Ship motion intelligent forecasting method and system integrated with system state estimation

Abstract

The invention belongs to the technical field of ship and ocean engineering, and discloses a ship motion intelligent forecasting method and system integrated with system state estimation, and the method comprises the steps: segmenting the time calendar data of the motion of a certain degree of freedom of a ship and the corresponding wave time calendar data, and obtaining a training set and a test set of time calendar forecasting; the optimal lag order # imgabs0 # is obtained through wave-ship motion system lag order analysis, a training set is divided through a sliding window according to the optimal lag order # imgabs1 #, and an ANN forecasting model is constructed according to the obtained data set; and obtaining a ship motion forecasting result according to the constructed ANN forecasting model. According to the method, wave-ship motion lag order calculation is integrated into an intelligent model ANN pretreatment method from the time sequence analysis theory, the mapping relation of the ANN model for capturing wave excitation and motion response is improved, great help is provided for the model to learn the mapping relation between the wave time calendar and the motion time calendar, and the method can be applied to the intelligent model ANN pretreatment method. And the intelligent forecasting precision of the ship motion is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship and ocean engineering, and in particular relates to a method and system for intelligent prediction of ship motion integrated with system state estimation. Background Art

[0002] The six-degree-of-freedom swaying motion of the ship is constantly generated due to environmental factors such as waves, wind, and currents. In medium and high sea conditions, this swaying motion will exceed the allowable range of marine operations, which has a significant impact on operational efficiency and navigation safety. Ship motion prediction refers to estimating the changes in the ship's navigation attitude in the future based on the ship's historical motion status and wave environment information. This is of great significance for ensuring the safety of various shipborne operations, improving the navigation safety of ships, and improving the efficiency of marine operations.

[0003] The swaying motion of a ship is caused by the excitation of the waves on the hull. There is a time lag between the ship motion response and the wave excitation. The lag time is called the system response time. Affected by sensors and other acquisition equipment and complex wave environments, the response time of the collected ship motion data and the wave data in the corresponding period has time-varying characteristics. The existing ship motion prediction method based on wave input does not consider the unstable characteristics of wave excitation and ship motion response time. It is difficult for the model to stably capture the mapping relationship between wave excitation and motion response, resulting in reduced model prediction accuracy.

[0004] In terms of ship motion prediction based on wave input, the solutions close to the present invention include Chapter 5 of the doctoral dissertation "Research on Phase-Analytical Wave Propagation and Ship Motion Stable Period Prediction Based on Deep Learning", the patent "System and Method for Wave Sensing and Ship Motion Prediction Using Operation Period Indicator" and "A Ship Motion Prediction Method Based on AR-OLSSVR Model"

[0005] The difference between this invention and the "Study on Phase-Analytical Wave Propagation and Ship Motion Stable Period Prediction Based on Deep Learning" is that:

[0006] 1. The present invention can firstly perform wave forecasting to obtain wave time history data for a period of time in the future, and then use the predicted future time history data as input to obtain ship motion time history data; it can also directly input the wave time history data of the current time to directly obtain the future ship motion time history data.

[0007] 2. The present invention analyzes the temporal correlation between the ship motion data and the wave data of the corresponding time period and performs cross-correlation analysis to obtain the optimal lag order. Based on the optimal lag order, the time domain correspondence between the ship time history segment and the wave time history segment is corrected to improve the forecast accuracy.

[0008] The present invention is different from the "System and method for wave sensing and ship motion prediction using operation period indicator" in that:

[0009] 1. The present invention predicts future ship motion time history data based on wave time history data, but the similar patent predicts future ship motion time history data based on radar images, and the input information of the two is different.

[0010] 2. The present invention innovates the ship motion prediction method based on wave input, specifically, innovates the algorithm model and provides a specific prediction principle, while the similar patent tends to propose a software system, focusing on functional design, and does not provide a specific prediction principle.

[0011] The difference between the present invention and "A ship motion prediction method based on AR-OLSSVR model" is that:

[0012] 1. The present invention predicts future ship motion history data based on wave history data, which essentially builds a model to learn the mapping relationship between wave information and ship motion information, and belongs to the modeling of mutual correlation between two sequences; however, the similar patent uses the historical ship motion data as input to extrapolate the ship motion history, which essentially builds a model to learn the relationship between historical ship motion and future ship motion, and belongs to the modeling of single sequence autocorrelation.

[0013] 2. The forecasting models used are different. The present invention improves the ANN model and is essentially implemented based on the neural network model; while the similar patent improves the autoregressive (AR) model and uses the classical time series analysis method.

[0014] In summary, the ship motion prediction method based on wave input is mainly based on the deep learning model, which realizes the prediction of the future motion of the ship by learning the time-dependent characteristics contained in the historical motion data of the ship and the wave environment information. However, the time-varying characteristics of the response time of the ship motion data and the wave data in the corresponding period are not taken into account, which makes the model unable to accurately capture the mapping relationship between wave excitation and motion response, resulting in poor model prediction effect.

[0015] Through the above analysis, the problems and defects of the existing technology are as follows: the existing ship motion prediction method does not take into account the time-varying characteristics of the response time of the ship motion data and the wave data in the corresponding time period, which makes the model unable to accurately capture the mapping relationship between wave excitation and motion response, and thus leads to poor model prediction effect. Summary of the invention

[0016] In order to overcome the problems existing in the related art, the disclosed embodiment of the present invention provides a method for intelligent prediction of ship motion integrating system state estimation, and the technical solution is as follows:

[0017] The present invention is implemented as follows: a method for intelligent prediction of ship motion incorporating system state estimation, comprising:

[0018] S1, segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain the training set and test set of time history forecast;

[0019] S2, through the hysteresis order analysis of the wave-ship motion system, the optimal hysteresis order is obtained ;

[0020] S3, based on the best lag order obtained Use sliding windows to divide the training set;

[0021] S4, constructing an ANN prediction model based on the obtained data set;

[0022] S5, obtaining the ship motion prediction result according to the constructed ANN prediction model.

[0023] In step S1, a training set and a test set of calendar forecast are obtained, including:

[0024] The known time history data of a certain degree of freedom motion of a ship D=[D1,d2…d T ], split into training set D tr =[d1,d2…d n ] and the test set D te =[d n+1 ,d n+2 …d T ], where d i Represents the data at each time point;

[0025] The data quantity of the training set and the test set is set according to actual needs, and the division of the wave history data is the same as that of the motion data.

[0026] In step S2, the optimal lag order is obtained include:

[0027] The formula for calculating the autocovariance coefficient of each series lag is as follows:

[0028] γ xx (k) = E[(x t -μ x )(x t+k -μ x )]=E[(x t -μ x )(x t-k -μ x )]

[0029] In the formula, γ xx (k) is xt The autocovariance of the sequence, E[] is the mathematical expectation, x t is the specific value of the sequence at time t, x t+k is the specific value of the sequence at time t+k, μ x is the mean;

[0030] The estimated value c of the cross-covariance between two series with lag order k xy (k) is obtained from the following formula:

[0031]

[0032] In the formula, c xy (k) is the estimated value of the cross-covariance, k is the lag order, n is the sequence length, y t+k ,y t-k are the specific values ​​of the sequence at time t+k and tk, respectively, t is the specific value at time t, x t ,y t The sample mean of ;

[0033] The formula for calculating the mutual correlation coefficient of two series with lag order k is:

[0034]

[0035] In the formula, ρ xy (k) is the cross-correlation function, σ x is the standard deviation of the x series, σ y is the standard deviation of the y series, γ xy (k) is the cross-covariance coefficient between sequence x and sequence y, s x ,s y is the standard deviation of the two series;

[0036] For the wave data W used for model training, W = [w1, w2…w t ] and the ship's certain degree of freedom motion data E = [e1, e2…e t ], using sliding windows for cutting, constructing the data into input and output matrices, the results are as follows:

[0037]

[0038] Where W in is the wave data set, e in is the motion dataset, w t is the wave data at a certain moment, e t is the motion data at a certain moment, i is the length of the segment segmentation of the model sliding window, and the input matrix and the output matrix are matched row by row to obtain several groups of input and output segment combinations;

[0039] For each input-output fragment combination (W in [j],D out [j]), and obtain the mutual correlation coefficient γ(k) of lag order k; the optimal lag order of the segment is obtained according to the statistics of the mutual correlation coefficients at each order Then we have:

[0040]

[0041] In the formula, is the optimal lag order function, γ() is the cross-covariance function, and f is the search range of order k.

[0042] In step S3, according to the obtained optimal lag order Use sliding windows to divide the training set, including:

[0043] The sliding window fragments are divided twice based on the optimal lag order, and the corrected input and output matrices are obtained as follows:

[0044]

[0045] In the formula, are the wave and motion matrices after the data is divided twice, i is the length of the segment of the model sliding window, and w * is the wave data at time *, D * is the motion data at time *, a∈N * It is half of the cropping length of the secondary segmentation, and the actual input and output length is i-2a. Finally, the corrected input and output matrix is ​​used for ANN training.

[0046] In step S4, an ANN prediction model is constructed according to the obtained data set, including:

[0047] An ANN prediction model is established, in which the number of input layer nodes is back, the number of hidden layer nodes is ahead, the number of output layer nodes is 1, and the activation function is the ReLU function. The expression is:

[0048]

[0049] In the formula, x is the independent variable of the function.

[0050] Furthermore, the number of neurons in the ANN prediction model constructed by the dataset is adjusted according to the actual situation, and the input is the wave time history data.

[0051] In step S5, the ship motion prediction result is obtained according to the constructed ANN prediction model, including:

[0052] The input is wave time history data, and the output is ship motion time history data; the training data set is X tr The input model is calculated and the calculation result is compared with Y in the training data set. tr Calculate the loss function, where X tr is the training set sample data, Y tr is the label value of the training set;

[0053] The back propagation algorithm is used to correct various parameters in the model according to the calculation results of the loss function, and the optimization is continuously iterated.

[0054] Furthermore, the loss function is calculated by selecting mean square error, root mean square error, mean absolute error, and mean absolute percentage error.

[0055] Furthermore, the optimal lag order of the current motion history and the corresponding historical wave history is used Optimal lag order between the motion history of the surrogate target ship and the corresponding wave history Based on the optimal lag order Carry out adaptive correction, input the corrected data into the constructed ANN prediction model, and obtain the ship motion prediction results.

[0056] Another object of the present invention is to provide a ship motion intelligent prediction system integrated with system state estimation, the system is used to control the ship motion intelligent prediction method integrated with system state estimation, the system comprises:

[0057] The data segmentation module is used to segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain a training set and a test set for time history forecasting;

[0058] The optimal hysteresis order calculation module is used to obtain the optimal hysteresis order through the hysteresis order analysis of the wave-ship motion system

[0059] The training set partitioning module is used to obtain the best lag order Use sliding windows to divide the training set;

[0060] ANN prediction model building module, used to build an ANN prediction model based on the obtained data set;

[0061] The ship motion prediction module is used to obtain the ship motion prediction results according to the constructed ANN prediction model.

[0062] In combination with all the above technical solutions, the beneficial effects of the present invention are as follows: starting from the theory of time series analysis, the present invention integrates the calculation of the lag order of wave-ship motion into the pre-processing method of the intelligent model ANN, so that the ANN model can calculate the system response duration of the input wave-motion time history segment and obtain the lag value based on the cross-correlation function analysis, and obtain the optimal lag order according to each group of segment combinations. Then the model adaptively corrects each wave-motion time history according to the selected optimal lag order, so that the mapping relationship between wave excitation and motion response basically corresponds.

[0063] The existing technology predicts ship motion data based on wave input, but does not consider the time lag relationship between two time history data. The model cannot accurately learn the mapping relationship between wave excitation and motion response, resulting in reduced performance of the model in motion prediction based on wave input and poor prediction effect. To this end, the present invention proposes an intelligent prediction method for ship motion that incorporates system state estimation, and uses the cross-correlation method to analyze the lag order of the wave-ship motion system to obtain the optimal lag order. Then the model uses a sliding window to divide the data so that the mapping relationship between wave excitation and motion response basically corresponds, thereby improving the ANN model's ability to capture the mapping relationship between wave excitation and motion response and improving the prediction accuracy.

[0064] The ship motion intelligent prediction method integrated with system state estimation of the present invention starts from the time series analysis theory, integrates the wave-ship motion lag order calculation into the neural network pre-processing method, so that the model can realize adaptive estimation according to each group of wave-motion time history fragment combination to obtain the lag time, and then correct the time domain correspondence between the ship time history fragment and the wave time history fragment according to the lag time, thereby improving the prediction accuracy. The present invention takes into account the time lag relationship between wave excitation and ship motion response, provides great help for the model to learn the mapping relationship between wave time history and motion time history, and greatly improves the prediction accuracy.

[0065] Since the duration of the excitation response is dynamically changing, the present invention takes into account the time lag relationship between wave excitation and motion response and integrates it into the model, so that the model can capture the mapping relationship between wave excitation and motion response, thereby improving the prediction accuracy. The present invention analyzes the temporal correlation between the ship motion data and the wave data of the corresponding period and performs cross-correlation analysis to obtain the optimal lag order. Based on the optimal lag order, the time domain correspondence between the ship time history segment and the wave time history segment is corrected, which solves the problem that the previous model is difficult to stably capture the mapping relationship between wave excitation and motion response, and effectively improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure;

[0067] Figure 1 is a flow chart of a method for intelligent prediction of ship motion incorporating system state estimation provided by an embodiment of the present invention;

[0068] Figure 2 is a schematic diagram of a method for intelligent prediction of ship motion incorporating system state estimation provided by an embodiment of the present invention;

[0069] Figure 3 This is a schematic diagram of the time history data of the fifth-level lower limit ship pitch motion provided by an embodiment of the present invention;

[0070] Figure 4 This is a schematic diagram of wave time history data of level 5 lower limit sea conditions provided by an embodiment of the present invention;

[0071] Figure 5 It is a comparison chart of the pitch motion forecast for the fifth lower limit sea state provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0073] Different from the previous ship motion prediction model based on wave input, the present invention analyzes the time lag relationship between wave excitation and ship motion response, calculates the system response time of the input wave-motion time history segment and obtains the lag value based on the cross-correlation function analysis, and obtains the optimal lag order according to each group of segment combinations; the model adaptively corrects each wave-motion time history according to the selected optimal lag order, so that the mapping relationship between wave excitation and motion response basically corresponds, so that the ANN prediction model can accurately capture the mapping relationship between wave time history and ship motion time history, thereby improving the accuracy of the results.

[0074] Embodiment 1, as Figure 1 and Figure 2 As shown, the method for intelligent prediction of ship motion integrating system state estimation provided by an embodiment of the present invention specifically includes the following steps:

[0075] S1, segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain the training set and test set of time history forecast;

[0076] The specific method is as follows: Taking the ship motion history data as an example: the known ship motion history data of a certain degree of freedom D = [d1, d2…d T ], split into training set D tr =[d1,d2…d n ] and the test set D te =[d n+1 ,d n+2 …d T ], where d i Represents the data at each time point;

[0077] The data quantity of the training set and the test set is set according to actual needs, and the division of the wave history data is the same as that of the motion data.

[0078] S2, through the hysteresis order analysis of the wave-ship motion system, the optimal hysteresis order is obtained

[0079] The specific method is: A stable Gaussian random process can be expressed by its mean μ and autocovariance function γ k To describe, or equivalently, by its mean μ, variance σ 2 and the autocorrelation function ρ k to describe.

[0080] The formula for calculating the autocovariance coefficient of each series lag is as follows:

[0081] γ cc (k) = E[(x t -μ x )(x t+k -μ x )]=E[(x t -μ x )(x t-k -μ x )]

[0082] In the formula, γ xx (k) is x t The autocovariance of the sequence, E[] is the mathematical expectation, x t is the specific value of the sequence at time t, x t+k is the specific value of the sequence at time t+k, μ x is the mean;

[0083] The estimated value c of the cross-covariance between two series with lag order k xy (k) is obtained from the following formula:

[0084]

[0085] In the formula, c xy(k) is the estimated value of the cross-covariance, k is the lag order, n is the sequence length, y t+k ,y t-k are the specific values ​​of the sequence at time t+k and tk, respectively, t is the specific value at time t, x t ,y t The sample mean of ;

[0086] The formula for calculating the mutual correlation coefficient of two series with lag order k is:

[0087]

[0088] In the formula, ρ xy (k) is the cross-correlation function, σ x is the standard deviation of the x series, σ y is the standard deviation of the y series, γ xy (k) is the cross-covariance coefficient between sequence x and sequence y, s x ,s y is the standard deviation of the two series;

[0089] For the wave data W used for model training, W = [w1, w2…w t ] and the ship's certain degree of freedom motion data E = [e1, e2…e t ], using sliding windows for cutting, constructing the data into input and output matrices, the results are as follows:

[0090]

[0091] Where W in is the wave data set, e in is the motion dataset, w t is the wave data at a certain moment, e t is the motion data at a certain moment, i is the length of the segment segmentation of the model sliding window, and the input matrix and the output matrix are matched row by row to obtain several groups of input and output segment combinations;

[0092] For each input-output fragment combination (W in [j],D out [j]), and obtain the mutual correlation coefficient γ(k) of lag order k; the optimal lag order of the segment is obtained according to the statistics of the mutual correlation coefficients at each order Then we have:

[0093]

[0094] In the formula, is the optimal lag order function, γ() is the cross-covariance function, and f is the search range of order k.

[0095] S3, based on the best lag order obtained Use sliding windows to divide the training set;

[0096] The sliding window fragments are divided twice based on the optimal lag order, and the corrected input and output matrices are obtained as follows:

[0097]

[0098] In the formula, are the wave and motion matrices after the data is divided twice, i is the length of the segment of the model sliding window, and w * is the wave data at time *, D * is the motion data at time *, a∈N * It is half of the cropping length of the secondary segmentation, and the actual input and output length is i-2a. Finally, the corrected input and output matrix is ​​used for ANN training.

[0099] S4, constructing an ANN prediction model based on the obtained data set;

[0100] An ANN prediction model is established, in which the number of input layer nodes is back, the number of hidden layer nodes is ahead, the number of output layer nodes is 1, and the activation function is the ReLU function. The expression is:

[0101]

[0102] In the formula, x is the independent variable of the function.

[0103] S5, obtaining the ship motion prediction result according to the constructed ANN prediction model.

[0104] The input is wave time history data, and the output is ship motion time history data; the training data set is X tr The input model is calculated and the calculation result is compared with Y in the training data set. tr Calculate the loss function, where X tr is the training set sample data, Y tr is the label value of the training set;

[0105] The back propagation algorithm is used to correct various parameters in the model according to the calculation results of the loss function, and the optimization is continuously iterated.

[0106] The loss function can be calculated by selecting mean square error, root mean square error, mean absolute error, and mean absolute percentage error.

[0107] Optimal lag order using the current motion history and the corresponding historical wave history Optimal lag order between the motion history of the surrogate target ship and the corresponding wave history Based on the optimal lag order Carry out adaptive correction, input the corrected data into the constructed ANN prediction model, and obtain the ship motion prediction results.

[0108] Embodiment 2, the ship motion intelligent prediction system integrated with system state estimation provided by the embodiment of the present invention comprises:

[0109] The data segmentation module is used to segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain a training set and a test set for time history forecasting;

[0110] The optimal hysteresis order calculation module is used to obtain the optimal hysteresis order through the hysteresis order analysis of the wave-ship motion system

[0111] The training set partitioning module is used to obtain the best lag order Use sliding windows to divide the training set;

[0112] ANN prediction model building module, used to build an ANN prediction model based on the obtained data set;

[0113] The ship motion prediction module is used to obtain the ship motion prediction results according to the constructed ANN prediction model.

[0114] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0115] The experimental data in this paper selects the ship pitching motion and the corresponding wave time history data under the 180° wave direction condition at the zero speed level 5 lower limit sea condition, such as Figure 3 and Figure 4 shown.

[0116] Step 1: Segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain the training set and test set of time history forecast;

[0117] Take the ship motion history data as an example: the known ship motion history data of a certain degree of freedom is D = [d1, d2…d T ], split into training set: D tr =[d1,d2…d n ], test set D te =[d n+1 ,d n+2 …d T]; the number of data in the training set and the test set can be set according to actual needs. The division of wave history data is the same as that of motion data.

[0118] Step 2: Analyze the hysteresis order of the wave-ship motion system to obtain the optimal hysteresis order

[0119] A stationary Gaussian random process can be represented by its mean μ and autocovariance function γ. k To describe, or equivalently, by its mean μ, variance σ 2 and the autocorrelation function ρ k to describe.

[0120] The formula for calculating the autocovariance coefficient of each series lag is as follows:

[0121] γ xx (k) = E[(x t -μ x )(x t+k -μ x )]=E[(x t -μ x )(x t-k -μ x )]

[0122] In the formula, γ xx (k) is x t The autocovariance of the sequence, E[] is the mathematical expectation, x t is the specific value of the sequence at time t, x t+k is the specific value of the sequence at time t+k, μ x is the mean;

[0123] The estimated value c of the cross-covariance between two series with lag order k xy (k) is obtained from the following formula:

[0124]

[0125] In the formula, c xy (k) is the estimated value of the cross-covariance, k is the lag order, n is the sequence length, y t+k ,y t-k are the specific values ​​of the sequence at time t+k and tk, respectively, t is the specific value at time t, c t ,y t The sample mean of ;

[0126] The formula for calculating the mutual correlation coefficient of two series with lag order k is:

[0127]

[0128] In the formula, ρ xy (k) is the cross-correlation function, σ x is the standard deviation of the x series, σ y is the standard deviation of the y series, γ xy (k) is the cross-covariance coefficient between sequence x and sequence y, s x ,s y is the standard deviation of the two series;

[0129] For the wave data W used for model training, W = [w1, w2…w t ] and the ship's certain degree of freedom motion data E = [e1, e2…e t ], using sliding windows for cutting, constructing the data into input and output matrices, the results are as follows:

[0130]

[0131] Where W in is the wave data set, e in is the motion dataset, w t is the wave data at a certain moment, e t is the motion data at a certain moment, i is the length of the segment segmentation of the model sliding window, and the input matrix and the output matrix are matched row by row to obtain several groups of input and output segment combinations;

[0132] For each input-output fragment combination (W in [j],D out [j]), and obtain the mutual correlation coefficient γ(k) of lag order k; the optimal lag order of the segment is obtained according to the statistics of the mutual correlation coefficients at each order Then we have:

[0133]

[0134] In the formula, is the optimal lag order function, γ() is the cross-covariance function, and f is the search range of order k.

[0135] Step 3: Based on the optimal lag order obtained Use sliding windows to divide the training set;

[0136] The sliding window fragments are divided twice based on the optimal lag order, and the corrected input and output matrices are obtained as follows:

[0137]

[0138] In the formula, are the wave and motion matrices after the data is divided twice, i is the length of the segment of the model sliding window, and w* is the wave data at time *, D * is the motion data at time *, a∈N * It is half of the cropping length of the secondary segmentation, and the actual input and output length is i-2a. Finally, the corrected input and output matrix is ​​used for ANN training.

[0139] Step 4: Construct an ANN prediction model based on the obtained data set;

[0140] The ANN prediction model is constructed based on the obtained data set, including:

[0141] An ANN prediction model is established, in which the number of input layer nodes is back, the number of hidden layer nodes is ahead, the number of output layer nodes is 1, and the activation function is the ReLU function. The expression is:

[0142]

[0143] In the formula, x is the independent variable of the function.

[0144] S5, obtaining the ship motion prediction result according to the constructed ANN prediction model.

[0145] The input is wave time history data, and the output is ship motion time history data; the training data set is X tr The input model is calculated and the calculation result is compared with Y in the training data set. tr Calculate the loss function, where X tr is the training set sample data, Y tr is the label value of the training set;

[0146] The back propagation algorithm is used to correct various parameters in the model according to the calculation results of the loss function, and the optimization is continuously iterated.

[0147] In the ship motion forecasting of the present invention, due to the lack of target ship motion history fragments, that is, future ship motion history fragments, it is impossible to obtain the optimal lag order of the target ship motion history and the corresponding wave history through cross-correlation calculation. However, the lag order changes relatively smoothly over time, so the best lag order of the current motion history and its corresponding historical wave history can be used. Approximate substitution. Based on the optimal lag order Carry out adaptive correction, input the corrected data into the constructed ANN prediction model, and obtain the ship motion prediction results.

[0148]

[0149] Where η prei is the motion prediction result, η expiis the actual result of the motion, and n is the number of sample points contained in the time sequence segment.

[0150]

[0151] In the formula, H s is the significant amplitude of the time-period pair;

[0152] Table 1 Table of the prediction error of pitch motion in the lower limit sea state of level 5

[0153]

[0154] according to Figure 5 As can be seen from Table 1, in this embodiment, the phase and amplitude of the prediction results of the ANN have large deviations from the true values, and the NRMSE reaches 32.11%, while the prediction results of the present invention have a high degree of consistency with the true values, and the NRMSE is only 8.91%, which has a significant advantage.

[0155] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for intelligent prediction of ship motion incorporating system state estimation, characterized in that: The method includes: S1, segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain the training set and test set of time history forecast; S2, through the hysteresis order analysis of the wave-ship motion system, the optimal hysteresis order is obtained S3, based on the best lag order obtained Use sliding windows to divide the training set; S4, constructing an ANN prediction model based on the obtained data set; S5, obtaining the ship motion prediction result according to the constructed ANN prediction model.

2. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 1 is characterized in that: In step S1, a training set and a test set of calendar forecast are obtained, including: The known time history data of a certain degree of freedom motion of a ship D=[d1,d2…d T ], split into training set D tr =[d1,d2…d n ] and the test set D te =[d n+1 ,d n+2 …d T ], where d i Represents the data at each time point; The data quantity of the training set and the test set is set according to actual needs, and the division of the wave history data is the same as that of the motion data.

3. The method for intelligent prediction of ship motion integrating system state estimation according to claim 1 is characterized in that: In step S2, the optimal lag order is obtained include: The formula for calculating the autocovariance coefficient of each series lag is as follows: c xx (k)=E[(x t -m x )(x t+k -m x )]=E[(x t -m x )(x t-k -m x )] In the formula, γ xx (k) is x t The autocovariance of the sequence, E[] is the mathematical expectation, x t is the specific value of the sequence at time t, x t+k is the specific value of the sequence at time t+k, μ x is the mean; The estimated value c of the cross-covariance between two series with lag order k xy (k) is obtained from the following formula: In the formula, c xy (k) is the estimated value of the cross-covariance, k is the lag order, n is the sequence length, y t+k ,y t-k are the specific values ​​of the sequence at time t+k and tk, respectively, t is the specific value at time t, x t ,y t The sample mean of ; The formula for calculating the mutual correlation coefficient of two series with lag order k is: In the formula, ρ xy (k) is the cross-correlation function, σ x is the standard deviation of the x series, σ y is the standard deviation of the y series, γ xy (k) is the cross-covariance coefficient between sequence x and sequence y, s x ,s y is the standard deviation of the two series; For the wave data W used for model training, W = [w1, w2…w t ] and the ship's certain degree of freedom motion data E = [e1, e2…e t ], using sliding windows for cutting, constructing the data into input and output matrices, the results are as follows: Where W in is the wave data set, e in is the motion dataset, w t is the wave data at a certain moment, e t is the motion data at a certain moment, i is the length of the segment segmentation of the model sliding window, and the input matrix and the output matrix are matched row by row to obtain several groups of input and output segment combinations; For each input-output fragment combination (W in [j],D out [j]), and obtain the mutual correlation coefficient γ(k) of lag order k; the optimal lag order of the segment is obtained according to the statistics of the mutual correlation coefficients at each order Then we have: In the formula, is the optimal lag order function, γ() is the cross-covariance function, and f is the search range of order k.

4. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 1 is characterized in that: In step S3, according to the obtained optimal lag order Use sliding windows to divide the training set, including: The sliding window fragments are divided twice based on the optimal lag order, and the corrected input and output matrices are obtained as follows: In the formula, are the wave and motion matrices after the data is divided twice, i is the length of the segment of the model sliding window, and w * is the wave data at time *, D * is the motion data at time *, a∈N * It is half of the cropping length of the secondary segmentation, and the actual input and output length is i-2a. Finally, the corrected input and output matrix is ​​used for ANN training.

5. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 1 is characterized in that: In step S4, an ANN prediction model is constructed according to the obtained data set, including: An ANN prediction model is established, in which the number of input layer nodes is back, the number of hidden layer nodes is ahead, the number of output layer nodes is 1, and the activation function is the ReLU function. The expression is: In the formula, x is the independent variable of the function.

6. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 5 is characterized in that: The number of neurons in the ANN prediction model constructed by the data set is adjusted according to the actual situation, and the input is wave time history data.

7. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 1 is characterized in that: In step S5, the ship motion prediction result is obtained according to the constructed ANN prediction model, including: The input is wave time history data, and the output is ship motion time history data; the training data set is X tr The input model is calculated and the calculation result is compared with Y in the training data set. tr Calculate the loss function, where X tr is the training set sample data, Y tr is the label value of the training set; The back propagation algorithm is used to correct various parameters in the model according to the calculation results of the loss function, and the optimization is continuously iterated.

8. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 7 is characterized in that: The loss function can be calculated by selecting mean square error, root mean square error, mean absolute error, and mean absolute percentage error.

9. The method for intelligent prediction of ship motion incorporating system state estimation according to claim 7, characterized in that: Optimal lag order using the current motion history and the corresponding historical wave history Optimal lag order between the motion history of the surrogate target ship and the corresponding wave history Based on the optimal lag order Carry out adaptive correction, input the corrected data into the constructed ANN prediction model, and obtain the ship motion prediction results.

10. An intelligent ship motion prediction system incorporating system state estimation, characterized in that: The system is used to control the ship motion intelligent prediction method integrated with system state estimation as described in any one of claims 1 to 9, and the system comprises: The data segmentation module is used to segment the ship's motion time history data of a certain degree of freedom and the corresponding wave time history data to obtain a training set and a test set for time history forecasting; The optimal hysteresis order calculation module is used to obtain the optimal hysteresis order through the hysteresis order analysis of the wave-ship motion system The training set partitioning module is used to obtain the best lag order Use sliding windows to divide the training set; ANN prediction model building module, used to build an ANN prediction model based on the obtained data set; The ship motion prediction module is used to obtain the ship motion prediction results according to the constructed ANN prediction model.

Images