A high-precision prediction system for ship energy consumption based on multi-model fusion
By using multi-model fusion and optimization algorithms, a high-precision ship energy consumption prediction system was constructed, which solved the problems of low prediction accuracy and poor adaptability in existing technologies, and achieved more accurate energy consumption prediction.
Patent Information
- Application Number
- CN202311158277.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-08
- Publication Date
- 2026-01-30
- Estimated Expiration
- 2043-09-08
AI Technical Summary
Existing ship energy consumption prediction models have low prediction accuracy and fail to comprehensively consider various influencing factors such as navigation weather data, resulting in weak adaptability of energy consumption prediction models to the navigation environment and difficulty in achieving accurate prediction.
A multi-model fusion approach is adopted, which combines a data collection and analysis module, a single-model algorithm ship energy consumption prediction module, and a multi-model fusion ship energy consumption prediction module. By combining the Stacking model, Bayesian optimization algorithm, and adaptive algorithm, the ship energy consumption prediction model is optimized and a high-precision prediction system is constructed.
It improves the accuracy and adaptability of ship energy consumption forecasting, achieving more accurate energy consumption prediction.
Smart Images
Figure CN117290673B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ship energy consumption prediction applications, and in particular to a high-precision ship energy consumption prediction system based on multi-model fusion. Background Technology
[0002] Intelligent energy efficiency management for ships is one of the effective ways to achieve green and intelligent development of ships. To conduct research on intelligent prediction methods for ship energy consumption and further improve the level of intelligent energy efficiency management, it is necessary to carry out research on more accurate and efficient ship energy consumption prediction methods and technologies. However, existing ship energy consumption prediction models have low prediction accuracy and do not comprehensively consider various variable influencing factors such as navigation weather data, resulting in weak adaptability of energy consumption prediction models to the navigation environment and difficulty in achieving accurate prediction of ship energy consumption. Summary of the Invention
[0003] This invention provides a high-precision prediction system for ship energy consumption based on multi-model fusion to overcome the above-mentioned technical problems.
[0004] A high-precision ship energy consumption prediction system based on multi-model fusion includes a data collection and analysis module, a single-model algorithm ship energy consumption prediction module, a multi-model fusion ship energy consumption prediction module, and a human-computer interaction module.
[0005] The data collection and analysis module is used to collect ship energy efficiency data and meteorological data, and to preprocess the ship energy efficiency data and meteorological data. After performing feature selection on the preprocessed ship energy efficiency data and meteorological data according to the feature selection method, a feature dataset is obtained. The feature dataset is then divided into training set and test set according to a certain ratio.
[0006] The single-model algorithm for ship energy consumption prediction is used to construct different types of ship energy consumption prediction models using a training set and test them using a test set. Based on the test results, performance evaluation indices for different models are calculated. The set of ship energy consumption prediction models that meet the constraints is then selected based on the values of these performance evaluation indices.
[0007] The multi-model fusion ship energy consumption prediction module is used to fuse a set of ship energy consumption prediction models based on the Stacking model fusion method, and represents the fused model as a fusion model. The module then optimizes the parameters of the fusion model using Bayesian optimization and adaptive algorithms, and predicts ship energy consumption based on the optimized fusion model.
[0008] The human-computer interaction module is used to display the analysis and processing, operation process and analysis results of the data collection and analysis module, the single-model algorithm ship energy consumption prediction module and the multi-model fusion ship energy consumption prediction module.
[0009] Preferably, the single-model algorithm ship energy consumption prediction module includes a prediction model unit based on statistical machine learning, a prediction model unit based on neural network algorithms, a prediction model unit based on ensemble learning algorithms, and a single-model comparison analysis unit. The prediction model unit based on statistical machine learning includes ship energy consumption prediction models based on multiple linear regression, support vector regression, and K-nearest neighbors. The prediction model unit based on neural network algorithms includes ship energy consumption prediction models based on backpropagation (BP) neural networks and long short-term memory (LSTM) neural networks. The prediction model unit based on ensemble learning algorithms includes ship energy consumption prediction models based on decision trees, random forests, extreme gradient boosting, extreme random trees, and AdaBoost. The single-model comparison analysis unit is used to test different types of ship energy consumption prediction models based on a test set and obtain test results. Based on the test results, it calculates performance evaluation indicators for different models and selects a set of ship energy consumption prediction models that meet the constraints based on the values of the performance evaluation indicators.
[0010] Preferably, the step of selecting a set of ship energy consumption prediction models that meet the constraints based on the values of performance evaluation indicators includes setting an objective function based on the performance evaluation indicators.
[0011] The first constraint condition is set based on the coefficient of determination in the performance evaluation index. Ship energy consumption prediction models that meet the first constraint condition are stored in the first filter model set. The objective function values of the first filter model set are calculated and sorted in descending order according to the objective function values.
[0012] A second constraint is set, which is a constraint on the number of models. Based on the second constraint, models are selected sequentially from the first filter model set and stored in the ship energy consumption prediction model set.
[0013] Preferably, the step of setting the objective function based on the performance evaluation index is to set the objective function according to formula (1).
[0014]
[0015] In the formula, R 2 MAPE, RMSE, MAE, and MSE are the coefficient of determination, mean absolute percentage error, root mean square error, mean absolute error, and mean square error, respectively; w1, w2, w3, w4, and w5 are different weighting coefficients that add up to 1.
[0016] Preferably, the step of optimizing the parameters of the fusion model based on the Bayesian optimization algorithm and the adaptive algorithm includes,
[0017] Step S1: Obtain the ship energy consumption prediction models from the ship energy consumption prediction model set and represent them as the basic models.
[0018] Step S2: Based on the Stacking model fusion method, fuse the obtained base models and represent the fused model as the fused model.
[0019] Step S3: Adjust the near-optimal hyperparameter combination parameters in the parameter domain space of the basic model by using Bayesian optimization algorithm and adaptive algorithm.
[0020] Step S4: Based on the approximate optimal hyperparameter combination of the obtained basic model, use it as the parameters of the fusion model to obtain the fusion model optimized by Bayesian optimization algorithm and adaptive algorithm.
[0021] Preferably, step S2 includes constructing a model with a two-layer structure. The first layer of the model is a basic learning layer, and the second layer of the model is a meta-learning layer. The basic learning layer is used to perform K-fold cross-validation training on the ship energy consumption prediction models in the ship energy consumption prediction model set according to the feature dataset and generate a new dataset. The new dataset is divided into a new training set and a new test set. The meta-learning layer is used to construct a meta-model and learn the meta-model according to the new dataset. The learning includes training and testing the meta-model according to the new training set and the new test set respectively. The trained and tested meta-model is represented as a fusion model.
[0022] This invention provides a high-precision prediction system for ship energy consumption based on multi-model fusion. An objective function is designed to calculate the performance evaluation index values of different types of ship energy consumption prediction models. Based on first and second constraints, several high-performing single models are selected, and these high-performing single models are fused according to a Stacking model to construct a ship energy consumption prediction fusion model based on the Stacking framework. The fusion model is then optimized using Bayesian optimization and adaptive algorithms, thereby improving the prediction accuracy of the ship energy consumption prediction fusion model. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 This is a system structure diagram of the present invention;
[0025] Figure 2This is a flowchart of the high-precision prediction process for ship energy consumption based on multi-model fusion, as described in this invention.
[0026] Figure 3 This is a schematic diagram illustrating the construction process of the ship energy consumption prediction model set of this invention;
[0027] Figure 4 This is a diagram of the Stacking model framework of the present invention;
[0028] Figure 5 This is a schematic diagram illustrating the process of establishing a Bayesian optimization fusion model based on the present invention.
[0029] Figure 6 This is a schematic diagram illustrating the specific implementation process of constructing the fusion model of the present invention. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0031] Figure 1 This is a system structure diagram of the present invention, such as... Figure 1 As shown, the method in this embodiment may include a data collection and analysis module, a single-model algorithm ship energy consumption prediction module, a multi-model fusion ship energy consumption prediction module, and a human-computer interaction module.
[0032] The data collection and analysis module is used to collect ship energy efficiency data and meteorological data, and to preprocess the ship energy efficiency data and meteorological data. After performing feature selection on the preprocessed ship energy efficiency data and meteorological data according to the feature selection method, a feature dataset is obtained. The feature dataset is then divided into training set and test set according to a certain ratio.
[0033] The single-model algorithm for ship energy consumption prediction is used to construct different types of ship energy consumption prediction models using a training set and test them using a test set. Based on the test results, performance evaluation indices for different models are calculated. The set of ship energy consumption prediction models that meet the constraints is then selected based on the values of these performance evaluation indices.
[0034] The multi-model fusion ship energy consumption prediction module is used to fuse a set of ship energy consumption prediction models based on the Stacking model fusion method, and represents the fused model as a fusion model. The module then optimizes the parameters of the fusion model using Bayesian optimization and adaptive algorithms, and predicts ship energy consumption based on the optimized fusion model.
[0035] The human-computer interaction module is used to display the analysis and processing, operation process and analysis results of the data collection and analysis module, the single-model algorithm ship energy consumption prediction module and the multi-model fusion ship energy consumption prediction module.
[0036] This invention provides a high-precision prediction system for ship energy consumption based on multi-model fusion. An objective function is designed to calculate the performance evaluation index values of different types of ship energy consumption prediction models. Based on first and second constraints, several high-performing single models are selected. These high-performing single models are then fused using a Stacking model to construct a Stacking-based ship energy consumption prediction fusion model. The fusion model is further optimized using Bayesian optimization and adaptive algorithms, thereby improving the prediction accuracy of the ship energy consumption prediction fusion model.
[0037] Specifically, this embodiment provides a detailed description of a high-precision ship energy consumption prediction system based on multi-model fusion. The data collection and analysis module can be divided into a data collection unit, a data preprocessing unit, and a data analysis unit. The single-model algorithm ship energy consumption prediction module can be divided into a prediction model unit based on statistical machine learning, a prediction model unit based on neural network algorithms, a prediction model unit based on ensemble learning algorithms, and a single-model comparative analysis unit. The multi-model fusion ship energy consumption prediction module includes adaptive fusion models, Bayesian optimization fusion models, and multi-model fusion. The human-computer interaction module can be divided into a display unit, a management unit, and an operation unit. The prediction process of this invention is as follows: Figure 2 As shown.
[0038] The data collection and analysis module is used to collect ship energy efficiency data and meteorological data, and to preprocess the ship energy efficiency data and meteorological data. It performs feature selection on the preprocessed ship energy efficiency data and meteorological data according to the feature selection method. The data collection and analysis module includes a data collection unit, a data preprocessing unit, and a data analysis unit. The data collection unit is used to acquire ship energy efficiency data through multi-source sensors installed on the ship and to acquire meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF). Specifically, it collects operational data such as latitude and longitude, speed, heading, draft, main engine speed, power, and fuel oil quantity during the ship's navigation through multi-source sensors installed on the ship, and collects meteorological data such as wind speed, wind direction, and characteristic wave height from the ECMWF.
[0039] The data preprocessing unit is used to preprocess ship energy efficiency data and meteorological data. The preprocessing includes outlier cleaning, feature selection, standardization, dimensionless elimination, and data scaling. The data analysis unit is used to normalize and perform correlation analysis on the preprocessed ship energy efficiency data and meteorological data, and then represent them as the first dataset. According to the Pearson correlation coefficient method, features with high correlation coefficients with ship energy consumption are selected from the first dataset to form a feature dataset, and the training set and test set are divided according to a certain ratio. Among them, the data normalization process adopts the Min-max method to map the ship energy efficiency data to the [0,1] interval. The known ship energy consumption data is transformed, and the specific formula (1) is as follows:
[0040]
[0041] In the formula: x min x represents the minimum value of the sample data; max x represents the maximum value of the sample data; k This represents any value in the sample data; This represents the normalized value of any number.
[0042] The correlation analysis uses the Pearson correlation coefficient method to analyze the correlation between ship energy consumption and various input characteristics. The specific formula (2) is as follows:
[0043]
[0044] In the formula: These represent the average values of sample features x and y, respectively; x i The value of feature x of the i-th sample; y i The value of feature y for the i-th sample is denoted by ; n represents the number of samples corresponding to the feature; r represents the correlation coefficient, used to measure the degree of correlation between feature x and feature y. r is in the range -1 < r < 1, r > 0 indicates a positive correlation, and r < 0 indicates a negative correlation.
[0045] The single-model algorithm for ship energy consumption prediction is used to construct different types of ship energy consumption prediction models using training set data, and to calculate the performance evaluation index of different models based on the test results of test set data. Based on the values of the performance evaluation index, a set of ship energy consumption prediction models that meet the constraints is selected.
[0046] The single-model algorithm ship energy consumption prediction module includes a prediction model unit based on statistical machine learning, a prediction model unit based on neural network algorithm, a prediction model unit based on ensemble learning algorithm, and a single-model comparative analysis unit.
[0047] The prediction model unit based on statistical machine learning includes a ship energy consumption prediction model based on multiple linear regression, a ship energy consumption prediction model based on support vector regression, and a ship energy consumption prediction model based on the K-nearest neighbor algorithm.
[0048] The prediction model unit based on neural network algorithms includes a ship energy consumption prediction model based on BP neural network and a ship energy consumption prediction model based on long short-term memory neural network.
[0049] The prediction model unit based on ensemble learning algorithms includes a ship energy consumption prediction model based on decision trees, a ship energy consumption prediction model based on random forest algorithms, a ship energy consumption prediction model based on extreme gradient boosting algorithms, a ship energy consumption prediction model based on extreme random trees, and a ship energy consumption prediction model based on AdaBoost.
[0050] The single-model comparison and analysis unit is used to test different types of ship energy consumption prediction models based on the test set and obtain test results. Based on the test results, performance evaluation indicators for different models are calculated. A set of ship energy consumption prediction models that meet the constraints is selected based on the values of the performance evaluation indicators. Selecting a set of ship energy consumption prediction models that meet the constraints based on the values of the performance evaluation indicators includes setting an objective function based on the performance evaluation indicators. Setting the objective function based on the performance evaluation indicators is based on formula (3).
[0051]
[0052] In the formula, R 2 MAPE, RMSE, MAE, and MSE are the coefficient of determination, mean absolute percentage error, root mean square error, mean absolute error, and mean square error, respectively; w1, w2, w3, w4, and w5 are different weighting coefficients that add up to 1. The relative importance of these indicators varies depending on the requirements of the model. Therefore, after comprehensively considering the predictive accuracy and generalization of the model, the weights of the indicators are assigned as shown in formula (4):
[0053]
[0054] The first constraint condition is set based on the coefficient of determination in the performance evaluation index. Ship energy consumption prediction models that meet the first constraint condition are stored in the first filter model set. The values of the objective function in the first filter model set are calculated and sorted in descending order according to the values of the objective function.
[0055] A second constraint is set, which is a constraint on the number of models. Based on the second constraint, models are selected sequentially from the first filter model set and stored in the ship energy consumption prediction model set.
[0056] The process of constructing a ship energy consumption prediction model set is as follows: Figure 3 As shown, the steps are as follows:
[0057] Step 1: Select multiple candidate base models for training and evaluation, and calculate the comprehensive evaluation index S and the coefficient of determination R for each model. 2 The values are stored in the first filter model set - style dictionary DataFrame;
[0058] Step 2: According to R 2 Constraints were considered, and to ensure the model's prediction accuracy was maximized, R was selected. 2 Models with a value greater than or equal to 0.9 are sorted in descending order according to the size of the comprehensive evaluation index S.
[0059] Step 3: Based on the model number constraint, select the n models with the largest S value that meet the conditions in sequence. In this embodiment, the value of n is 4.
[0060] Step 4: Construct a set of ship energy consumption prediction models based on all selected models.
[0061] The multi-model fusion ship energy consumption prediction module is used to fuse a set of ship energy consumption prediction models based on the Stacking model fusion method, and the resulting fused model is shown in the model framework diagram below. Figure 4 As shown, the fusion model based on the Stacking model fusion method, after fusing the ship energy consumption prediction model set, is represented as a fused model. This involves constructing a two-layer model: a first layer (basic learning layer) and a second layer (meta-learning layer). The basic learning layer trains the ship energy consumption prediction models in the set using K-fold cross-validation based on the feature dataset, generating a new dataset. This new dataset is then divided into a training set and a test set. The meta-learning layer constructs a meta-model, trains and tests it using the new training and test sets, and the trained and tested meta-model is represented as the fused model. To avoid overfitting, K-fold cross-validation is used. Its advantage lies in considering the characteristics of the samples in the dataset, ensuring that the sub-training and validation sets do not repeat learning, thus reducing the probability of overfitting and improving the performance of the fused model. In this embodiment, K is set to 5.
[0062] The parameters of the fusion model are optimized using Bayesian optimization and adaptive algorithms. This optimization includes...
[0063] Step S1: Obtain the ship energy consumption prediction models from the ship energy consumption prediction model set and represent them as the basic models.
[0064] Step S2: Based on the Stacking model fusion method, fuse the obtained base models and represent the fused model as the fused model.
[0065] Step S3: Adjust the near-optimal hyperparameter combination parameters of the basic model in the parameter domain space using Bayesian optimization and adaptive algorithms; the process of establishing the fusion model based on Bayesian optimization is as follows. Figure 5 As shown, step S3 can be divided into the following processes:
[0066] S31. Based on the parameter domain space and the number of hyperparameter combinations θ1-θ of the basic models M1-M5 n Randomly generate initial sample points x i As input to the Gaussian model, the training set is used as input to the base model, utilizing the base model's performance on the validation set. i Value (i.e., 1-S value) and sample point x i The Gaussian model is modified to make it closer to the true distribution of the function;
[0067] S32. Using the sampling function, select the parameter combination point x to be evaluated next from the modified Gaussian model. i This allows the Gaussian model to more quickly and accurately approximate the true distribution of the objective function compared to other candidate sets. The sampling function can be: expectation boosting strategy sampling function, probability boosting strategy sampling function, Gaussian process upper confidence boundary strategy sampling function, information-based Thompson sampling function, and entropy increasing strategy sampling function.
[0068] S33, the new x i and y i The input is fed into the Gaussian model, and the process jumps to S32 to correct the Gaussian process. The next evaluation point is then selected based on an adaptive algorithm until a termination condition is met, where the fitness function value satisfies a threshold. At this point, the minimum y... i The corresponding x i The value is the optimal combination of hyperparameters in the parameter domain space, thus obtaining the approximate optimal combination of hyperparameters for the basic models M1-M5;
[0069] Step S4: Based on the approximately optimal hyperparameter combination of the obtained basic models M1-M5, use it as the parameters of the fusion model to obtain the fusion model optimized by Bayesian optimization algorithm and adaptive algorithm.
[0070] Ship energy consumption is predicted based on the optimized fusion model.
[0071] Specifically, the schematic diagram of the specific implementation process of the fusion model construction is as follows: Figure 6 As shown, the main steps include:
[0072] Step A1: Select the base model. Multiple base models M1, M2, ..., Mn are selected using the base model selection strategy. n ;
[0073] Step A2: Data partitioning. The processed ship energy efficiency data D is divided into a training set X and a test set C in an 8:2 ratio. The training dataset X is then divided into 5 non-overlapping parts using cross-validation, denoted as P1 to P5.
[0074] Step A3: A cross-validation process. Model A1, trained on training sets P1-P4, makes predictions on the test set P5;
[0075] Step A4: Single-model cross-validation process. After the first cross-validation is completed, the predicted value of the current test set P5 is obtained, denoted as r1; and so on, eventually r2, r3, r4, and r5 are obtained. Their averages are used to form a prediction matrix for the training set X, denoted as R1. After the entire cross-validation is completed, the predicted value of model A1 for the test set C of the original dataset is obtained, denoted as s1; and so on, s2, s3, s4, and s5 are obtained, and their averages are taken to obtain S1.
[0076] Step A5: Repeat step A4 for each model to obtain the R2, R3, R4, R5 and S2, S3, S4, S5 matrices. Combine the R1-R5 models to form a new training set X. x S1-S5 form a new test set C c ;
[0077] Step A6: Meta-model training. In the second layer, the new training set X... x The input is fed into the next layer's meta-model, RR model, for training. The RR model learns from the training set, assigning weights to the predictions of the base model, thus making the predictions more accurate. The training process ends, and the final fused model is obtained after fusion.
[0078] Step A7: Model Prediction. Based on the new test set C c The trained meta-model is used for prediction, and the result is output as the final prediction.
[0079] The human-computer interaction module is used to display the analysis and processing, operation process and analysis results of the data collection and analysis module, the single-model algorithm ship energy consumption prediction module and the multi-model fusion ship energy consumption prediction module. The human-computer interaction module includes a display unit, a management unit and an operation unit. The display unit is used to display on a display device, the management unit is used to determine the information that can be displayed, and the operation unit is used to display the information determined by the management unit through the display unit.
[0080] A high-precision prediction method for ship energy consumption based on multi-model fusion includes the following steps:
[0081] Step 1-1: Collect relevant data through the data collection and analysis module, preprocess the data, perform correlation analysis, select appropriate feature values, and transmit them to the single-model algorithm ship energy consumption prediction module.
[0082] Steps 1-2: The single-model algorithm ship energy consumption prediction module receives the data selected by the data collection and analysis module, performs single-model testing, compares and analyzes the performance of different models, and selects multiple models with better performance to construct a ship energy consumption prediction model set.
[0083] Steps 1-3: The multi-model fusion ship energy consumption prediction module uses the Stacking model fusion method to fuse the basic models in the ship energy consumption prediction model set built by the single-model algorithm ship energy consumption prediction module to form a fused model. Then, the fused model is optimized using Bayesian optimization algorithm and adaptive algorithm, and the ship energy consumption is predicted based on the optimized fused model.
[0084] Overall beneficial effects:
[0085] This invention provides a high-precision prediction system for ship energy consumption based on multi-model fusion. An objective function is designed to calculate performance evaluation index values for different types of ship energy consumption prediction models. Based on first and second constraints, several high-performing basic models are selected. These high-performing basic models are then fused using a Stacking model to construct a Stacking-based ship energy consumption prediction fusion model. Finally, the fusion model is optimized using Bayesian optimization and adaptive algorithms, thereby improving the prediction accuracy of the ship energy consumption prediction fusion model.
[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A ship energy consumption high-precision prediction system based on multi-model fusion, characterized in that, The system comprises a data collection and analysis module, a single model algorithm ship energy consumption prediction module, a multi-model fusion ship energy consumption prediction module, and a human-computer interaction module, The data collection and analysis module is used for collecting ship energy efficiency data and meteorological data, pre-processing the ship energy efficiency data and meteorological data, performing feature selection on the pre-processed ship energy efficiency data and meteorological data according to a feature selection method to obtain a feature data set, and dividing the feature data set into a training set and a test set according to a certain proportion, The single model algorithm ship energy consumption prediction module is used for constructing different types of ship energy consumption prediction models using the training set and testing using the test set, calculating performance evaluation indexes of different models according to the test results, and selecting a ship energy consumption prediction model set satisfying a constraint condition according to the values of the performance evaluation indexes, wherein the ship energy consumption prediction model set satisfying the constraint condition according to the values of the performance evaluation indexes comprises setting a target function according to the performance evaluation indexes, A first constraint condition is set according to a determination coefficient in the performance evaluation indexes, ship energy consumption prediction models satisfying the first constraint condition are stored in a first filtered model set, the values of the target function in the first filtered model set are calculated respectively and sorted in descending order according to the values of the target function, A second constraint condition is set, the second constraint condition is a model number constraint, and models are selected from the first filtered model set according to the second constraint condition and stored in a ship energy consumption prediction model set. The target function set according to the performance evaluation indexes is a target function set according to formula (1), wherein R 2 , MAPE, RMSE, MAE and MSE are the coefficient of determination, mean absolute percentage error, root mean square error, mean absolute error and mean square error, respectively; w1, w2, w3, w4 and w5 are different weight coefficients and add up to 1; The multi-model fusion ship energy consumption prediction module is used for fusing the ship energy consumption prediction model set based on a Stacking model fusion method, representing the fused model as a fusion model, and performing parameter optimization on the fusion model according to a Bayesian optimization algorithm and a self-adaptive algorithm, and predicting ship energy consumption according to the optimized fusion model, The human-computer interaction module is used for displaying the analysis and processing, running process, and analysis results of the data collection and analysis module, the single model algorithm ship energy consumption prediction module, and the multi-model fusion ship energy consumption prediction module.
2. The ship energy consumption high-precision prediction system based on multi-model fusion according to claim 1, characterized in that, The single model algorithm ship energy consumption prediction module comprises a statistical machine learning-based prediction model unit, a neural network algorithm-based prediction model unit, an integrated learning algorithm-based prediction model unit, and a single model comparative analysis unit, the statistical machine learning-based prediction model unit comprises a multivariate linear regression model-based ship energy consumption prediction model, a support vector regression-based ship energy consumption prediction model, and a K-nearest neighbor algorithm-based ship energy consumption prediction model, the neural network algorithm-based prediction model unit comprises a BP neural network-based ship energy consumption prediction model and a long short-term memory neural network-based ship energy consumption prediction model, the integrated learning algorithm-based prediction model unit comprises a decision tree-based ship energy consumption prediction model, a random forest algorithm-based ship energy consumption prediction model, an extreme gradient boosting algorithm-based ship energy consumption prediction model, an extreme random tree-based ship energy consumption prediction model, and an AdaBoost-based ship energy consumption prediction model, the single model comparative analysis unit is used for testing different types of ship energy consumption prediction models according to a test set and obtaining test results, calculating performance evaluation indexes of different models according to the test results, and selecting a ship energy consumption prediction model set meeting a constraint condition according to a value of the performance evaluation indexes.
3. The ship energy consumption high-precision prediction system based on multi-model fusion according to claim 1, characterized in that, The parameter optimization of the fusion model according to the Bayesian optimization algorithm and the adaptive algorithm comprises, Step S1: obtaining a ship energy consumption prediction model in a ship energy consumption prediction model set and representing the ship energy consumption prediction model as a base model, Step S2: fusing the obtained base model based on a Stacking model fusion method, and representing the fused model as a fusion model, Step S3: adjusting an approximately optimal hyperparameter combination parameter in a parameter domain space of the base model by using the Bayesian optimization algorithm and the adaptive algorithm, Step S4: obtaining the approximately optimal hyperparameter combination of the base model as a parameter of the fusion model to obtain the fusion model optimized based on the Bayesian optimization algorithm and the adaptive algorithm.
4. The ship energy consumption high-precision prediction system based on multi-model fusion according to claim 3, characterized in that, The step S2 comprises constructing a model comprising two layers, a first layer of the model is a base learning layer, and a second layer of the model is a meta learning layer, the base learning layer is used for performing K-fold cross-validation training on ship energy consumption prediction models in a ship energy consumption prediction model set according to feature data sets and generating new data sets, the new data sets are divided into new training sets and new test sets, the meta learning layer is used for constructing a meta model and learning the meta model according to the new data sets, the learning comprises training and testing the meta model according to the new training sets and the new test sets, respectively, and representing the trained and tested meta model as the fusion model.
Citation Information
Patent Citations
Stacking model fusion-based ship motion black box identification modeling and motion prediction method
CN114357872A