Dynamic ship speed prediction method based on latent invariant feature transformation

Abstract

The application provides a dynamic ship speed prediction method based on latent invariant feature conversion. The method comprises the following steps: obtaining data and preprocessing the data to obtain a data set; performing ship-water speed prediction at the t'+1th to t'+kth moment, which comprises the following steps: performing data decomposition on the training set, constructing a model according to the decomposed data, inputting the parameters at the t'+1th moment into the model to obtain the ship-water speed prediction value at the t'+1th moment, updating the ship-water speed prediction value at the t'+1th moment to the training set, performing data decomposition on the updated training set and iterating the model, and inputting the parameters at the t'+a+1th moment into the model for the ath iteration; and after k-1 iterations, the ship-water speed prediction values at the t'+1th to t'+kth moment are obtained. The technical scheme of the application solves the problem of short dynamic ship speed prediction period in the prior art.

Description

Technical Field

[0001] This invention relates to the field of ship speed prediction technology, and more particularly to a dynamic ship speed prediction method based on potential invariant feature transformation. Background Technology

[0002] Traditional methods for ship speed prediction have limitations. For example, physics-based and data-driven model-based methods often fail to deliver accurate predictions because key features such as the real-time power of the ship's engines cannot be obtained in advance. Furthermore, existing dynamic prediction models struggle to capture unpredictable speed trends, hindering accurate long-term speed forecasts. More importantly, significant differences in loading conditions and navigation areas lead to varying speed distributions across different voyages, failing to meet the model's fundamental assumption of consistent distribution. This negatively impacts the accuracy and generalization ability of the prediction model. Consequently, existing prediction models generally have short prediction cycles. Summary of the Invention

[0003] To address the aforementioned technical problem of short dynamic prediction cycles for ship speed, this invention provides a dynamic ship speed prediction method based on latent invariant feature transformation. The invention primarily utilizes data decomposition of the training set, constructs a model based on the decomposed data, inputs parameters at time t'+1 into the model to obtain the predicted ship speed relative to the water at time t'+1, updates the training set with this predicted speed, and iterates the updated training set. After k-1 iterations, the predicted ship speed relative to the water from time t'+1 to time t'+k is obtained, thus achieving dynamic ship speed prediction.

[0004] The technical means employed in this invention are as follows:

[0005] On the one hand, the present invention provides a dynamic ship speed prediction method based on latent invariant feature transformation, comprising:

[0006] Data is acquired and preprocessed to obtain a dataset, which includes a training set containing ship speeds relative to water from time 1 to time t'.

[0007] Predicting the ship's speed relative to the water from time t'+1 to t'+k includes:

[0008] The training set is decomposed into data, and a model is constructed based on the decomposed data. The parameters at time t'+1 are input into the model to obtain the predicted value of the ship's speed relative to the water at time t'+1. The predicted value of the ship's speed relative to the water at time t'+1 is updated to the training set. The updated training set is decomposed into data and iterated in the model. For the a-th iteration, the parameters at time t'+a+1 are input, where a is a positive integer.

[0009] After k-1 iterations, the predicted ship speed relative to water is obtained from time t'+1 to time t'+k.

[0010] Further, the data is preprocessed to obtain a dataset, including:

[0011] The data set is obtained by extracting the ship schedule and ship geographical location information from the data, deleting data on ships docking in port, data on ships sailing distances less than a preset range, and data with missing or abnormal features.

[0012] Furthermore, the data is preprocessed to obtain a dataset, which also includes:

[0013] Obtain the ship's land speed, ocean current land speed, and the angle between the ship's heading and the ocean current direction at time i, and calculate the ship's water speed at time i.

[0014] Calculate the absolute error between the ship's speed relative to the water at time i and the average value of the dataset. Calculate a complementary error function based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, delete the ship data at time i.

[0015] Furthermore, the calculation of the ship's velocity relative to the water at time i is performed as follows:

[0016]

[0017] Among them, V stw Let V be the ship's velocity relative to the water at time i. sog Let V be the ship's ground speed at time i. current Let be the ocean current velocity relative to the ground at time i. Let be the angle between the ship's heading and the direction of the ocean current at the i-th moment.

[0018] Further, the calculation of the absolute error between the ship's speed relative to the water at time i and the average value of the dataset involves calculating a complementary error function based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, the ship data at time i is deleted. The calculation is performed as follows:

[0019]

[0020] Where erfc is the complementary error function, Δi is the absolute error between the ship's speed relative to the water at time i and the average value of the dataset, σ is the standard deviation of the dataset, and 0.2 is the preset value.

[0021] Further, the step of decomposing the training set into data, constructing a model based on the decomposed data, and inputting the parameters at time t'+1 into the model to obtain the predicted ship speed relative to the water at time t'+1 includes:

[0022] The training set is decomposed into q components, and q prediction models are established for each of the q components. Each model is the sum of the q prediction models. The parameters at time t'+1 are input into each of the q prediction models, and each of the q prediction models obtains q prediction values. The q prediction values ​​are added together to obtain the predicted value of the ship's speed relative to the water at time t'+1.

[0023] Further, establishing q prediction models corresponding to the q components includes:

[0024] The q components include a first component and a second component. A global attention long short-term memory neural network model is constructed for the first component, and a temporal random forest model is constructed for the second component.

[0025] Further, the step of inputting the parameters at time t'+1 into the q prediction models includes:

[0026] The global attention long short-term memory neural network model is input with the sea conditions at time t'+1, the draft at time t'+1, and the residual component difference value of the ship's speed relative to water from time t'-n+1 to time t', where n is the preset input historical time series length;

[0027] The sea conditions at time t'+1, the draft at time t'+1, and the intrinsic mode function values ​​of the ship's speed relative to water from time t'-n+1 to time t' are input into the temporal random forest model.

[0028] Furthermore, the dataset also includes a test set, which includes the ship's speed relative to the water from time t'+1 to time t'+k, used to determine the accuracy of the predicted ship speed relative to the water at time t'+k.

[0029] On the other hand, the present invention also provides a storage medium comprising a stored program, wherein, when the program is executed, the above-described dynamic ship speed prediction method based on latent invariant feature transformation is performed.

[0030] Compared with the prior art, the present invention has the following advantages:

[0031] 1. The dynamic ship speed prediction method based on latent invariant feature transformation provided by this invention decomposes the training set into data and constructs a model based on the decomposed data. It can extract latent invariant features from the training set and use the latent invariant features to construct a model for predicting ship speed relative to water. It has higher accuracy and stronger robustness, and also has a longer prediction period.

[0032] 2. The dynamic ship speed prediction method based on potential invariant feature transformation provided by this invention avoids the direct use of real-time monitoring data from ship engines. The historical ship speed data and weather and sea state data used can be easily obtained through shipborne automatic identification systems or midday reports and marine weather forecasts, respectively. The requirements for shipborne monitoring equipment are not high, and it has high accuracy. It can predict the ship's speed over a long period of time in advance, thereby providing data support for subsequent navigation, collision avoidance and other operations. It can also be used for post-experimental analysis.

[0033] In summary, the technical solution of this invention offers advantages such as easy data acquisition, high accuracy, and the ability to dynamically predict ship speed in water. Therefore, the technical solution of this invention solves the problem of short dynamic prediction cycles for ship speed in existing technologies.

[0034] Based on the above reasons, this invention can be widely applied in fields such as ship speed prediction. Attached Figure Description

[0035] 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.

[0036] Figure 1 This is a flowchart of a dynamic ship speed prediction method based on latent invariant feature transformation provided by the present invention.

[0037] Figure 2 Another flowchart of the dynamic ship speed prediction method based on latent invariant feature transformation provided by the present invention. Detailed Implementation

[0038] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0039] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0040] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a dynamic ship speed prediction method based on latent invariant feature transformation provided by the present invention, demonstrating a specific embodiment of the dynamic ship speed prediction method based on latent invariant feature transformation provided by the present invention, including:

[0041] S101: Acquire data and preprocess the data to obtain a dataset, which includes a training set, which includes the ship's water velocity from time 1 to time t'.

[0042] S102: Predict the ship's speed relative to the water from time t′+1 to time t′+k, including:

[0043] S1021: Decompose the training set into data, construct a model based on the decomposed data, input the parameters at time t′+1 into the model, obtain the predicted value of the ship's speed relative to the water at time t′+1, update the predicted value of the ship's speed relative to the water at time t′+1 into the training set, decompose the updated training set into data and iterate the model. For the a-th iteration, input the parameters at time t′+a+1, where a is a positive integer.

[0044] S1022: After k-1 iterations, the predicted values ​​of the ship's speed relative to the water are obtained from time t'+1 to time t'+k.

[0045] Understandably, in step S101, the data includes dynamic navigation information of the ship and corresponding meteorological and sea state characteristics. The dynamic navigation information includes time, ship's latitude and longitude coordinates, heading, ship's speed relative to land, and ship's draft. The meteorological and sea state characteristics include wind direction, wind speed, wave direction, wave height, swell direction, swell height, current direction, and current speed. This data can be easily obtained through the ship's automatic identification system and ship midday reports, requiring minimal requirements for the ship's monitoring equipment and possessing high accuracy. It allows for the prediction of the ship's speed over a longer period, providing data support for subsequent navigation, collision avoidance, and other operations, or for post-event experimental analysis. In step S1021, after updating the predicted ship speed relative to water at time t'+1 to the training set, the updated training set is decomposed and substituted into the model. The parameters at time t'+2 are then input into the model to obtain the predicted ship speed relative to water at time t'+2, completing one iteration. The number of components in the training set after each data decomposition is the same. In step S1022, k-1 is a positive integer and k-1≥a. After k-1 iterations, the predicted ship speed relative to water is output from time t'+1 to time t'+k. By decomposing the training set, a model is constructed based on the decomposed data. In this process, latent invariant features can be extracted from the training set, and a model can be built using these latent invariant features to predict ship speed relative to water, resulting in higher accuracy, stronger robustness, and a longer prediction period.

[0046] In some alternative embodiments, refer to Figure 2 , Figure 2 Another flowchart of the dynamic ship speed prediction method based on latent invariant feature transformation provided by the present invention includes preprocessing the data to obtain a dataset, including:

[0047] The data set is obtained by extracting the ship schedule and ship geographical location information from the data, deleting data on ships docking in port, data on ships sailing distances less than a preset range, and data with missing features and anomalies.

[0048] It is understandable that deleting data where the ship's sailing distance is less than the preset range means deleting data with shorter sailing distances, as well as data on ships docking in port, retaining only data on transoceanic voyages. On this basis, data with missing features and anomalies are deleted to obtain a cleaner dataset, thereby ensuring the accuracy of ship speed predictions based on this dataset.

[0049] In some alternative embodiments, reference continues to be made to... Figure 2 The dataset is obtained by preprocessing the data, and also includes:

[0050] Obtain the ship's velocity relative to the ground, the ocean current's velocity relative to the ground, and the angle between the ship's heading and the ocean current direction at time i. Calculate the ship's velocity relative to the water at time i. Calculate the absolute error between the ship's velocity relative to the water at time i and the average value of the dataset. Calculate the complementary error function based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, delete the ship data at time i.

[0051] Understandably, the data includes multiple voyages of the ship. In order to eliminate the influence of ocean currents on the ship, the ship's land speed is first converted into the ship's water speed. Then, the ship's water speed for each voyage is analyzed, and a complementary error function is calculated for the ship's water speed. If the result of the complementary error function calculation is less than the preset value, the ship's water speed will be regarded as an outlier and deleted.

[0052] In some alternative embodiments, the ship's velocity relative to the water at time i is calculated as follows:

[0053]

[0054] Among them, V stw Let V be the ship's velocity relative to the water at time i. sog Let V be the ship's velocity relative to the ground at time i. current Let be the ocean current velocity relative to the ground at time i. Let be the angle between the ship's bow and the direction of the ocean current at time i.

[0055] In some optional embodiments, the absolute error between the ship's speed relative to the water at time i and the average value of the dataset is calculated. A complementary error function is then calculated based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, the ship data at time i is deleted. The calculation is performed as follows:

[0056]

[0057] Where erfc is the complementary error function, Δi is the absolute error between the ship's speed relative to the water at time i and the average value of the dataset, σ is the standard deviation of the dataset, and 0.2 is the preset value.

[0058] In some alternative embodiments, reference continues to be made to... Figure 2 The training set is decomposed, and a model is constructed based on the decomposed data. The parameters at time t′+1 are input into the model to obtain the predicted value of the ship's speed relative to the water at time t′+1, including:

[0059] The dataset is decomposed into q components, and q prediction models are established for each of the q components. The model is the sum of the q prediction models. The parameters at time t′+1 are input into the q prediction models respectively, and the q prediction models obtain q prediction values ​​respectively. The q prediction values ​​are added together to obtain the predicted value of the ship's speed relative to the water at time t′+1.

[0060] Understandably, the training set includes the ship's velocity relative to the water from time 1 to time t', i.e., v 1：t’ The training set is decomposed into q components, i.e. For each of the q components, q prediction models are established. The parameters at time t'+1 are input into each of the q prediction models, resulting in q predicted values ​​for each model. The sum of these q predicted values ​​yields the predicted ship velocity relative to the water at time t'+1, i.e., p. t’+j j=1; the predicted ship speed relative to water at time t'+1 is updated to the training set, the updated training set is decomposed and substituted into the model, and the parameters at time t'+2 are input into the model to obtain p. t’+j j = 2, ..., until p is obtained. t′+j , j=m, and perform iterative prediction in units of m until the predicted value of the ship's speed relative to the water at time t′+k is obtained.

[0061] In some alternative embodiments, reference continues to be made to... Figure 2 For each of the q components, establish q prediction models, including:

[0062] The q components include the first component and the second component. A global attention long short-term memory neural network model is constructed for the first component, and a temporal random forest model is constructed for the second component.

[0063] It is understandable that the first component is the residual (RES), the second component is the intrinsic mode function (IMF), and n is the preset historical time series length of the input. The time series length refers to the time series length of the ship's velocity relative to the water input to the model. By decomposing a certain time series, it can be found that the image corresponding to RES is approximately monotonic, and its velocity is in a relatively large range relative to the IMF. Therefore, the velocity with the largest proportion and the most stable change among the components of the ship's velocity relative to the water can be extracted, namely the velocity generated by the engine. Due to the different geographical locations and external environments of the ship in different voyages, the engine state varies greatly in different voyages. Therefore, to achieve knowledge transfer between different voyages, it is necessary to align the features in the engine state. Since the engine state generally does not change instantaneously, the difference between adjacent time points is minimal. Therefore, the potential invariant features are contained in the velocity difference between adjacent time points, and the velocity distribution between different voyages can be aligned by using invariant features. On the other hand, in addition to the large differences in velocity distribution, the velocity change trends also differ between different voyages. However, there is a certain similarity in the long-term change trend of RES between different voyages. Therefore, we utilize a global attention network to handle these invariant features implicit in longer time series. This approach maximizes the alignment of features across different voyages, enabling domain adaptation between them. In summary, for the RES component, we construct a global attention long short-term memory neural network incorporating a difference mechanism.

[0064] For the IMF components, firstly, we remove IMF components with low Hearst coefficients; here, we assume the first c IMFs are removed. This eliminates unpredictable random fluctuations in ship speed. Next, while the distribution range of IMF components doesn't differ significantly across voyages, the speed differences between adjacent time points are large. Therefore, differencing cannot further reduce the distribution differences between the test and training sets. Furthermore, the external environment faced by ships across different voyages typically lacks time-series trend characteristics. Therefore, for modeling speed changes in IMFs due to weather and sea conditions, we employ a semi-dynamic prediction model that better handles data with unclear time-series characteristics compared to a purely dynamic model. In summary, we construct independent temporal random forest models for each IMF component.

[0065] In some alternative embodiments, reference continues to be made to... Figure 2 The parameters at time t'+ are input into the q prediction models respectively, including:

[0066] Input the sea conditions at time t'+1, the draft at time t'+1, and the residual component difference values ​​of the ship's water speed from time t'-n+1 to time t' into the global attention long short-term memory neural network model;

[0067] Input the sea conditions at time t'+1, the draft at time t′+1, and the intrinsic mode function values ​​of the ship's water velocity from time t'-n+1 to time t' into the temporal random forest model.

[0068] It is understandable that the sea conditions affecting speed are defined as M. t The draft is defined as U. t The actual and predicted values ​​of water velocity are defined as v, respectively. t and P t Use superscripts I and R to denote the components after velocity decomposition, for example... Where I represents the intrinsic mode function and R represents the residual; the length of the input historical time series is defined as n. When predicting the ship's velocity relative to the water at time t′+1, M′ is input into the global attention long short-term memory neural network model. t+1 、U′ t+1 , Output Input M' to the temporal random forest model t+1 、U' t+1 , Output h represents the h-th second component, and the q-th component includes one first component and q-1 second components. The predicted ship speed relative to the water at time t'+1 is... and The sum of.

[0069] In some alternative embodiments, reference continues to be made to... Figure 2 The dataset also includes a test set, which includes the ship's water speed from time t'+1 to time t'+k, used to determine the accuracy of the predicted ship water speed from time t'+1 to time t'+k.

[0070] During model training, to adjust the hyperparameters involved, a standard 5-fold cross-validation experiment is established. The entire dataset is divided into 5 subsets. For each fold, 3 subsets are used as training data, and the remaining two subsets are used as validation and test data, respectively. Then, a grid search is used to search the hyperparameter space. The search process and model performance are evaluated using three widely used validation metrics: mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 ).

[0071] This invention was experimentally compared with the Integrated Moving Average Autoregressive Moving Average (ARIMA), Extended Kalman Filter (EKF), Long Short-Term Memory (LSTM), Temporal Random Forest (SRF), and Self-Attention Network (XFMR) models on the same dataset, as shown in Table 1. This invention, regardless of single-step or multi-step prediction, showed lower mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 All three error metrics are superior to the other five models.

[0072] Table 1. Comparison of results for single-step and multi-step prediction

[0073]

[0074] Based on the same inventive concept, the present invention also provides a storage medium including a stored program, wherein, when the program is executed, the dynamic ship speed prediction method based on latent invariant feature transformation as described in any of the above embodiments is performed.

[0075] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0076] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0078] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0080] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0081] 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 dynamic ship speed prediction based on latent invariant feature transformation, characterized in that, Comprise: Obtaining data and pre-processing the data to obtain a data set, the data set including a training set, the training set including ship speed against water from time 1 to time n; and time n. A method for predicting a ship's speed over water at a future time t is provided, comprising: +1 to a ship's speed over water at a future time t, comprising: The training set is decomposed into data, a model is constructed based on the decomposed data, and the first data is input into the model. The parameters at time +1 are used to obtain the first... The predicted ship speed relative to the water at time +1, the first... The predicted ship speed relative to water at time +1 is updated to the training set. The updated training set is then decomposed and iterated in the model. For the a-th iteration, the input is... The parameter at time +a+1, where a is a positive integer; The training set is divided into q components, and q prediction models are established corresponding to the q components, the q components include a first component and a second component, a global attention long short-term memory neural network model is constructed for the first component, and a time series random forest model is constructed for the second component; The model is a sum of the q prediction models, respectively inputting the first moment parameter into the global attention long short-term memory neural network model; inputting the first moment sea condition, the first moment draft and the first moment residual component difference value of the ship-to-water speed from the first moment to the nth moment into the time series random forest model, n being a preset input history time length; inputting the first moment sea condition, the first moment draft and the first moment intrinsic mode function value of the ship-to-water speed from the first moment to the nth moment into the time series random forest model. The q prediction models respectively obtain q predicted values, and the q predicted values are added to obtain the first The ship-to-water speed prediction value at the time of +1. after k-1 iterations, the first +1 to the kth moment of the ship-to-water speed prediction.

2. The dynamic ship speed prediction method based on latent invariant feature transformation according to claim 1, characterized in that, The data is preprocessed to obtain a data set, comprising: Extracting the ship schedule and the ship geographic location information from the data, deleting the ship port stay data, deleting the data with a ship sailing distance less than a preset range, and deleting the data with missing and abnormal features to obtain the data set. 3.The dynamic ship speed prediction method based on latent invariant feature transformation of claim 1, wherein, The data is preprocessed to obtain a data set, and the preprocessing further comprises: Get the Calculate the ship's land speed, the ocean current's land speed, and the angle between the ship's heading and the ocean current direction at time t. The ship's speed relative to the water at any given moment; Calculate the first The absolute error between the ship's speed relative to the water at a given time and the average value of the dataset is used to calculate a complementary error function based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, the ship is deleted. Ship data at any given time.

4. The dynamic ship speed prediction method based on latent invariant feature transformation according to claim 3, characterized in that, The calculating the first moment of the ship-to-water speed is calculated in the following way: ； wherein, the first time instant, the first time instant, the first time instant, the first time instant.

5. The dynamic ship speed prediction method based on latent invariant feature transformation according to claim 3, characterized in that, The calculation of the first The absolute error between the ship's speed relative to the water at a given time and the average value of the dataset is used to calculate a complementary error function based on the absolute error and the standard deviation of the dataset. If the result of the complementary error function calculation is less than a preset value, the ship is deleted. Ship data at any given time is calculated as follows: 0.2； wherein, is the complementary error function, is the average value of the data set, is the absolute error of the ship speed over water at the time instant from the average value of the data set, is the standard deviation of the data set, and 0.2 is the preset value. 6.The dynamic ship speed prediction method based on latent invariant feature transformation of claim 1, wherein, The data set also includes a test set comprising the first +1 time to the kth +1 time to the kth +1 time to the kth +1 time to the kth 7. A storage medium, characterized by The storage medium comprises a stored program, wherein when the program runs, the program executes the dynamic ship speed prediction method based on latent invariant feature transformation in any one of claims 1 to 6.

Images