that the model learns the relationship between the input data and fuel consumption prediction in order to construct the target fuel consumption prediction model.

6. A method for predicting ship fuel consumption, characterized in that, The method is applied to a target fuel consumption prediction model constructed using the construction method described in any one of claims 1 to 5; the method includes the following steps: Obtain the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset; The voyage dataset is input into several trained base models for calculation, and prediction results of several base models are obtained; a meta-feature set is constructed based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the voyage dataset. The meta-feature set is input into the target fuel consumption prediction model for calculation to obtain the fuel consumption prediction result for the target voyage.

7. The ship fuel consumption prediction method according to claim 6, characterized in that, The target fuel consumption prediction model is trained in the following way: Obtain the sample dataset; The sample dataset is input into several base models for calculation to obtain the fuel consumption prediction results of each base model; the sample dataset is input into a preset fuel consumption physical model for calculation to obtain the fuel consumption prediction results of the physical model. The target value dataset is obtained by calculating the fuel consumption prediction results of each base model and the fuel consumption prediction results of the physical model. The fuel consumption prediction pre-model is trained based on the sample dataset and the target value dataset. The fuel consumption prediction pre-model is adjusted according to the training results until the preset requirements are met. The trained target fuel consumption prediction model is obtained based on the adjustment results.

8. A ship fuel consumption prediction system, characterized in that, The system is applied to the ship fuel consumption prediction method as described in any one of claims 6 to 7; the system comprises: The data acquisition module is used to acquire the voyage dataset of the target voyage and the physical meta-feature extraction results corresponding to the voyage dataset; The data processing module is used to input the voyage dataset into several trained base models for calculation, and obtain the prediction results of several base models; and to construct a meta-feature set based on the prediction results of several base models and the physical meta-feature extraction results corresponding to the voyage dataset. The fuel consumption prediction module is used to input the meta-feature set into the target fuel consumption prediction model for calculation, and obtain the fuel consumption prediction result for the target voyage.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the ship fuel consumption prediction method according to any one of claims 5 to 6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the ship fuel consumption prediction method according to any one of claims 5 to 6.

Images