Ocean-going ship speed loss prediction method based on hybrid prediction model

Abstract

The application discloses a kind of based on hybrid prediction model's ocean ship speed loss prediction method, including constructing semi-empirical model, according to semi-empirical model, ship power parameter, ocean ship navigation data set is estimated and added to the speed loss of ship in ocean ship navigation data set, based on Z score standardization method, ocean ship navigation data set is pretreated, parameter in ocean ship navigation data set is as feature, the correlation coefficient between each feature is calculated and correlation coefficient matrix is constructed, according to correlation coefficient matrix, obtain the feature set related to speed loss, based on multilayer artificial neural network, construct speed loss hybrid prediction model, obtain training set and train in model, and based on grid search method, the hyperparameter of model is optimized, according to the speed loss hybrid prediction model after optimization, the speed loss of ocean ship is predicted. Based on hybrid prediction model, the prediction accuracy of the speed loss of ocean ship is improved.

Description

Technical Field

[0001] This invention relates to the field of ship navigation technology, and in particular to a method for predicting speed loss of ocean-going vessels based on a hybrid prediction model. Background Technology

[0002] Involuntary speed loss during ship navigation reduces energy efficiency. To improve ship energy efficiency, a reliable method for describing speed loss is typically needed for ship operation. This prediction can be used in business management, such as parts procurement, finished product assembly, and product shipment, effectively reducing delivery delays and inventory backlogs. Therefore, the correct application of "stall" is of great significance for improving ship operating efficiency and increasing corporate financial benefits in areas such as voyage planning, ship position estimation, safe production management, and reducing marine accidents.

[0003] Previous studies have primarily used semi-empirical models (SEM) to estimate speed loss and machine learning models to predict speed loss. However, SEM suffers from poor predictive performance, while machine learning models offer better predictive performance but have limited applicability, requiring large amounts of historical data for modeling and thus are unsuitable for modeling new ships and ships after dry-docking. Summary of the Invention

[0004] This invention provides a method for predicting speed loss of ocean-going vessels based on a hybrid prediction model, in order to overcome the above-mentioned technical problems.

[0005] A method for predicting speed loss of ocean-going vessels based on a hybrid prediction model includes,

[0006] Step 1: Construct a semi-empirical model, which is used to estimate the speed loss of ocean-going vessels during navigation.

[0007] Step 2: Obtain the ocean-going vessel navigation dataset and vessel dynamic parameters. Estimate the vessel's speed loss based on the semi-empirical model, vessel dynamic parameters, and ocean-going vessel navigation dataset. Add the estimated speed loss to the ocean-going vessel navigation dataset. Preprocess the ocean-going vessel navigation dataset using the Z-score normalization method.

[0008] Step 3: Using the parameters in the ocean-going vessel navigation dataset as features, calculate the correlation coefficient between each feature using the Pearson correlation coefficient method, construct a correlation coefficient matrix based on the correlation coefficients between each feature, and obtain the feature set related to speed loss based on the correlation coefficient matrix.

[0009] Step 4: Construct a hybrid prediction model for speed loss based on a multi-layer artificial neural network. Obtain data corresponding to the feature set and speed loss from the preprocessed ocean-going vessel navigation dataset and use it as the training set. Input the training set into the hybrid prediction model for speed loss training. Optimize the hyperparameters of the hybrid prediction model for speed loss using a grid search method to obtain the optimized hybrid prediction model for speed loss.

[0010] Step 5: Predict the speed loss of ocean-going vessels based on the optimized speed loss hybrid prediction model.

[0011] Preferably, the semi-empirical model is constructed using formula (1).

[0012]

[0013] Where: speed loss ΔV is the actual speed of the ship V real With still water velocity V calm The difference between them, P e P represents the effective power of the host. b η is the braking power of the ship's main engine. S For shaft transmission efficiency, η D For the main engine propulsion efficiency, the total resistance R TOTAL =R CALM +R AA +R AW R CALM R AA R AW These are hydrostatic resistance, wind resistance, and wave resistance, respectively.

[0014] Preferably, the feature set related to velocity loss includes five features: ground speed, heading, meaningful wave height, mean wave direction, and mean wave period.

[0015] Preferably, the preprocessing of the ocean-going vessel navigation dataset based on the Z-score standardization method includes obtaining the mean and standard deviation of each parameter, and preprocessing the value of each parameter according to formula (2).

[0016]

[0017] Where X represents the original data. s is the mean and s is the standard deviation.

[0018] Preferably, the optimization of the hyperparameters of the speed loss hybrid prediction model based on the grid search method includes determining the hyperparameters and their value ranges, obtaining different combinations of hyperparameters through exhaustive search, calculating the loss function of the speed loss hybrid prediction model based on the different combinations of hyperparameters, and selecting the optimal combination of hyperparameters as the hyperparameters of the speed loss hybrid prediction model based on the value of the loss function.

[0019] This invention provides a method for predicting speed loss of ocean-going vessels based on a hybrid prediction model. The invention employs a hybrid prediction model based on a semi-empirical model, establishing a semi-empirical model by considering the inherent characteristics of the vessel, such as main engine braking power, shaft transmission efficiency, and main engine propulsion efficiency. Simultaneously, by combining the semi-empirical model with an artificial neural network model, it overcomes the drawback of machine learning models requiring large datasets for training. A feasible prediction model can be trained using only the shipborne automatic measurement data from a single voyage, improving the generalization performance of the hybrid prediction model and making it applicable to new vessels with limited data or vessels whose characteristics have undergone significant changes after major overhauls. Furthermore, a grid search method is used to optimize the hyperparameters of the artificial neural network model, improving the prediction accuracy of the hybrid model. Attached Figure Description

[0020] 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.

[0021] Figure 1 This is a flowchart of the method of the present invention;

[0022] Figure 2 This is a schematic diagram of data fusion according to the present invention;

[0023] Figure 3 This is a heatmap of the correlation coefficient matrix of each variable in the dataset used in this invention;

[0024] Figure 4 This is a diagram of the artificial neural network structure in the hybrid prediction model of this invention;

[0025] Figure 5 This is a scatter plot of the predicted stall value and the actual stall value of the hybrid prediction model of this invention. Detailed Implementation

[0026] 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.

[0027] Figure 1 This is a flowchart of the method of the present invention, as shown below. Figure 1 As shown, the method in this embodiment may include:

[0028] Step 1: Construct a semi-empirical model, which is used to estimate the speed loss of ocean-going vessels during navigation.

[0029] Step 2: Obtain the ocean-going vessel navigation dataset and vessel dynamic parameters. Estimate the vessel's speed loss based on the semi-empirical model, vessel dynamic parameters, and ocean-going vessel navigation dataset. Add the estimated speed loss to the ocean-going vessel navigation dataset. Preprocess the ocean-going vessel navigation dataset using the Z-score normalization method.

[0030] Step 3: Using the parameters in the ocean-going vessel navigation dataset as features, calculate the correlation coefficient between each feature using the Pearson correlation coefficient method, construct a correlation coefficient matrix based on the correlation coefficients between each feature, and obtain the feature set related to speed loss based on the correlation coefficient matrix.

[0031] Step 4: Construct a hybrid prediction model for speed loss based on a multi-layer artificial neural network. Obtain data corresponding to the feature set and speed loss from the preprocessed ocean-going vessel navigation dataset and use it as the training set. Input the training set into the hybrid prediction model for speed loss training. Optimize the hyperparameters of the hybrid prediction model for speed loss using a grid search method to obtain the optimized hybrid prediction model for speed loss.

[0032] Step 5: Predict the speed loss of ocean-going vessels based on the optimized speed loss hybrid prediction model.

[0033] Based on the above scheme, this invention provides a method for predicting speed loss of ocean-going vessels based on a hybrid prediction model. It employs a hybrid prediction model based on a semi-empirical model, establishing a semi-empirical model by considering the inherent characteristics of the vessel, such as main engine braking power, shaft transmission efficiency, and main engine propulsion efficiency. Simultaneously, by combining the semi-empirical model with an artificial neural network model, it overcomes the drawback of machine learning models requiring large datasets for training. A feasible prediction model can be trained using only the shipborne automatic measurement data from a single voyage, improving the generalization performance of the hybrid prediction model and making it applicable to new vessels with limited data or vessels whose characteristics have significantly changed after major overhauls. Furthermore, a grid search method is used to optimize the hyperparameters of the artificial neural network model, improving the prediction accuracy of the hybrid model.

[0034] Specifically, this embodiment provides detailed information on the velocity loss prediction method, including:

[0035] Step 1: Construct a semi-empirical model, which is used to estimate the speed loss of ocean-going vessels during navigation. The semi-empirical model is constructed using formula (1).

[0036]

[0037] Where: speed loss ΔV is the actual speed of the ship V real With still water velocity V calm The difference between them, P e P represents the effective power of the host. b η is the braking power of the ship's main engine. S For shaft transmission efficiency, η D For the main engine propulsion efficiency, the total resistance R TOTAL =R CALM +R AA +R AW R CALM R AA R AW These are hydrostatic resistance, wind resistance, and wave resistance, respectively.

[0038] Step 2: Obtain the ocean-going vessel navigation dataset. The dataset contains variables (or features, fields) such as ship speed, pitch, rudder angle, and meteorological information including wind, waves, and currents. The abbreviations and units of each variable are shown in Table 1. Among them, the relevant variables for ship navigation can be obtained from the deck log and engine log. Meteorological data are obtained by accessing the public dataset of the European Centre for Medium-Range Weather Forecasts (ECMWF) and the real-time ocean surface current analysis (OSCAR) published by the National Centers for Environmental Prediction (NCEP).

[0039] Table 1 Abbreviations and Units for Each Variable

[0040]

[0041] The ship's dynamic parameters, including speed, draft, and main engine power, are obtained. The speed loss of the ship in the ocean-going vessel navigation dataset is estimated using a semi-empirical model, and the estimated speed loss is added to the ocean-going vessel navigation dataset. Specifically, the stall value calculated by the semi-empirical model is added to the dataset as a column based on the "time" and "ship position" data. An example of data fusion is shown below. Figure 2 As shown, its main process is as follows:

[0042] (1) Calculate the meteorological data at the ship's position at that time using the proportional interpolation method based on the meteorological data in the navigation dataset;

[0043] (2) The calculated meteorological data and stall values ​​are aligned row by row into the navigation dataset according to the "time", "longitude" and "latitude" information to form a new dataset for training machine learning models.

[0044] Furthermore, since each feature value in the dataset has a different numerical range, this will lead to a decrease in the accuracy and convergence of the model. To eliminate this adverse effect, the ocean-going vessel navigation dataset is preprocessed based on the Z-score standardization method, that is, each column of data in the dataset is standardized according to the Z-score calculation formula. The preprocessing of the ocean-going vessel navigation dataset based on the Z-score standardization method includes obtaining the mean and standard deviation of each parameter, and preprocessing the value of each parameter according to formula (2).

[0045]

[0046] Where X represents the original data. s is the mean and s is the standard deviation.

[0047] Step 3: Use the parameters in the ocean-going vessel navigation dataset as features, calculate the correlation coefficient between each feature based on the Pearson correlation coefficient method, and construct a correlation coefficient matrix based on the correlation coefficient between each feature. Obtain the feature set related to speed loss based on the correlation coefficient matrix. The feature set related to speed loss includes five features: ground speed, heading, meaningful wave height, mean wave direction, and mean wave period.

[0048] Specifically, we use features (parameters or variables) from the dataset as input. To reveal the relationship between stall and various navigation parameters, we employ correlation analysis to calculate the correlation coefficient matrix of each feature in the dataset. Correlation analysis studies whether there is a linear correlation between two sets of variables; therefore, it assumes a linear correlation between the variables, and the results describe the degree of linear correlation. This invention calculates the Pearson correlation coefficients between the variables to construct a correlation coefficient matrix, thereby identifying several variables highly correlated with ship stall. Simultaneously, high correlation between dependent variables can lead to multicollinearity. Based on maritime practice, this invention selects variables that may cause multicollinearity, retaining only one representative variable. The specific analysis is as follows:

[0049] We found that the features significantly associated with speed loss are (the values ​​in parentheses are correlation coefficients, close to 1 indicates positive correlation, and close to -1 indicates negative correlation) ground speed (-0.97), heading (-0.65), significant wave height (0.72), swell height (0.76), mean wave period (0.59), mean wave direction (0.62), mean swell period (0.54), mean wind and wave direction (0.62), and mean swell direction (0.56). Among these, ground speed and heading showed a high negative correlation.

[0050] Furthermore, we observed a strong positive correlation between the mean swell direction and the mean swell period (0.94), mean wave direction (0.97), mean wave period (0.9), swell height (0.77), and significant wave height (0.77), and a negative correlation with ground speed (-0.57) and heading (-0.71). Other features also exhibited multicollinearity. Therefore, to avoid multicollinearity, we selected ground speed, heading, significant wave height, mean wave direction, and mean wave period as input variables after feature selection, based on our nautical practice. The correlation coefficient matrix of each variable is shown below. Figure 3 As shown.

[0051] Step 4: Construct a hybrid prediction model for speed loss based on a multi-layer artificial neural network. Obtain data corresponding to the feature set and speed loss from the preprocessed ocean-going vessel navigation dataset and use it as the training set. Input the training set into the hybrid prediction model for speed loss training, and optimize the hyperparameters of the model using a grid search method.

[0052] The optimization of hyperparameters for the speed loss hybrid prediction model based on the grid search method includes determining the hyperparameters and their value ranges, as shown in Table 2. Different combinations of hyperparameters are obtained through exhaustive search. The loss function of the speed loss hybrid prediction model is calculated based on these combinations. The optimal hyperparameter combination is selected as the hyperparameters for the speed loss hybrid prediction model based on the values ​​of the loss function. Specifically, the hyperparameter optimization method based on the grid search method is used to optimize the hyperparameters of the ANN model, thereby achieving higher accuracy and wider applicability in speed loss prediction.

[0053] Table 2 Hyperparameters and their ranges

[0054]

[0055] Obtain the optimized hybrid prediction model for speed loss.

[0056] Specifically, in this embodiment, the basic process of a neural network can be represented by the following formula, and the neural network model structure used in this embodiment is as follows: Figure 4 As shown,

[0057]

[0058] Where w ij Let x be the weights of the i-th layer and neuron j. i Let b be the input value of the i-th layer. i Here, u represents the bias weights of the i-th layer, n is the total number of connected neurons, and u represents the bias weights of the i-th layer. i This refers to the output of a unit or neuron in a neural network model.

[0059] Generally speaking, the layers of a neural network are fully connected, meaning that any neuron in the nth layer is connected to any neuron in the (n+1)th layer. Although ANNs appear large and complex overall, from a microscopic local model perspective, they are essentially a linear relationship plus an activation function. Mathematically, the activation function maps input data to 0 to 1 (where tanh maps to -1 to +1). The reason for this mapping, besides regularizing the data, is likely to control the data, keeping it within a certain range. There are also other detailed functions; for example, the Sigmoid function focuses on small changes in the data before and after the center point, ignoring changes at extreme points. Typically, Sigmoid is used in fully connected layers, while ReLU is used in convolutional layers. This invention uses three hidden layers with full connections, therefore the Sigmoid activation function is chosen, and its calculation formula is as follows:

[0060]

[0061] Here, x represents the input variable. Analysis of the function expression and the Sigmoid curve reveals that the Sigmoid activation function has the advantage of a smooth curve that is differentiable everywhere, thus it is often used for regression and fitting problems.

[0062] Step 5: Predict the speed loss of ocean-going vessels based on the optimized speed loss hybrid prediction model.

[0063] In this embodiment, the preliminary stall prediction value is calculated by a semi-empirical model (SEM) and used as a new feature to construct a new dataset. Then, the input of the hybrid prediction model (HPM) consists of five variables or features obtained from the feature selection process: ground velocity, heading, meaningful wave height, mean wave direction, and mean wave period, plus the stall prediction value calculated by SEM.

[0064] For the new dataset, the model parameters were reselected. During the parameter selection phase, a grid search method was used to optimize the model's hyperparameters within their range of values. The grid search method is essentially an exhaustive search; within the range of hyperparameter values, it iterates through all possibilities, trying each one, and the best-performing parameter combination is the final hyperparameter combination. After hyperparameter tuning, the model's prediction accuracy improved to some extent, and the accuracy of the hybrid prediction model was as follows: Figure 5 As shown.

[0065] Overall beneficial effects:

[0066] This invention provides a method for predicting speed loss of ocean-going vessels based on a hybrid prediction model. The invention employs a hybrid prediction model based on a semi-empirical model, establishing a semi-empirical model by considering the inherent characteristics of the vessel, such as main engine braking power, shaft transmission efficiency, and main engine propulsion efficiency. Simultaneously, by combining the semi-empirical model with an artificial neural network model, it overcomes the drawback of machine learning models requiring large datasets for training. A feasible prediction model can be trained using only the shipborne automatic measurement data from a single voyage, improving the generalization performance of the hybrid prediction model and making it applicable to new vessels with limited data or vessels whose characteristics have undergone significant changes after major overhauls. Furthermore, a grid search method is used to optimize the hyperparameters of the artificial neural network model, improving the prediction accuracy of the hybrid model.

[0067] 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 method for predicting speed loss of ocean-going vessels based on a hybrid prediction model, characterized in that, include, Step 1: Construct a semi-empirical model, which is used to estimate the speed loss of ocean-going vessels during navigation. The semi-empirical model is constructed using formula (1). (1) Among them: velocity loss The actual speed of the ship real With still water speed calm The difference between them This refers to the effective power of the host unit. For the main engine braking power of the ship, For shaft transmission efficiency. For main engine propulsion efficiency, total drag , R CALM , R AA , R AW These are hydrostatic resistance, wind resistance, and wave resistance, respectively; the feature set related to velocity loss includes five features: ground speed, heading, significant wave height, mean wave direction, and mean wave period. Step 2: Obtain the ocean-going vessel navigation dataset and vessel dynamic parameters. Estimate the vessel's speed loss based on the semi-empirical model, vessel dynamic parameters, and ocean-going vessel navigation dataset. Add the estimated speed loss to the ocean-going vessel navigation dataset. Preprocess the ocean-going vessel navigation dataset using the Z-score normalization method. Step 3: Using the parameters in the ocean-going vessel navigation dataset as features, calculate the correlation coefficient between each feature using the Pearson correlation coefficient method, construct a correlation coefficient matrix based on the correlation coefficients between each feature, and obtain the feature set related to speed loss based on the correlation coefficient matrix. Step 4: Construct a hybrid prediction model for speed loss based on a multi-layer artificial neural network. Obtain data corresponding to the feature set and speed loss from the preprocessed ocean-going vessel navigation dataset and use it as the training set. Input the training set into the hybrid prediction model for speed loss and train it. Optimize the hyperparameters of the hybrid prediction model for speed loss using a grid search method to obtain the optimized hybrid prediction model for speed loss. Calculate the initial stall prediction value using a semi-empirical model. The stall prediction value is used as a new feature to construct a new dataset. The input to the hybrid prediction model consists of five variables or features obtained from the feature selection process: ground speed, heading, meaningful wave height, mean wave direction, and mean wave period, plus the stall prediction value calculated by the semi-empirical model. Step 5: Predict the speed loss of ocean-going vessels based on the optimized speed loss hybrid prediction model.

2. The method for predicting speed loss of ocean-going vessels based on a hybrid prediction model according to claim 1, characterized in that, The preprocessing of the ocean-going vessel navigation dataset based on the Z-score standardization method includes obtaining the mean and standard deviation of each parameter, and preprocessing the value of each parameter according to formula (2). (2) in, X The original data, The average is... s The standard deviation is denoted as .

3. The method for predicting speed loss of ocean-going vessels based on a hybrid prediction model according to claim 1, characterized in that, The optimization of the hyperparameters of the speed loss hybrid prediction model based on the grid search method includes determining the hyperparameters and their value ranges, obtaining different combinations of hyperparameters through exhaustive search, calculating the loss function of the speed loss hybrid prediction model based on the different combinations of hyperparameters, and selecting the optimal combination of hyperparameters as the hyperparameters of the speed loss hybrid prediction model based on the value of the loss function.

Images