Electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation

Abstract

The invention discloses an electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation, and belongs to the technical field of intelligent shipping and ship energy management. The method aims at solving the problems that in an existing electric ship energy consumption prediction technology, a model is single, prediction precision is poor, and generalization ability is insufficient. In order to achieve the target, the method comprises the following steps: firstly, collecting historical operation data of the electric ship, and preprocessing; then, respectively constructing a physical model and a data driving model to describe a physical mechanism and a statistical rule of energy consumption; thirdly, adopting three fusion methods of stacking, mixing and voting to organically combine a physical model and a data-driven model, and improving prediction precision and generalization ability; and finally, evaluating the fusion model from multiple dimensions of prediction precision, calculation efficiency, model generalization ability and the like, and optimizing and deploying the model according to an evaluation result. According to the invention, high-precision energy consumption prediction, assistance energy scheduling optimization, battery management strategy formulation and carbon emission monitoring can be provided for a ship energy efficiency management system, and energy conservation and carbon reduction of ships are promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent shipping and ship energy management, and in particular to a method for predicting energy consumption of electric ships based on multi-model fusion and multi-dimensional evaluation. Background Art

[0002] As a key area of energy consumption and greenhouse gas emissions, the shipping industry is facing unprecedented pressure for green transformation. Emissions of carbon dioxide, nitrogen oxides, and particulate matter generated by traditional fuel-powered ships during operation have become a key factor restricting the sustainable development of the shipping industry. Against this backdrop, electric-powered ships, with their significant advantages such as zero tailpipe emissions, high energy efficiency, and low noise, are gradually becoming the mainstream development direction for ships operating in ports, short-distance passenger ferries, and inland cargo ships. However, the operating characteristics of electric ships are more complex than those of traditional fuel-powered ships. Their energy consumption level is not only limited by the hull structure and propulsion system parameters, but is also significantly affected by multiple factors such as navigation status, hydrological and meteorological conditions, battery system status, and load conditions.

[0003] Currently, electric ship energy consumption prediction technology is mainly divided into two categories: simulation modeling methods based on physical mechanisms and statistical prediction methods based on data-driven.

[0004] 1) Physics-based simulation modeling: This approach simulates the energy consumption characteristics of a ship under different operating conditions by establishing dynamic, propulsion, and battery system models during navigation. For example, computational fluid dynamics (CFD) technology is used to simulate hull resistance and energy consumption is calculated based on propulsion system efficiency curves. However, this method relies heavily on ship design parameters and operating environment assumptions, making it difficult to accurately reflect actual energy consumption levels under complex navigation conditions.

[0005] 2) Data-driven statistical prediction methods: This approach collects historical ship operating data and applies statistical models such as regression analysis and time series forecasting to establish a mapping between energy consumption and influencing factors. For example, a multivariate linear regression model is used to analyze the correlation between speed, load, and energy consumption. While this method offers high prediction accuracy when sufficient data is available, it has stringent requirements for data quality and lacks in-depth analysis of physical mechanisms, limiting the model's generalization capabilities.

[0006] For example, CN118982098A discloses a method and system for predicting energy consumption for electric bulk carriers based on multi-source data. This method collects multi-source data from the operation of electric bulk carriers, including hull status data, hydrological environment data, and weather environment information, and constructs a data network model to predict energy consumption. However, while this method takes into account the impact of multi-source data, it is still insufficient in terms of model fusion and performance evaluation, making it difficult to adapt to complex and changing navigation conditions. Furthermore, it lacks quantitative comparison and optimization of the effects of different models.

[0007] In addition, CN117290673A proposes a high-precision ship energy consumption prediction system based on multi-model fusion. This system improves prediction accuracy by constructing different types of ship energy consumption prediction models and using a stacking model fusion method for model fusion. However, this method focuses primarily on predicting the overall energy consumption of a ship and fails to adequately consider factors unique to electric ships, such as the battery system status and energy recovery efficiency. Furthermore, the model's generalization capabilities remain to be improved when dealing with the complex and changing operating conditions of electric ships.

[0008] Furthermore, CN119903346A discloses a method and device for predicting battery energy consumption for electric ships. This method achieves prediction of battery energy consumption for electric ships by constructing an initial prediction model based on an echo state network and optimizing the model using a particle swarm algorithm. However, this method primarily focuses on predicting battery energy consumption and has little coverage of the overall energy consumption of the ship. Furthermore, when processing multi-source heterogeneous data, there is still room for improvement in data preprocessing and feature engineering to further enhance prediction accuracy and model generalization.

[0009] Although the above methods have made some progress in the field of electric ship energy consumption prediction, the following major problems still exist: 1) Model Simplification: Existing technologies often employ a single modeling strategy, failing to fully leverage the complementary strengths of physical modeling and data-driven approaches. For example, while physical models can reveal energy consumption mechanisms, they struggle to address complex operating conditions; while data models can capture statistical patterns, they lack interpretability.

[0010] 2) Insufficient prediction accuracy: Since the energy consumption of electric ships is affected by multiple factors such as hull structure, propulsion system, navigation status, hydrological and meteorological conditions, and battery system status, existing methods are difficult to fully characterize the energy consumption characteristics, resulting in large prediction errors.

[0011] 3) Weak generalization ability: Existing models are mostly trained for specific ships or specific navigation scenarios. When the application conditions change (such as route adjustments or battery aging), the model performance drops significantly, making it difficult to meet actual engineering needs.

[0012] 4) Lack of a unified evaluation system: Existing research focuses on model development and lacks a systematic evaluation mechanism for energy consumption prediction performance, resulting in a lack of quantitative basis for model optimization and deployment.

[0013] Therefore, there is an urgent need for an energy consumption prediction method that integrates multiple modeling methods, combines multi-source environmental and equipment information input, and has higher prediction accuracy and stability. This method can meet the energy consumption modeling needs of electric ships under complex operating conditions and multi-segment conditions, and provide basic support for intelligent shipping systems. Based on this need, this paper proposes an electric ship energy consumption prediction method based on a multi-model fusion method and multi-dimensional performance evaluation, aiming to solve the problems existing in the existing technology and improve the accuracy and reliability of electric ship energy consumption prediction. Summary of the Invention

[0014] The technical problem to be solved by the present invention is to provide an electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation, so as to solve the problems of insufficient accuracy, weak generalization ability and lack of a unified evaluation system in the field of intelligent shipping and ship energy management technology; by providing an electric ship energy consumption prediction method that integrates physical modeling, data-driven and machine learning technologies, to overcome the specific limitations of the existing technology such as single model, limited prediction accuracy and insufficient generalization ability.

[0015] In order to achieve the above objectives, the present invention adopts the following technical solutions: Specifically, it includes four main steps: data collection, data preprocessing, model construction and energy consumption visualization output.

[0016] 1. Data collection: Collect multi-source heterogeneous data related to electric ship operation, including basic ship parameter data, navigation condition data, environmental data, battery system data, route and operation data, and auxiliary system data.

[0017] 2. Data preprocessing: Preprocess the collected data, including data cleaning (filling missing values, correcting outliers, and suppressing high-frequency noise), feature selection and reconstruction (calculating the correlation coefficient and mutual information between features and target variables, and using dimensionality reduction methods such as principal component analysis (PCA) and Lasso regression), timestamp alignment, and frequency synchronization (interpolation and resampling operations, or downsampling and synchronous compression operations, for data with different time precisions and sampling frequencies).

[0018] 3. Model Construction: Multiple single prediction models are constructed, and the Pearson correlation coefficient is used for model selection. Three fusion methods, stacking, blending, and voting, are applied to achieve model integration. The stacking method combines the prediction results of multiple different base models and uses a meta-model to further optimize the predictions. The blending method splits the training dataset into a training set and a validation set. The base models are trained on the training set and predicted on the validation set. A new training set is constructed by combining the predictions of all base models on the validation set, and the meta-model is trained. The voting method combines the prediction results of multiple base models and obtains the final prediction value through weighted averaging.

[0019] 4. Energy consumption visualization output: Generate high-precision energy consumption prediction curves, construct three-dimensional energy consumption prediction surface maps, and build dynamic decision-making panels; specifically, generate line graphs showing energy consumption changes over time, overlay comparison graphs of model prediction values and actual historical energy consumption data, multi-dimensional energy consumption distribution graphs or heat maps based on key variables, and energy consumption-range matching graphs.

[0020] 5. Multi-dimensional performance evaluation: Optimize the energy consumption prediction model in terms of prediction accuracy, computational efficiency, and model generalization capability to improve its practicality and reliability.

[0021] The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation provided by the present invention has the following beneficial effects: 1. This invention solves the problem of insufficient prediction accuracy of electric ship energy consumption. By integrating multi-source heterogeneous data and adopting a complex multi-model fusion strategy, it significantly improves the prediction accuracy and robustness.

[0022] 2. The present invention integrates multiple prediction models such as physical modeling, data-driven and machine learning, giving full play to the complementary advantages of various models, effectively capturing the nonlinear energy consumption characteristics of electric ships under different navigation conditions and complex environments, and improving the prediction accuracy of the model in practical applications; experimental results show that the prediction accuracy of the method of the present invention is significantly improved compared with the existing methods when actual electric ship operation data is used to test the method.

[0023] 3. The present invention overcomes the limitation of weak generalization ability in the existing technology. By optimizing the data processing process and model fusion strategy, the generalization ability of the model is enhanced, enabling it to adapt to complex and changing operating environments.

[0024] 4. The present invention adopts multi-dimensional feature engineering and model training strategies, combined with multi-source heterogeneous input data, and can be applied to energy consumption prediction tasks under different ship types, routes and working conditions, ensuring that the model has good adaptability in multiple scenarios.

[0025] 5. This invention proposes for the first time a comprehensive integration method for multi-source heterogeneous data for electric ship energy consumption prediction, and constructs a multi-dimensional visualization interface, providing ship operators with an intuitive and comprehensive energy efficiency analysis tool, which has not been reported in the prior art.

[0026] 6. The present invention constructs a three-dimensional energy consumption prediction surface, a dynamic time series diagram, and an interactive decision panel to intuitively display energy consumption trends, sensitive factors, and change trajectories, assisting ship operators in energy efficiency analysis and navigation decision-making, and enhancing the practicality of the prediction results.

[0027] 7. The present invention solves the problem of the lack of a unified energy consumption prediction and evaluation system in the existing technology. By comprehensively evaluating the model effect from multiple dimensions such as prediction error, computational efficiency, model stability, and interpretability, it provides a quantitative basis for model optimization and deployment, improves the scientific nature of model screening and optimization, and ensures the reliability and effectiveness of the model in practical applications.

[0028] 8. The present invention integrates the results of a single model by stacking integration, weighted voting and hybrid fusion, and combines Pearson correlation analysis to improve the fusion accuracy and improve the overall model robustness while ensuring computational efficiency.

[0029] 9. The present invention provides high-precision prediction input for the ship energy efficiency management system, supports energy scheduling optimization, battery management strategy formulation and carbon emission monitoring, promotes the realization of ship energy conservation and carbon reduction goals, and contributes to the development of green shipping and smart maritime.

[0030] 10. The present invention not only improves the accuracy and robustness of electric ship energy consumption prediction, but also provides powerful decision-making support for ship operators through a multi-dimensional visualization interface, and has significant practicality and application value.

[0031] 11. The high-precision energy consumption prediction of the present invention provides strong support for ship energy efficiency management systems, helping to achieve key tasks such as energy scheduling optimization, battery management strategy formulation and carbon emission monitoring.

[0032] 12. The present invention provides strong support for ship energy efficiency management and the realization of energy conservation and carbon reduction goals by improving prediction accuracy and robustness, enhancing model generalization capabilities, providing intuitive and comprehensive energy efficiency analysis tools, establishing a systematic multi-dimensional performance evaluation mechanism, optimizing the efficiency and stability of model fusion strategies, and realizing intelligent and engineering-based electric ship energy consumption modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 It is a technical flow chart of the present invention; Figure 2It is a decomposition diagram of the technical process of the present invention; Figure 3 is a schematic diagram of the stacking frame structure of the present invention; Figure 4 is a schematic diagram of the hybrid frame structure of the present invention; Figure 5 Schematic diagram of the voting framework structure of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions of the present invention are further described below with reference to the embodiments and accompanying drawings: Example 1 like Figure 1 As shown, this embodiment provides an electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation, including the following steps: Step 1: Data Collection Step 1.1: Clarify the data source (1) Ship basic parameter data: Extract basic information such as hull size, propulsion system specifications, motor parameters, battery capacity, etc. from ship design documents and technical manuals; (2) Navigational condition data: The sensor network on the ship collects data such as speed, acceleration, rudder angle, propulsion instructions, etc. in real time, and records the operation phase (such as acceleration, cruising, and deceleration); (3) Environmental data: Combined with weather station data, ocean buoy data and the ship's own environmental sensors, environmental parameters such as water flow, wind speed, wind direction, temperature, and humidity are obtained; (4) Battery system data: Real-time monitoring of battery SOC (State of Charge), voltage, current, temperature and state of health (SOH) through the battery management system (BMS); (5) Route and operation data: Extract route planning, port call times, electricity price information, and voyage schedules from the Freight Management System (FMS); (6) Auxiliary system data: records the power consumption of non-propulsion loads such as navigation, communication, lighting, and air conditioning, and is obtained through the ship's power monitoring system.

[0035] Step 1.2: Data storage and management Use database systems (such as MySQL and MongoDB) to store collected data to ensure data security and accessibility; classify and store data to facilitate subsequent processing and analysis.

[0036] Step 2: Data Preprocessing Step 2.1: Timestamp alignment and frequency synchronization Identify differences in time accuracy and sampling frequency between different data sources; perform interpolation processing (such as linear interpolation and spline interpolation) on low-frequency data and downsample high-frequency data to ensure that all data are aligned in the time dimension; use synchronous compression technology to reduce data redundancy and improve processing efficiency.

[0037] Step 2.2: Data Cleaning (1) Missing value processing: Use mean filling, median filling or machine learning-based prediction filling methods to handle missing values; (2) Outlier correction: Use box plots, Z-score and other methods to identify outliers, and use the median to replace or delete outliers; (3) Noise suppression: Apply low-pass filters (such as moving average filters and Kalman filters) to suppress high-frequency noise.

[0038] Step 2.3: Feature selection and reconstruction Calculate the Pearson correlation coefficient and mutual information between the features and the target variable (energy consumption), and evaluate the importance of the features using principal component analysis (PCA) to reduce the dimension of the features and retain the main information; Lasso (Least Absolute Shrinkage and Selection Operator) regression is used for feature selection to eliminate redundant features and construct an optimized feature set.

[0039] Step 3: Model construction Step 3.1: Single model construction Build multiple single prediction models such as linear regression, decision tree, random forest, support vector machine, etc.; train the model using training set data and adjust model parameters to optimize performance.

[0040] Step 3.2: Model selection and fusion The Pearson correlation coefficient is used to evaluate the complementarity between the individual models, and the models with high correlation and complementarity are selected for fusion: (1) Stacking method: The prediction results of multiple basic models are used as input to train a meta-model (such as a neural network) for the final prediction; (2) Hybrid method: The training dataset is divided into a training set and a validation set. The basic model is trained on the training set and predicted on the validation set. The prediction results of all basic models are combined to construct a new training set and train the meta-model. (3) Voting method: The prediction results of multiple basic models are weighted averaged to obtain the final prediction value.

[0041] Step 4: Energy consumption visualization output (1) Generation of energy consumption prediction curve: Draw a line graph of energy consumption over time to show the comparison between predicted energy consumption and actual energy consumption.

[0042] (2) Comparison chart of predicted value and actual value: Draw a superimposed comparison chart of the model prediction value and the actual historical energy consumption data to intuitively demonstrate the prediction accuracy.

[0043] (3) Multi-dimensional energy consumption distribution diagram: Construct a multi-dimensional energy consumption distribution map or heat map based on key variables (such as speed, load, and wind speed) to analyze the impact of different factors on energy consumption.

[0044] (4) Energy consumption-range matching diagram: Draw an energy consumption-range matching diagram, mark key operating nodes (such as maximum energy consumption points and abnormal peak points), and assist in energy efficiency analysis and scheduling decisions.

[0045] (5) Dynamic decision panel: Build a dynamic decision-making panel that integrates the above-mentioned visualization charts to provide real-time energy consumption forecasts, energy efficiency analysis, and scheduling recommendations.

[0046] Step 5: Multi-dimensional performance evaluation (1) Selection of evaluation indicators: The mean square error (MSE), mean absolute error (MAE), and coefficient of determination ( , R-squared) and other indicators to evaluate the model prediction accuracy; record the model training time, prediction time and other indicators to evaluate the computational efficiency; and evaluate the model generalization ability through cross-validation, leave-one-out method and other methods.

[0047] (2) Model optimization and iteration: The model is optimized and adjusted based on the evaluation results, such as adjusting model parameters, adding or reducing features, changing fusion methods, etc. The model is updated regularly to adapt to changes in the operating conditions of electric ships.

[0048] The method proposed in this example significantly improves the accuracy of electric ship energy consumption predictions, significantly enhancing the model's generalization and adaptability. The visual output of energy consumption prediction results provides ship operators with an intuitive and comprehensive energy efficiency analysis tool, facilitating energy efficiency optimization and scheduling of electric ships. Furthermore, a multi-dimensional performance evaluation mechanism provides a quantitative basis for continuous model optimization, ensuring its practicality and reliability.

[0049] Example 2 In another preferred embodiment, based on Example 1, this embodiment further describes the electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation. The prediction method of the present invention includes the following steps: Step 1: Data Collection. The core task of the data collection phase is to collect information about the target vessel from various data sources to support subsequent data preprocessing and modeling. The data sources are mainly divided into three categories: engine energy consumption data, navigation status data, and marine environment data. The collected data is integrated into a multi-source dataset, laying the data foundation for analyzing vessel energy consumption. Step 2: Data preprocessing. A series of operations are performed to improve the quality and validity of the data. First, frequency synchronization is applied to ensure that the data collection frequency of different data sources is consistent to ensure data consistency. Then, data cleaning is performed to deal with missing values, noise, and outliers to reduce the impact of data noise on the model. After that, feature selection is used to extract key features and remove redundant information. At the same time, feature reconstruction generates new features to enhance the predictive ability of the model. Step 3: Data modeling. Data modeling is the core step. First, the current mainstream ship energy consumption modeling methods (eight different single models) are selected. Then, the Pearson correlation coefficient method is used to measure the correlation between different single models as the criterion for selecting individual models. The fusion model combines multiple single models through three different methods (stacking, mixing, and voting) to improve the model's predictive performance. Step 4: Visualize the ship's energy consumption. This is based on the dynamic prediction results of the fusion model, combined with real-time marine environmental compensation corrections. The prediction uncertainty is quantified through confidence intervals, and the physical constraints of the ship's power system are coupled with historical pattern verification to generate a high-precision energy consumption time series curve. Finally, visualization is achieved with a three-dimensional energy consumption surface and a dynamic decision dashboard, providing a quantitative basis for energy efficiency optimization.

[0050] The data collection in step 1 is intended to provide a data foundation with complete structure, consistent time and reliable quality for subsequent energy consumption prediction modeling. By deploying sensors in various subsystems of electric ships, energy consumption-related information covering multiple dimensions such as power system, operating status, environmental conditions, auxiliary loads and ship structure is collected in real time, forming an original database with wide coverage and rich data dimensions.

[0051] The sources of collected data include six aspects: first, basic ship parameter data, such as hull type, main scale parameters (length, width, draft, etc.), propulsion system type, motor rated power, maximum speed, battery capacity and energy density, etc. These structural static parameters play a basic supporting role in the construction of energy consumption model; second, navigation condition data, including dynamic state information such as speed, acceleration, propulsion instructions, rudder angle changes, and operation stages (such as acceleration, cruising, deceleration, and berthing) at different time points, which directly reflects the changes in ship propulsion load; third, environmental data, including real-time acquisition of water flow velocity and direction, water depth, wave height and period, wind speed and direction, temperature, humidity and other hydrological and meteorological factors, These environmental disturbance factors are the main external variables that affect changes in navigation energy consumption; fourthly, battery system data, such as state of charge (SOC), voltage, current, temperature, discharge rate, battery health status, cycle life and abnormal records, reflect the actual output capacity and operating status of the battery system; fifthly, route and operation data, such as planned routes, port stop times, electricity price information, voyage plans and cargo conditions, etc. These operational information directly determines the ship's scheduling strategy and energy consumption behavior; sixthly, auxiliary system data, including the power consumption of non-propulsion loads such as navigation systems, communication systems, lighting, cooling and ventilation, and air conditioning. Although the power consumption is relatively low, it also has a cumulative effect that cannot be ignored during long-term operation.

[0052] Step 2: Data preprocessing: After data collection is completed, Figure 2 As shown in the figure, first, the timestamps of data from different sources and with different time precisions are aligned and the frequencies are synchronized. Due to the differences in the sampling frequencies of sensors in various subsystems, interpolation and resampling operations need to be performed based on a unified time axis. For example, linear interpolation and spline interpolation are used to fill in the missing time points of low-frequency data, and high-frequency data is downsampled and synchronously compressed to ensure the consistency and comparability of multi-source data in the time dimension. In addition, a frequency coordination mechanism is used to process asynchronous data sources to ensure that the data-driven model can learn based on fully aligned input features during modeling.

[0053] Perform systematic data cleaning to improve the quality and robustness of the original data; for missing data points, use a variety of interpolation strategies to fill them according to the variable type and distribution characteristics, such as mean / median filling, Nearest neighbor interpolation, sliding window averaging, etc.; for outliers that deviate significantly from the normal range, combined with The (Zero-mean normalization--score) standardized detection method is used for screening and correction, and the non-physical values are eliminated by combining the historical sample rules and boundary rules. For high-frequency noise and short-term spike interference, sliding filtering, wavelet denoising, exponential smoothing and other methods are applied to suppress data fluctuations and improve data continuity and stability.

[0054] After completing the initial cleaning and normalization, the feature engineering stage begins. Targeted at the modeling goal of ship energy consumption prediction, the original variables are screened and transformed. By calculating the correlation coefficient and mutual information between the features and the target variables, combined with dimensionality reduction methods such as principal component analysis (PCA) and Lasso regression, redundant information or weakly correlated features are eliminated and key variables are retained. At the same time, feature variables with strong nonlinear distributions or scale differences are logarithmically transformed and standardized to improve the convergence speed and accuracy of model training. For categorical variables such as navigation status and control mode, One-Hot encoding (also known as one-bit effective encoding) or label encoding is used for numerical processing to facilitate direct model processing.

[0055] Through the above-mentioned multi-source data collection and rigorous data preprocessing process, a data set with unified structure, time synchronization, complete features, and noise suppression is finally formed, which serves as the high-quality input basis for the subsequent multi-model fusion energy consumption prediction method of the present invention; this step ensures that the model has real, comprehensive and highly interpretable data support during the learning process, thereby improving the accuracy and stability of the prediction results from the source.

[0056] Step 3: Data modeling and building a stacking model, as follows: Stacking is a fusion method that improves the overall model performance by combining the prediction results of multiple different base models. The goal of stacking is to capture different patterns or structures in the data through different base models, thereby enhancing the prediction performance.

[0057] Stacking combines multiple layers of models for learning, usually divided into two levels, such as Figures 3 to 5 shown.

[0058] Level 0: This layer contains multiple different basic models, which can be various algorithms, such as linear regression, decision tree, random forest, support vector machine, etc. Before using the level 0 basic model for training and prediction, The ship energy consumption dataset is first split into a training set and a test set according to a certain ratio (usually 8:2). In order to prevent data leakage during model training, fold cross validation. The model is The training is done on 1 fold of data and the remaining 1 fold is used for prediction. Each base model gets a prediction value in the form of a one-dimensional array. The prediction output of the base model is: (1) Where, 、 、…、 Represents basic models 1, 2, ..., The predicted output of Representative data samples; 、 、…、 Represents basic models 1, 2, ; Indicates the sequence number of the base model.

[0059] A new training set can be obtained through the basic model layer. The training set is as follows: (2) Where, represents the number of rows in the training set, represents the sequence number of the basic model in the training set, and 、 Respectively represent the first basic model for the first row, The predicted value of the row data, 、 Respectively represent The basic model is for the first row, The predicted value of the row data.

[0060] From formula (2), we can see that the new training set This is done by applying The results of the fold cross validation are obtained using the base model at level 0. The size of the new training set, in terms of the number of samples (rows), is the same as that of the original training set. However, the number of features in the new training set has changed. The original training set had 9 features, but after passing through the first layer of the base model, it has features (number of base models used).

[0061] In the test set, the trained base model is used to make predictions. times, thus obtaining predicted values, and these values are averaged. Similarly, by combining The predictions of the base model are used to obtain a new test set. As shown below: (3) Where, represents the number of rows in the test set, represents the predicted number of repetitions; and 、 、 、 Respectively represent the first basic model for the first time in the first row and the first time in the first row sequence The first and second Rank The predicted value of the secondary data; 、 、 、 Respectively represent The basic model is used for the first time in the first row and the first time in the first row. sequence The first and second Rank The predicted value of the data.

[0062] 2) Level 1: Metamodel layer The new training set obtained from layer 0 is used to train the meta-model, and then the trained meta-model is used to predict the new test set to obtain the final prediction results.

[0063] Stacking improves the accuracy of final predictions by combining the predictions of multiple base models and using a meta-model to further optimize the predictions. Through cross-validation and combining different models, stacking enhances the generalization ability of the model, reduces overfitting, and has been widely used in many complex prediction tasks.

[0064] Step 3: Data modeling and building a blending model, as follows: Mixing is similar to stacking, but unlike stacking, it does not use cross-validation; instead, it splits the training dataset into training and validation sets. In hybrid, the base models are trained on the training set and predicted on the validation set. A new training set is constructed by combining the predictions of all the base models on the validation set. As shown below: (4) Where, Indicates the ratio of the training set divided into the training set and the validation set, usually ranging from 0.1 to 0.4.

[0065] Similarly, the trained base model is used to directly predict the test set; since no fold( ) cross validation, so there is no need to average the predicted values; by combining the predictions of all base models, a new test set is obtained, the new test set As shown below: (5) Then, the new training set is used to train the meta-model, and the trained meta-model is used to predict the new test set to generate the final prediction results.

[0066] The hybrid model does not require complex cross-validation; instead, it directly uses the split validation set for prediction and training, making the process simple and easy to implement. However, since the hybrid model only uses the prediction results of the validation set, the relatively small number of validation samples may lead to insufficient performance of the meta-model, affecting the final prediction accuracy.

[0067] Step 3: Data modeling and building a voting model, as follows: Voting is a simple and effective multi-model fusion method. Its core idea is to combine the prediction results of multiple base models and use a "voting" mechanism to determine the final prediction result.

[0068] First, multiple different regression models (base models) are trained on the same dataset; then, the trained models are used to predict the test data, and finally, the final prediction value is obtained by weighted average. , the formula is as follows: (6) Where, Representative The weights of the models; if the weight values are the same, it is a simple average algorithm; Indicates the The predicted values of the model test data.

[0069] Step 4: Generate visualization of ship energy consumption to enhance the readability of model output and the intuitiveness of engineering applications. This step specifically includes the structured organization of prediction results, graphical display, and multi-dimensional comparative analysis, as follows: The moment-by-moment predicted energy consumption data output by the model is reorganized into a time series to form a standardized data matrix. Each record contains the predicted timestamp, predicted energy consumption value (in kWh), corresponding navigation state identifier (such as acceleration, cruising, berthing), geographic location information, and a snapshot of the main input variables. This data structure is used for subsequent graph processing and dynamic querying.

[0070] The system outputs the prediction results in graphical form. First, it generates a line graph showing the change of energy consumption over time, with the horizontal axis representing the sailing time and the vertical axis representing the energy consumption value per unit time, which intuitively reflects the high and low fluctuations in energy consumption and the time series relationship; second, it superimposes and compares the model prediction value with the actual historical energy consumption data, and draws a two-line comparison graph or error bar graph to show the prediction accuracy and model residual range; third, it draws a multi-dimensional energy consumption distribution graph or heat map based on key variables such as speed, load, wind speed, etc., to reveal the sensitivity and nonlinear relationship between input variables and energy consumption output.

[0071] In a specific implementation, the system can further provide an energy consumption-range matching diagram, that is, projecting the predicted energy consumption value according to the corresponding flight segment, outputting the energy consumption curve per unit distance (in kWh / km), and marking key operating nodes such as maximum energy consumption points, abnormal peak points, feedback energy points, etc., to assist users in diagnosing range energy efficiency.

[0072] In terms of fusion model evaluation, the system supports visual comparison of the prediction outputs of multiple sub-models and the integrated model output, and statistically annotates the corresponding average errors, graphically clarifying the performance improvement brought about by model fusion.

[0073] The system integrates the above-mentioned graphical output results and structured energy consumption forecast data into the forecast analysis report module, which can generate a comprehensive report containing energy consumption forecast overview, operating condition segmentation evaluation, model residual statistics, etc., and supports output as a visual report for user browsing, scheduling optimization or decision-making reference.

[0074] In the preferred solution, in the data collection step of Step 1, the data sources include basic ship parameter data, navigation condition data, environmental data, battery system data, route and operation data, and auxiliary system data; the above settings can comprehensively cover the key information of ship operation, provide a solid foundation for subsequent data analysis and decision support, ensure the effectiveness and accuracy of the solution, optimize ship performance, and improve operational efficiency.

[0075] In the preferred solution, the data preprocessing in Step 2 also includes timestamp alignment and frequency synchronization. This involves interpolating and resampling data with varying time precision and sampling frequency, or performing downsampling and synchronous compression operations to ensure consistency and comparability across the time dimension of multi-source data. This setup significantly improves the efficiency and accuracy of data fusion. This step also incorporates an outlier detection and correction mechanism to automatically identify and correct unreasonable values in the data, further enhancing the reliability of the data and the accuracy of the analysis results.

[0076] In the preferred solution, in the data preprocessing step of Step 2, data cleaning includes filling missing values, correcting outliers and suppressing high-frequency noise to improve the quality and robustness of the original data; the above settings ensure the consistency and reliability of the data and provide a solid foundation for subsequent data analysis; in addition, normalization is also performed in the data preprocessing stage to eliminate the influence of different dimensions on model training.

[0077] In the preferred solution, in the data preprocessing step of Step 2, feature selection and reconstruction include calculating the correlation coefficient and mutual information between the features and the target variable, using principal component analysis PCA, Lasso regression and other dimensionality reduction methods to eliminate redundant or weakly correlated features and retain key variables; the above settings are intended to improve model efficiency and prediction accuracy; in addition, missing values are interpolated, such as mean interpolation, Nearest neighbor interpolation, etc., to ensure data integrity; and detect and correct outliers to avoid their negative impact on model training.

[0078] In the preferred solution, in the model building step of Step 3, the stacking method combines the prediction results of multiple different basic models and uses the meta-model to further optimize the prediction to improve the prediction accuracy; the above settings enable the model to fully utilize the advantages of each basic model. At the same time, the introduction of the meta-model reduces the deviation that may be caused by a single model, thereby improving the stability and reliability of the overall prediction.

[0079] In the preferred solution, in the model construction step of Step 3, the hybrid method divides the training data set into a training set and a validation set. The basic model is trained on the training set and predicted on the validation set. By combining the predictions of all basic models on the validation set, a new training set is constructed to train the meta-model. The above settings are intended to improve the generalization ability of the model. The meta-model further learns based on the new training set and optimizes the prediction performance. This process is iterative until the model performance is stable to ensure the prediction accuracy of the final model on unseen data.

[0080] In the preferred solution, during the model building step (Step 3), a voting method combines the prediction results of multiple base models, obtaining the final prediction value through weighted averaging. This setup significantly improves the stability and accuracy of the prediction. For further optimization, a model fusion strategy is introduced, utilizing ensemble learning methods to combine the advantages of different base models, thereby further enhancing prediction performance.

[0081] In the preferred solution, the energy consumption visualization output of Step 4 includes generating a line graph showing energy consumption changes over time, an overlay comparison graph of model prediction values and actual historical energy consumption data, a multi-dimensional energy consumption distribution graph or heat map based on key variables, and an energy consumption-range matching graph; the above settings are intended to intuitively display energy consumption trends, prediction accuracy, energy consumption distribution characteristics and their relationship with range, provide decision makers with a comprehensive basis for energy consumption analysis and optimization, and promote efficient energy utilization and navigation strategy optimization.

[0082] In the preferred scheme, the method also includes a multi-dimensional performance evaluation step to optimize the energy consumption prediction model in terms of prediction accuracy, computational efficiency, model generalization ability, etc., so as to improve its practicality and reliability; the above settings further ensure that the model can maintain stable performance when facing different scenarios and data distributions. At the same time, by introducing a real-time feedback mechanism and dynamically adjusting the model parameters, rapid response and accurate prediction of energy consumption changes are achieved.

[0083] In summary, the electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation provided by the present invention has achieved key innovations and breakthroughs in the field of intelligent shipping and ship energy management technology. This method proposes a comprehensive and detailed solution to the core problems existing in the prediction of electric ship energy consumption, such as insufficient accuracy, weak generalization ability, and lack of a unified evaluation system. By integrating multi-source heterogeneous data for the first time, covering multi-dimensional information such as basic ship parameters, navigation conditions, environment, battery system, route and operation, and auxiliary systems, the comprehensiveness and accuracy of the prediction model are ensured. In the data preprocessing link, a refined process is adopted, including data cleaning, feature selection and reconstruction, and timestamp alignment and frequency synchronization processing, which significantly improves the quality and availability of the data.

[0084] This invention innovatively proposes a model selection and fusion strategy based on the Pearson correlation coefficient. Using three fusion methods, stacking, blending, and voting, it achieves intelligent integration of multiple single prediction models, significantly improving prediction accuracy and robustness. Furthermore, the constructed multidimensional visualization interface, including energy consumption prediction curves and predicted vs. actual value comparison charts, provides ship operators with an intuitive and comprehensive energy efficiency analysis tool, a first in the art.

[0085] Furthermore, this invention overcomes the limitations of traditional electric ship energy consumption prediction methods, which rely solely on a single data source or simple model fusion. By integrating heterogeneous data from multiple sources and employing a complex multi-model fusion strategy, it significantly improves prediction accuracy. In terms of data processing, the invention utilizes a variety of advanced algorithms, including feature importance assessment based on the Pearson correlation coefficient and mutual information, PCA dimensionality reduction, and Lasso regression feature selection, demonstrating significant innovation.

[0086] Overall, this invention not only improves the accuracy and robustness of electric vessel energy consumption forecasting but also provides scientific decision-making support for ship operators through a multidimensional visualization interface, contributing to energy efficiency optimization and intelligent scheduling decisions. By integrating multi-source heterogeneous data, optimizing data processing processes, integrating multiple prediction models, and constructing a multidimensional visualization interface, this invention has brought substantial progress to the fields of intelligent shipping and energy management for electric vessels, demonstrating significant practicality and application value.

Claims

1. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation is characterized by: The following steps are involved: Step 1: Data collection: Collect multi-source heterogeneous data related to electric ship operation; Step 2: Data preprocessing, preprocessing the collected data, including data cleaning, feature selection, and feature reconstruction; Step 3: Model construction: construct multiple single prediction models, use the Pearson correlation coefficient for model selection, and apply three fusion methods: stacking, mixing, and voting to achieve model integration; Step 4: Energy consumption visualization output, generate high-precision energy consumption prediction curve, build a three-dimensional energy consumption prediction surface map, build a dynamic decision panel, and assist energy efficiency optimization and scheduling.

2. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 1 is characterized by: In the data collection step of Step 1, the data sources include basic ship parameter data, navigation condition data, environmental data, battery system data, route and operation data, and auxiliary system data.

3. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 2 is characterized by: The data preprocessing in Step 2 also includes timestamp alignment and frequency synchronization processing, which is to perform interpolation and resampling operations, or downsampling and synchronous compression operations on data with different time precisions and sampling frequencies to ensure the consistency and comparability of multi-source data in the time dimension.

4. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 3 is characterized by: In the data preprocessing step of Step 2, data cleaning includes filling missing values, correcting outliers, and suppressing high-frequency noise to improve the quality and robustness of the original data.

5. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 4 is characterized by: In the data preprocessing step of Step 2, feature selection and reconstruction include calculating the correlation coefficient and mutual information between the features and the target variable, using principal component analysis (PCA) and Lasso regression dimensionality reduction methods to eliminate redundant or weakly correlated features and retain key variables.

6. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 5 is characterized by: In the model building step of Step 3, the stacking method combines the prediction results of multiple different basic models and uses a meta-model to further optimize the prediction to improve the prediction accuracy.

7. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 6 is characterized by: In the model construction step of Step 3, the hybrid method divides the training data set into a training set and a validation set. The base model is trained on the training set and predicted on the validation set. By combining the predictions of all base models on the validation set, a new training set is constructed to train the meta-model.

8. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 7 is characterized by: In the model building step of Step 3, the voting method combines the prediction results of multiple basic models and obtains the final prediction value through weighted average.

9. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 8 is characterized by: The energy consumption visualization output of Step 4 includes generating a line graph showing energy consumption changes over time, a superimposed comparison graph of model prediction values and actual historical energy consumption data, a multi-dimensional energy consumption distribution graph or heat map based on key variables, and an energy consumption-range matching graph.

10. The electric ship energy consumption prediction method based on multi-model fusion and multi-dimensional evaluation according to claim 9 is characterized by: The method also includes a multi-dimensional performance evaluation step to optimize the energy consumption prediction model in terms of prediction accuracy, computational efficiency, model generalization ability, etc., to improve its practicality and reliability.

Images