Ship speed prediction method and system

Abstract

The invention provides a ship navigation speed prediction method and system, and the method comprises the following steps: collecting historical ship navigation data and historical meteorological data, and constructing a data set; selecting a plurality of machine learning models as base learners and one machine learning model as a meta learner in the algorithm set to form a plurality of stacking models, and training and verifying the plurality of stacking models by adopting the data set to obtain an optimal stacking model; and adopting the optimal stacking model to predict the ship navigational speed. The advantages of various models are integrated through the constructed stacking model, the complex relation between data can be more effectively captured, the prediction precision and reliability are improved, and ship speed prediction is accurately carried out.

Description

Technical Field

[0001] The present invention relates to the field of ship navigation technology, and in particular to a ship speed prediction method and system. Background Art

[0002] The maritime industry is the cornerstone of global trade, with more than 80% of the world's cargo being transported via maritime routes. The efficiency of these maritime operations depends largely on the ability to effectively predict and manage ship speeds. Ship speed prediction is of great significance in areas such as marine transportation, navigation safety, and ship management. Traditional ship speed prediction methods include physical models and empirical models / semi-empirical models. The advantage of physical models is that they have a solid theoretical foundation and can simulate the motion characteristics of ships more accurately. However, physical models require more computing resources and expertise to implement. Empirical models / semi-empirical models involve fewer physical principles than physical models, and the difficulty of building prediction models based on historical data and expert experience is reduced. The application of the above methods cannot take into account the specific characteristics of the ship, which will produce great uncertainty.

[0003] With the rapid development of information technology, monitoring and sensor data related to ship navigation are easy to obtain, which provides new possibilities for speed prediction and promotes the popularity of data-driven methods in the shipping industry. In particular, the application of technologies such as machine learning has become a hot topic in the field of speed prediction. From the perspective of prediction model implementation, data-driven speed prediction models basically do not require physical knowledge related to ships, and are simpler and easier to operate, with high prediction performance and good stability. Existing technologies usually use methods such as Random Forest (RF) and Gaussian Process Regression (GPR) to study the problem of ship speed prediction. However, the method of predicting speed based on machine learning relies on a large amount of data, and there are challenges in data quality and data reliability. In addition, current research uses a single machine learning algorithm to construct a prediction model. When targeting different ship data sets, the prediction accuracy is reduced and the robustness of the model is poor. Summary of the invention

[0004] The present invention proposes a ship speed prediction method and system to solve the technical problems that the existing model needs to rely on a large amount of data for training when predicting ship speed, and the prediction accuracy is reduced when targeting different ship data sets, and the robustness of the model is poor.

[0005] In order to solve the above technical problems, the present invention provides a method for predicting ship speed, comprising the following steps:

[0006] Step S1: Collect historical ship navigation data and historical meteorological data to construct a data set;

[0007] Step S2: Select several machine learning models from the algorithm set as base learners and a machine learning model as a meta-learner to form several stacking models, and use the dataset to train and validate the several stacking models to obtain the optimal stacking model;

[0008] Step S3: Use the optimal stacking model to predict the ship speed.

[0009] Preferably, step S11 includes:

[0010] Step S11: Perform unit data cleaning through adaptive abnormal data identification and interpolation;

[0011] Step S12: Align multi-source data through spatio-temporal attribute matching to perform multi-source data fusion and obtain a dataset.

[0012] Preferably, step S11 includes: converting the ship navigation data into the following matrix A form:

[0013]

[0014] where q tn is the eigenvalue of the nth feature at time t;

[0015] Represent matrix A as A = UΣV through SVD T , where U and V are the left singular matrix and the right singular matrix respectively, and Σ is a diagonal matrix containing singular values σ i ;

[0016] Reconstruct the data matrix by selecting the first k significant singular values Use the reconstructed matrix to fill in the missing or abnormal values

[0017] Preferably, step S12 includes: for a given time point t, use the meteorological data y(t1) and y(t2) at the previous and subsequent time points, where t1 < t < t2, to obtain the meteorological data at time t:

[0018]

[0019] Apply the IDW method to map the meteorological data to the actual navigation position of the ship, then the meteorological data value Z(P) at the target position P is:

[0020]

[0021] where n represents the number of known data points used for interpolation; z i represents the meteorological data value of the ith known observation point; d irepresents the Euclidean distance between the target point P and the i-th known observation point; p is the weight index.

[0022] Preferably, the data set in step S1 is processed by the following method to obtain a high-quality feature data set;

[0023] Assume the heading of the ship is θ, and convert the absolute feature into the relative feature according to the following conversion formula:

[0024]

[0025] in, is the original meteorological characteristic value, is the relative meteorological characteristic value after transformation, θ represents the heading of the ship;

[0026] Let X be the original data set, x be the data point to be transformed, and perform scale unification to obtain the scale-unified data point x ′ The values ​​are:

[0027]

[0028] Where median(X) is the median of data set X, and IQR(X) is the interquartile range of data set X.

[0029] Preferably, in step S2, a base learner is selected from the algorithm set by the following method:

[0030] Step S21: using the data set to train and verify all machine learning models in the algorithm set;

[0031] Step S22: Evaluate the performance of all machine learning models using four indicators: mean square error, root mean square error, average percentage error, and coefficient of determination;

[0032] Step S23: Select the top q machine learning models as base learners.

[0033] Preferably, the expression of the mean square error in step S22 is:

[0034]

[0035] The expression of the root mean square error is:

[0036]

[0037] The expression of the average percentage error is:

[0038]

[0039] The expression of the determination coefficient is:

[0040]

[0041] Where n is the number of samples in the validation set; y i is the actual speed of the ship; is the predicted value of the ship’s speed; It is the average value of the actual speed of the ship.

[0042] Preferably, in step S22, a multi-attribute decision-making method is used to evaluate the performance of the model:

[0043]

[0044] In the formula, S is the comprehensive evaluation index of model prediction performance, I ij represents the value of the prediction performance evaluation index j of model I on the i-th data set, W j is the weight of the predicted performance evaluation index j, and m represents the number of performance evaluation indicators.

[0045] Preferably, step S3 comprises:

[0046] Step S31: Each base learner performs independent modeling based on the input data to generate a set of prediction results;

[0047] Step S32: Integrate the output results of each base learner into a new feature matrix;

[0048] Step S33: Input the new feature matrix into the meta-learner to obtain the ship speed prediction result.

[0049] The present invention also provides a ship speed prediction system, which is implemented based on the above-mentioned ship speed prediction method, and includes a data collection module, a data preprocessing module, a model building module and a ship speed prediction module;

[0050] The data collection module is used to collect historical ship navigation data and historical meteorological data;

[0051] The data preprocessing module is used to perform single-source data cleaning and multi-source data fusion on the historical navigation data of the ship and the historical meteorological data to establish a data set;

[0052] The model building module is used to select several machine learning models as base learners and one machine learning model as a meta learner through an algorithm concentration to build a stacking model;

[0053] The ship speed prediction module is used to process current data using the stacking model to predict the ship speed.

[0054] The beneficial effects of the present invention include at least: the stacked model integrates the advantages of various models by integrating multiple machine learning algorithms, which can more effectively capture the complex relationship between data and improve the accuracy and reliability of prediction; compared with a single model, the stacked model shows higher stability and adaptability when dealing with complex marine environments and ship characteristics, and multi-model integration can effectively reduce the overfitting or underfitting problems that may exist in a single model, and improve the robustness of the model; accurate ship speed prediction can optimize route planning, increase cargo turnover speed, control operating costs, and help evaluate emissions during ship operation and formulate appropriate green shipping strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0056] Figure 2 A schematic diagram of the structure of a stacking model according to an embodiment of the present invention;

[0057] Figure 3 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0059] Example 1

[0060] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting ship speed, comprising the following steps:

[0061] Step S1: Collect historical ship navigation data and historical meteorological data to construct a data set.

[0062] Specifically, the ship's navigation data includes the ship's AIS data and the ship's voyage report data obtained from the shipping company, which specifically includes the ship's MMSI number, ship name, ship length, ship width, ship's maximum design speed, timestamp, longitude, latitude, ground heading, ship speed and other information.

[0063] According to the time and space range of ship navigation data, historical wind field and wave data are obtained from the weather forecast center through an automated program. The wind field data includes two variables, "10mu-component of wind" and "10m v-component of wind", with a data resolution of 0.25° (longitude) × 0.25° (latitude) × 1 hour (time); the wave data includes three variables, "Mean wave direction", "Mean wave period" and "Significant height of combined wind waves and swell", with a data resolution of 0.5° (longitude) × 0.5° (latitude) × 1 hour (time). The ocean current data are obtained from the Physical Ocean Distributed Active Archive Center, including two variables, "Ocean surface zonal currents" and "Ocean surface meridional currents", with a data resolution of 1 / 3° (longitude) × 1 / 3° (latitude) × 1 / 72 year (time).

[0064] Step S2: Select several machine learning models as base learners and one machine learning model as a meta-learner in the algorithm set to form several stacking models, and use the data set to train and verify the several stacking models to obtain the optimal stacking model.

[0065] Specifically, the base learner is combined with different meta-learners, and the optimal stacked ensemble learning model for ship speed prediction is obtained through training and verification.

[0066] Step S3: Use the optimal stacking model to predict the ship speed.

[0067] Specifically, Figure 2 As shown, based on constructing the optimal stacked ensemble learning model, this embodiment is illustrated with six base learners h1, h2, h3, ..., h6. Each base learner performs independent modeling on the input data feature set X to generate a set of prediction results. Where k is the number of the base learner and i is the number of the data feature. The output results of these base learners are integrated into a new feature matrix Y to form a high-level feature representation of the original data:

[0068]

[0069] This new feature matrix Y is then fed into the meta-learner h meta , a higher-level decision generation is performed based on the output of the base learner, and finally a comprehensive prediction result is obtained:

[0070]

[0071] Example 2

[0072] Based on Example 1, in this example, a high-quality feature dataset is obtained through feature engineering, that is, single-source data cleaning and multi-source data fusion are performed.

[0073] Among them, single-source data cleaning includes adaptive outlier data identification and imputation. Specifically, outlier identification is performed through random forests. By constructing a set of multiple decision trees T(x), the outlier probability P(A|x i ) is evaluated, where A represents an outlier event. By calculating the predicted value of each data point and combining the non-linear relationship between features, outliers deviating from the normal pattern are dynamically identified. The SVD method is used to impute outliers and missing values. Specifically, the ship navigation data is converted into the following matrix A form:

[0074]

[0075] q tn is the feature value of the nth feature at time t. The matrix A is represented as A = UΣV T by SVD, where U and V are the left singular matrix and the right singular matrix respectively, and Σ is a diagonal matrix containing singular values σ i . By selecting the first k significant singular values, the data matrix is reconstructed to effectively reduce noise and retain the main features. The imputation process uses the reconstructed matrix to fill in missing or outlier values to ensure that the imputation result not only conforms to the overall data trend but also captures potential structural information.

[0076] Multi-source data fusion includes spatio-temporal attribute matching and alignment of multi-source datasets. To ensure the consistency of different data sources in the time and space dimensions, a spatio-temporal coordination fusion method is adopted. Specifically, first, the linear interpolation method is applied to complete the time attribute consistency. For a given time point t, the meteorological data y(t1) and y(t2) at the previous and subsequent time points are used, where t1 < t < t2, to obtain the meteorological data at time t:

[0077]

[0078] Subsequently, the IDW method is applied to map the meteorological data to the actual navigation position of the ship. Then, the meteorological data value at the target position P is:

[0079]

[0080] In the formula, n represents the number of known data points used for interpolation; z i ​​represents the meteorological data value of the i-th known observation point; d i represents the Euclidean distance between the target point P and the i-th known observation point; p is the weight index. Through this spatiotemporal coordination fusion method, the ship navigation data is effectively integrated with the meteorological data.

[0081] In this embodiment, the historical navigation data of 5 container ships are obtained, and a usable ship navigation data set is obtained through the above data cleaning work.

[0082] Example 3

[0083] Based on Example 1 or Example 2, this embodiment takes into account the particularity of the factors affecting ship navigation and needs to process and transform some features, and uses a two-order feature transformation method for processing.

[0084] Specifically, the first order is relative vector conversion, which converts meteorological data into relative meteorological data closely related to the ship. Assuming the heading of the ship is θ, the absolute feature is converted into relative feature according to the following conversion formula:

[0085]

[0086] in, is the original meteorological characteristic value, is the relative meteorological characteristic value after conversion.

[0087] The second stage is the scale uniform transformation, which uniformly transforms the uneven feature data into a high-quality feature data set. Specifically, let X be the original feature data set and x be the data point to be transformed. The value of the data point after scale uniformization is:

[0088]

[0089] Where median(X) is the median of data set X, and IQR(X) is the interquartile range of data set X.

[0090] Example 4

[0091] This embodiment provides a method for selecting a base learner and a meta learner based on Embodiment 1.

[0092] After obtaining high-quality feature data sets of 5 container ships, each data set was divided into a training set and a validation set in a ratio of 7:3 based on the principle of randomness.

[0093] Determine the performance evaluation indicators of the machine learning model and construct a multi-attribute decision-making method to measure the comprehensive score of the model, including:

[0094] In this embodiment, mean squared error (MSE), root mean squared error (RMSE), mean absolute percentage error (MAPE) and coefficient of determination (R 2 )These four indicators comprehensively evaluate the performance of the ship speed prediction model based on the validation set. MSE measures the average of the squares of the differences between the predicted values ​​and the true values. It imposes heavier penalties on larger errors, thereby reflecting the overall size of the error. The smaller the MSE value, the better the model performance. Obviously, RMSE is the square root of MSE, which provides a balanced assessment of model performance, penalizing both large and small errors. Compared with MSE, RMSE is less sensitive to outliers, but can still effectively capture the overall size of the error. Similarly, the smaller the RMSE value, the better the model performance. MAPE measures the percentage average of the absolute value of the difference between the predicted values ​​and the true values. It provides a relative error measure that makes comparison between different data sets or models of different scales possible. The lower the MAPE value, the better the model performance. R 2 It is used to measure the degree of fit of the model to the data, indicating the ratio of the variability explained by the model to the total variability. Its value ranges from 0 to 1, and the closer it is to 1, the better the model fit.

[0095] The calculation formulas for the above performance evaluation indicators are as follows:

[0096]

[0097] Where n is the number of samples in the validation set; y i is the actual speed of the ship; is the predicted value of the ship’s speed; It is the average value of the actual speed of the ship.

[0098] Exemplarily, this embodiment provides a comprehensive evaluation function of the speed prediction performance of the prediction model based on a plurality of machine learning algorithm performance evaluation indexes on the basis of the four model performance evaluation indicators described above, and uses a multi-attribute decision-making method to construct the following:

[0099]

[0100] In the formula, S is the comprehensive evaluation index of model prediction performance, I ij represents the value of the prediction performance evaluation index j of model I on the i-th data set, W jis the weight of the prediction performance evaluation index j, and m represents the number of performance evaluation indicators. Through this comprehensive evaluation function, we can clearly understand the performance of the machine learning algorithm in the speed prediction task.

[0101] When in use, an algorithm set is established with multiple types of machine learning models, and an algorithm selector constructed with the above-mentioned speed prediction performance comprehensive evaluation function is used to select six machine learning models with better performance from the algorithm set as base learners, including:

[0102] Twelve different machine learning models, including XGBoost, LightGBM, CatBoost, HGBT, ET, RF, DT, GBDT, SVM, KNN, MLR and Ridge, were used as the construction algorithm set. The data set was imported into the 12 machine learning models for training and verification. The test result data was input into the algorithm selector, and the ranking results of the comprehensive evaluation index of speed prediction performance were obtained as shown in Table 1:

[0103] Table 1

[0104]

[0105] According to the comprehensive evaluation index ranking in the table, ET, RF, CatBoost, XGBoost, LightGBM, and HGBT are selected as base learners.

[0106] Based on the multi-level learning mechanism, the base learner is combined with different meta-learners, and the optimal meta-learner is obtained through training and verification. Finally, the optimal stacked ensemble learning model for ship speed prediction is constructed.

[0107] When used, each machine learning algorithm in the algorithm set is used as a meta-learner h meta , use the data set for training and verification, input the test data into the algorithm selector, and again use the speed prediction performance comprehensive evaluation index ranking results as the basis to obtain the optimal meta-learner. Finally, the optimal ship speed prediction stacked ensemble learning model is obtained.

[0108] Specifically, based on the stacking of the six selected base learners, each type of machine learning model in the algorithm set is used as a meta-learner to form 12 different stacking models, namely Stacking-XGBoost, Stacking-LightGBM, Stacking-CatBoost, Stacking-HGBT, Stacking-ET, Stacking-RF, Stacking-DT, Stacking-GBDT, Stacking-SVM, Stacking-KNN, Stacking-MLR and Stacking-Ridge. Subsequently, the data sets input by these 12 stacking models are trained and verified, and the ranking results of their speed prediction performance comprehensive evaluation index are obtained through the speed prediction performance comprehensive evaluation function as shown in Table 2:

[0109] Table 2

[0110]

[0111] According to the sorting results, it is obvious that the comprehensive evaluation index of Stacking-SVM is the highest, indicating that Stacking-SVM has the best comprehensive performance. Therefore, SVM is selected as the best meta-learner in this embodiment.

[0112] Example 5

[0113] Based on the above method, this embodiment provides a ship speed prediction system such as Figure 3 As shown, it includes a data collection module, a data preprocessing module, a model building module and a ship speed prediction module.

[0114] The data collection module is used to collect historical ship navigation data and historical meteorological data.

[0115] Specifically, it is used to obtain ship navigation data from shipping companies through APIs, and further, based on the time and spatial scope of the ship navigation data, obtain meteorological data from institutions such as ECMWF, NCEP and NASA through automated programs.

[0116] The data preprocessing module is used to clean single-source data and fuse multi-source data for historical ship navigation data and historical meteorological data to establish a data set.

[0117] Specifically, it is used to clean single-source data and fuse multi-source data of ship navigation data and historical meteorological and oceanographic data to establish a fused data set.

[0118] Among them, the fused data set can be further processed through the feature engineering module. Based on the two-order feature transformation method, a high-quality feature data set can be constructed through relative vector transformation and scale uniform transformation.

[0119] The model building module is used to select several machine learning models as base learners and one machine learning model as a meta-learner through an algorithm concentration to build a stacked model.

[0120] Specifically, a comprehensive evaluation function of the speed prediction performance of the prediction model is constructed using a multi-attribute decision-making method to form an algorithm selector that meets the actual application scenario. An algorithm set based on multi-type machine learning models is established. Through a multi-level learning mechanism, the optimal base learner and meta-learner are selected according to the training and verification results, and finally the optimal ship speed prediction stacked ensemble learning model is constructed.

[0121] The ship speed prediction module is used to process the current data using a stacking model to predict the ship speed.

[0122] Example 6

[0123] This embodiment provides a device for constructing a ship speed prediction stacking model, which is mainly composed of a processor and a memory, wherein the memory is responsible for storing a computer program, and the program contains an instruction set for guiding the operation of the processor. The processor is not only the core computing and control unit of the device, but also can take a variety of forms, including but not limited to a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), and other programmable logic devices or hardware components. These components enable the device to efficiently execute single or multiple instructions, and are suitable for loading and executing specific instructions to complete the process of constructing a ship speed prediction model.

[0124] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. Only the preferred embodiments of the present invention are expressed. The description is more specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. As long as there is no contradiction in the combination of these technical features, they should be considered as the scope recorded in this specification.

[0125] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for predicting ship speed, characterized in that: It includes the following steps: Step S1: Collect historical ship navigation data and historical meteorological data to construct a data set; Step S2: Select several machine learning models as base learners and one machine learning model as a meta-learner from the algorithm set to form several stacking models, and use the data set to train and validate the several stacking models to obtain the optimal stacking model; Step S3: Use the optimal stacking model to predict the ship speed.

2. A ship speed prediction method according to claim 1, wherein: Step S11 includes: Step S11: Perform unit data cleaning through adaptive abnormal data identification and interpolation; Step S12: Align multi-source data through spatio-temporal attribute matching to perform multi-source data fusion to obtain a data set.

3. A ship speed prediction method according to claim 2, characterized in that: Step S11 includes: Convert the ship navigation data into the following matrix A form: In the formula, q tn is the eigenvalue of the nth feature at time t; By SVD, the matrix A is represented as A = UΣV T , where U and V are left and right singular matrices respectively, and Σ is a diagonal matrix containing singular values ​​σ i ; Reconstruct the data matrix by selecting the first k significant singular values Fill missing or outlier values ​​using the reconstructed matrix 4. A ship speed prediction method according to claim 2, characterized in that: Step S12 includes: For a given time point t, use the meteorological data y(t1) and y(t2) at the previous and subsequent time points, where t1 < t < t2, to obtain the meteorological data at time t: Apply the IDW method to map the meteorological data to the actual navigation position of the ship, then the meteorological data value Z(P) at the target position P is: Where n represents the number of known data points used for interpolation; z i represents the meteorological data value of the i-th known observation point; d i represents the Euclidean distance between the target point P and the i-th known observation point; p is the weight index.

5. A ship speed prediction method according to claim 1, characterized in that: The data set in Step S1 is processed by the following method to obtain a high-quality feature data set; Let the heading of the ship be θ, and convert the absolute features into relative features according to the following conversion formula: in, is the original meteorological characteristic value, is the relative meteorological characteristic value after transformation, θ represents the heading of the ship; Let X be the original data set, x be the data point to be transformed, and perform scale unification to obtain the scale-unified data point x ′ The values ​​are: In the formula, median(X) is the median of the data set X, and IQR(X) is the interquartile range of the data set X.

6. A ship speed prediction method according to claim 1, characterized in that: In Step S2, the base learners are selected from the algorithm set by the following method: Step S21: Use the data set to train and validate all machine learning models in the algorithm set; Step S22: Evaluate the performance of all machine learning models through four indicators: mean squared error, root mean squared error, mean percentage error, and coefficient of determination; Step S23: Select the top q machine learning models with performance as the base learners.

7. A ship speed prediction method according to claim 6, characterized in that: The expression of the mean squared error in Step S22 is: The expression of the root mean squared error is: The expression of the mean percentage error is: The expression of the coefficient of determination is: Where n is the number of samples in the validation set; y i is the actual speed of the ship; is the predicted value of the ship’s speed; It is the average value of the actual speed of the ship.

8. A ship speed prediction method according to claim 6, characterized in that: In Step S22, a multi-attribute decision-making method is used to evaluate the performance of the model: In the formula, S is the comprehensive evaluation index of model prediction performance, I ij represents the value of the prediction performance evaluation index j of model I on the i-th data set, W j is the weight of the predicted performance evaluation index j, and m represents the number of performance evaluation indicators.

9. A ship speed prediction method according to claim 1, characterized in that: Step S3 includes: Step S31: Each base learner independently models according to the input data to generate a set of prediction results; Step S32: Integrate the output results of each base learner into a new feature matrix; Step S33: Input the new feature matrix into the meta-learner to obtain the ship speed prediction result.

10. A ship speed prediction system, implemented based on a ship speed prediction method according to any one of claims 1 to 9, characterized in that: It includes a data collection module, a data preprocessing module, a model construction module, and a ship speed prediction module; The data collection module is used to collect ship historical navigation data and historical meteorological data; The data preprocessing module is used to perform single-source data cleaning and multi-source data fusion on the ship historical navigation data and historical meteorological data to establish a data set; The model construction module is used to select several machine learning models as base learners and one machine learning model as a meta-learner from the algorithm set to construct a stacking model; The ship speed prediction module is used to process current data using the stacking model to predict the ship speed.

Images