Ship maneuvering motion prediction method and system based on local gaussian process regression

Abstract

The application discloses a ship maneuvering motion prediction method and system based on local Gaussian process regression, performs a clustering analysis algorithm on pre-collected ship motion data, divides the ship motion data into a plurality of clusters, and determines a cluster center corresponding to each cluster; the distances between a to-be-predicted sample and each cluster center are calculated, the nearest cluster is obtained, and a prediction result of the to-be-predicted sample is calculated on the nearest cluster. Based on the local data idea of clustering analysis, the ship motion data set is automatically divided into a plurality of clusters according to similarity, the prediction result of the to-be-predicted sample is calculated on the nearest cluster, and calculation on the entire data set is not necessary, so that the prediction can be accelerated and a certain prediction accuracy can be ensured. Compared with the traditional GPR method, the application has higher calculation efficiency and the loss of accuracy is not obvious.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of ship maneuvering motion prediction, and particularly relates to a ship maneuvering motion prediction method and system based on local Gaussian process regression. BACKGROUND

[0002] The prediction of ship maneuvering motion is the basis for studying ship maneuverability and developing ship motion controllers. At present, the common method for predicting ship maneuvering motion is to establish a ship dynamics model, and then to calculate the motion response of the ship within a period of time in a continuous iterative manner.

[0003] With the development of artificial intelligence technology, ship maneuvering motion prediction methods combined with machine learning technology have been widely applied. In these methods, the Gaussian Process Regression (GPR) algorithm has shown good performance in the prediction of ship maneuvering motion because of its strong interpretability, good performance in processing small sample and noisy data sets, and the ability to quantify the uncertainty of the prediction results. These methods include the multi-input multi-output GPR method, the GPR method considering input noise, the GPR method based on genetic algorithm optimization, the GPR method considering ship propeller torque, and the GPR method based on hybrid kernel function.

[0004] However, as a new method, the ship maneuvering motion prediction technology based on the GPR method has not been fully studied in terms of computational efficiency. In the prediction process, the time complexity of the GPR method for calculating the mean and variance of a single prediction sample is O(n) and O(n 2), where n is the sample size of the training dataset. Considering that the prediction of ship maneuvering motion is a continuous iterative process, if n is large or the prediction duration is long, the time-consuming calculation of the GPR method in the whole prediction process will be large, which will limit the practical engineering application value of the GPR method, especially in the design of the model-based controller of the intelligent ship. For the classic GPR method, the task with thousands of training samples is difficult to handle. However, due to the strong nonlinearity of ship maneuvering motion, a large number of training samples are usually required to identify a robust and accurate ship dynamics model by using a non-parametric modeling method. At present, there are few available technologies for improving the prediction of ship maneuvering motion based on the GPR method, and most of them use the inducing point method: experts use a sparse GPR method based on data similarity to predict the three-degree-of-freedom maneuvering motion of the KVLCC2 tanker; experts propose a new non-parametric online identification method, which combines input noise with the completely independent training condition approximation algorithm and is applied to the prediction of ship maneuvering motion. Although this method can improve the calculation efficiency of the GPR method to a certain extent, the selection of the inducing set is difficult, and if it is not properly selected, the problem of non-convex optimization may be encountered. Therefore, improving the calculation efficiency of the GPR method in predicting ship maneuvering motion is a problem to be solved. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a ship maneuvering motion prediction method and system based on local Gaussian process regression, which solves the problems of high calculation complexity and low efficiency in predicting ship maneuvering motion in the prior art.

[0006] The technical problems to be solved by the present application are solved by the following technical solutions:

[0007] The ship maneuvering motion prediction method based on local Gaussian process regression performs a clustering analysis algorithm on the pre-acquired ship motion dataset, divides it into a plurality of clusters, and determines the cluster center corresponding to each cluster; calculates the distance between the to-be-predicted sample and each cluster center, obtains the nearest cluster, and calculates the prediction result of the to-be-predicted sample on the nearest cluster.

[0008] The specific steps of performing a clustering analysis algorithm on the pre-acquired ship motion data are as follows:

[0009] Step 1, collect the ship motion data acquired during the self-propelled model test or the actual ship test, including the longitudinal oscillation velocity u, the transverse oscillation velocity v, the yaw angle velocity r and the rudder angle δ;

[0010] Step 2, apply the ship motion data acquired during the self-propelled model test or the actual ship test to construct a dataset where xi For the input of the i-th sample, input x i The specific mathematical structure is x i =[u i ,v i ,r i ,δ i ];y i For the output of the i-th sample, the mathematical meaning of the output in the three motion directions of the ship (pitch, sway, and yaw) is respectively pitch acceleration udot, sway acceleration vdot, and yaw acceleration rdot; i The matrix X is composed of (i = 1, 2, ..., n), i.e., X = [x1, x2, ..., xn]. n ] T ;y i The matrix consisting of (i = 1, 2, ..., n) is denoted as y, i.e.: y = [y1, y2, ..., yn]. n ] T ;

[0011] Step 3: For the n samples x in X i The dataset consisting of (i = 1, 2, ..., n) is standardized, and several cluster centers are randomly initialized.

[0012] Step 4: Calculate x for each sample in the dataset. i (i = 1, 2, ..., n) and the Euclidean distance between the cluster centers, and assign each sample to the cluster with the closest distance;

[0013] Step 5: Recalculate the center of each cluster and determine if convergence has occurred. If convergence has occurred, output the clustering results; otherwise, return to step 4.

[0014] For the sample x to be predicted * f(x) * The mean m(f(x)) * This can be expressed as:

[0015]

[0016] in, k f Let K be the covariance function, and K be the covariance matrix. Let I be the variance of the observed noise. n Let be the identity matrix of dimension n.

[0017] x * The prediction results are mainly composed of x from the training dataset. * The selection is based on samples with high similarity.

[0018] The sample x to be predicted is calculated using the following formula.* forecast result:

[0019]

[0020]

[0021] wherein m(f(x * )) is the mean of f(x * ); cov(f(x * )) is the covariance of f(x * ); X s is the distance of the clustering center to x * the nearest clustering, y s = [y s1 , y s2 ... y sm ] T (m < n) is the corresponding observation value vector of X s , and I m is a unit matrix with the dimension of m.

[0022] The calculation load of the mean and variance of f(x * ) is O(m) and O(m 2 ), respectively.

[0023] The ship maneuvering motion forecasting model comprises a data processing module, a motion forecasting module, and a forecasting result output display module; wherein the data processing module applies the method to perform clustering analysis on the pre-collected ship motion data, obtains a plurality of clusters, and calculates the distance of the to-be-forecasted sample to each clustering center; the motion forecasting module is used to calculate the forecasting result of the to-be-forecasted sample on the nearest clustering; and the forecasting result output display module is used to display the forecasting result of the motion variable in the ship maneuvering motion process.

[0024] The forecasting result display includes but is not limited to a curve comparison chart mode and a table mode.

[0025] The forecasting result includes but is not limited to the ship surge speed, the ship sway speed, the bow yaw angle speed, and the time consumed for forecasting.

[0026] The forecasting result includes a curve comparison chart of the measured value, the forecasting value by the traditional method, and the forecasting value by the model.

[0027] Compared with the prior art, the present application has the following beneficial effects:

[0028] 1. Based on the local data idea of clustering analysis, the ship motion data set is automatically divided into a plurality of clusters according to the similarity, and the forecasting result of the to-be-forecasted sample is calculated on the clustering X s (X s is the clustering center distance to x* The local Gaussian process regression method can not only accelerate the prediction, but also ensure the prediction accuracy without calculating on the whole data set X.

[0029] 2. The ship maneuvering motion prediction result based on the measured data shows that the technical scheme has higher calculation efficiency and unobvious accuracy loss compared with the traditional GPR method. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The local Gaussian process regression method flow chart of the application.

[0031] Figure 2 The k-means clustering analysis algorithm execution flow chart of the application.

[0032] Figure 3 The ship maneuvering motion data set clustering analysis result schematic view of the application.

[0033] Figure 4 The surge velocity prediction result schematic view of the 20° / 10° Z-shaped maneuvering motion of the application.

[0034] Figure 5 The sway velocity prediction result schematic view of the 20° / 10° Z-shaped maneuvering motion of the application.

[0035] Figure 6 The yaw angle velocity prediction result schematic view of the 20° / 10° Z-shaped maneuvering motion of the application. DETAILED DESCRIPTION

[0036] In order to enable the personnel in the technical field to better understand the scheme of the present application, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application.

[0037] Based on the embodiments in the present application, all other embodiments obtained by the personnel in the field without making creative labor should belong to the protection scope of the present application.

[0038] The ship maneuvering motion prediction method based on the local Gaussian process regression comprises the following steps: performing a clustering analysis algorithm on the pre-acquired ship maneuvering data, dividing the ship maneuvering data into a plurality of clusters, and determining a cluster center corresponding to each cluster; calculating the distance between a to-be-predicted sample and each cluster center, obtaining a cluster closest to the to-be-predicted sample; and calculating the prediction result of the to-be-predicted sample on the cluster closest to the to-be-predicted sample.

[0039] The specific embodiments are as shown in the following. Figures 1 to 6

[0040] ​The ship maneuvering motion prediction method based on local Gaussian process regression comprises the following steps:

[0041] Step 1, collect ship motion data collected from self-propelled model test or real ship test, including four signals of surge velocity u, sway velocity v, yaw angle velocity r and rudder angle δ. The collected data is constructed into a data set Wherein x i is the input of the i-th sample, the specific mathematical structure of the input x i is x i =[u i ,v i ,r i ,δ i ]; y i is the output of the i-th sample, and the mathematical meaning of the output is the longitudinal acceleration udot, the sway acceleration vdot and the yaw acceleration rdot in the three motion directions of the ship (longitudinal motion, sway motion and yaw motion). x i The matrix composed of x n (i=1, 2...n) is X, that is, X=[x1, x2...x T ] i The matrix composed of y n (i=1, 2...n) is y, that is: y=[y1, y2...y T ] n .

[0042] Step 2, execute k-means clustering analysis algorithm, for input matrix X=[x1, x2,...,x T , divide the n samples in X into k clusters according to the similarity based on Euclidean distance:

[0043] X→[X1,X2,...,X s ,...,X k ] T

[0044] size(X1)+size(X2)+...size(X s )+...size(X k )=n (1)

[0045] In formula (1), X1, X2,...,X s ,...,X k are k clusters divided from X=[x1, x2,...,x n ] T . The execution process of the clustering analysis algorithm is as follows:

[0046] Step 2.1 Standardize the dataset and randomly initialize several cluster centers;

[0047] Step 2.2 Calculate the Euclidean distance between each sample in the dataset and the cluster centers, and divide each sample to the cluster closest to it;

[0048] Step 2.3 Recalculate the center of each cluster, and determine whether to converge, if convergent, output the clustering result; otherwise, return to execute step 2.2.

[0049] Step 3, in the process of predicting the ship maneuvering motion, for the sample to be predicted x * , the prediction result based on the Gaussian process regression algorithm is calculated on the cluster X s = [x s1 , x s2 ,..., x sm ] T (m < n) (X s is the cluster closest to x * in all clusters) without having to calculate on the entire dataset X, and the basis for doing so is analyzed as follows:

[0050] For the sample to be predicted x * , the mean value m(f(x * )) of f(x * ) can be expressed as:

[0051]

[0052] In equation (2), k f is the covariance function, K is the covariance matrix, is the variance of the observation noise, and I n is the unit matrix with dimension n.

[0053] Equation (2) shows that the prediction result of the sample to be predicted x * mainly depends on the coefficients a and the outputs of the n covariance functions k f (x i , x * ). For most covariance functions k f , if the distance between x i and x * is larger (i.e., the similarity between x i and x * is lower), the output of the covariance function will be closer to 0, so according to equation (2), the prediction result of x * is mainly determined by the samples in the training dataset that are similar to x * .

[0054] Step 4: Based on step 3, calculate x. * The forecast results. In the traditional Gaussian process regression method, the forecast results are calculated as follows:

[0055]

[0056]

[0057] In equation (3), m(f(x) * f(x) * The mean of ); in equation (4), cov(f(x) * f(x) * Covariance. The matrices in equations (3) and (4) are based on the entire dataset. Established when the matrix After the inversion operation is completed and the result is stored, the computational loadings for the mean and variance for each sample to be predicted are O(n) and O(n) respectively. 2 If the dataset size n is large or the duration of the ship maneuvering motion to be predicted is long, the prediction calculation process will be very time-consuming. According to step 3, the sample x to be predicted... * The forecast results can be displayed in cluster X s =[x s1 x s2 , ..., x sm ] T Calculations are performed on (m < n) (X) s For all clusters, the distance x between the cluster centers * The most recent clustering can be used instead of computing on the entire dataset X, which not only speeds up the forecast but also ensures a certain level of forecast accuracy. At this point, f(x) * The formulas for calculating the mean and variance of () are:

[0058]

[0059]

[0060] In equation (5), y s =[y s1 y s2 , ..., y sm ] T (m < n) is X s The corresponding observation vector, I m Let f(x) be an identity matrix of dimension m. From equations (5) and (6), we can see that f(x) is equal to f(x) in this case. * The calculated loads for the mean and variance of ) are O(m) and O(m) respectively. 2 ).

[0061] Based on the above steps, the algorithm execution flow of the present application is as shown in Figure 1 .

[0062] Case implementation: the three-degree-of-freedom maneuvering motion of KVLCC2 tanker is predicted, and the collected ship motion data set includes 15° / 5°, 20° / 5°, 30° / 5° and 35° / 5° Z-shaped test, a total of 1274 sample points. The sampling period of the samples in the data set is 0.05s.

[0063] The k-means clustering analysis algorithm is executed. In the k-means clustering analysis algorithm, the cluster number k is 3, and the clustering analysis result is as shown in Figure 3 and Table 1.

[0064] Table 1 shows the sample size of each cluster, KVLCC2

[0065] Cluster A Cluster B Cluster C Total number Sample size 471 420 383 1274

[0066] After the clustering analysis is completed, the 20° / 10° Z-shaped maneuvering motion not involved in the training set is predicted based on formula (5) and formula (6), and the prediction result is as shown in Figures 4 to 6 , and the calculation time consumed in the prediction process is as shown in Table 2. Among them, “measured value” is the result collected from the self-propelled model test, “GPR” is the prediction result using the traditional GPR method, and “the present application” is the prediction result using the technical solution in the present application. From the results given in Figures 4 to 6 , it can be seen that the prediction result of the present application is in good agreement with the measured value as a whole, and is similar to the prediction result of the traditional GPR method, which indicates that the calculation of the prediction result based on clustering does not bring too much accuracy loss; from the results given in Table 2, it can be seen that the time consumed in the prediction process of the present application is much less than that of the traditional GPR method, indicating that the present application has very high calculation efficiency for the prediction of ship three-degree-of-freedom maneuvering motion.

[0067] Table 2 Time consumed in 20° / 10° Z-shaped maneuvering motion prediction

[0068] Conventional GPR method The present invention Time / second consumed for prediction 9.059 1.789

[0069] The ship maneuvering motion prediction model comprises a data processing module, a motion prediction module, and a prediction result output display module; wherein the data processing module applies the method to cluster analyze the pre-collected ship motion data, obtains a plurality of clusters, and calculates the distance from the to-be-predicted sample to each cluster center; the motion prediction module is used to calculate the prediction result of the to-be-predicted sample on the nearest cluster; and the prediction result output display module is used to display the prediction result.

[0070] The prediction result display includes but is not limited to a curve comparison graph mode and a table mode.

[0071] The prediction result includes but is not limited to ship surge speed, sway speed, yaw angle speed, time consumed for prediction.

[0072] The prediction result includes a curve comparison chart of measured value, traditional method prediction value and the model prediction value.

[0073] Those skilled in the art should understand that the skilled in the art can realize the variants in combination with the prior art and the above-mentioned embodiments, and such variants do not affect the essential content of the scheme, and are not described here.

[0074] It should be understood that the present scheme is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in the ordinary way in the art; any person skilled in the art can make many possible changes and modifications to the present scheme or modify it into equivalent embodiments without departing from the scope of the present scheme, which does not affect the essential content of the present scheme. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present scheme without departing from the content of the present scheme still belongs to the protection scope of the present scheme.

Claims

1. A method for ship maneuvering motion prediction based on local Gaussian process regression, characterized in that: The pre-collected ship motion dataset is subjected to a clustering analysis algorithm to divide it into several clusters, and the cluster center of each cluster is determined. The distance between the sample to be predicted and each cluster center is calculated, and the cluster with the closest distance is obtained. The prediction result of the sample to be predicted is calculated on the cluster with the closest distance. The specific steps of performing the clustering analysis algorithm on the pre-collected ship motion data are as follows: Step 1: Collect ship motion data obtained during model aircraft testing or full-scale ship testing, including pitch and roll speed. sway speed Bow roll rate and rudder angle These four signals; Step 2: Construct a dataset using the ship motion data collected during model aircraft tests or full-scale ship tests. ,in For the first Input of each sample, input The specific mathematical structure is ; For the first The output of each sample, in terms of its mathematical meaning in the three motion directions of the ship—pitch, sway, and yaw—represents pitch acceleration, yaw acceleration, and yaw acceleration, respectively. sway acceleration and bow roll acceleration ; The matrix formed is ,Right now ; The matrix formed is denoted as ,Right now: ; Step 3, for middle Sample The dataset is standardized and several cluster centers are randomly initialized. Step 4: Calculate the value of each sample in the dataset. Calculate the Euclidean distance from the cluster center and assign each sample to the cluster with the closest distance. Step 5: Recalculate the center of each cluster and determine if convergence has occurred. If convergence has occurred, output the clustering results. Otherwise, return to step 4.

2. The ship maneuvering motion prediction method based on local Gaussian process regression according to claim 1, characterized in that: For the sample to be predicted , mean This can be expressed as: in, , Let covariance function be used. Let covariance matrix be the variance matrix. The variance of the observed noise. For dimension is The identity matrix.

3. The ship maneuvering motion prediction method based on local Gaussian process regression according to claim 2, characterized in that: The prediction results are mainly composed of the training dataset and The selection is based on samples with high similarity.

4. The ship maneuvering motion prediction method based on local Gaussian process regression according to claim 3, characterized in that: The sample to be predicted is calculated using the following formula. Forecast results: in, for The mean; for covariance; Distance between cluster centers Recent clustering, for The corresponding observation vector, For dimension is The identity matrix.

5. The ship maneuvering motion prediction method based on local Gaussian process regression according to claim 4, characterized in that: The calculation loads for the mean and variance are respectively and .

6. A ship maneuvering motion prediction model, characterized in that: The system includes a data processing module, a motion prediction module, and a prediction result output display module. The data processing module applies the method described in any one of claims 1 to 5 to perform cluster analysis on the pre-collected ship motion data, obtains several clusters, and calculates the distance from the sample to be predicted to the center of each cluster. The motion prediction module is used to calculate the prediction result for the sample to be predicted in the cluster with the closest distance. The prediction result output display module is used to display the prediction result.

7. The ship maneuvering motion prediction model according to claim 6, characterized in that: The forecast results are displayed in modes including, but not limited to, curve comparison charts and tables.

8. The ship maneuvering motion prediction model according to claim 7, characterized in that: The forecast results include, but are not limited to, the ship's pitch speed, sway speed, bow roll rate, and the time taken for the forecast.

9. The ship maneuvering motion prediction model according to claim 8, characterized in that: The forecast results include measured values, forecast values ​​using traditional methods, and curve comparison graphs.

Images