CN114330828A - Method for forecasting movement rest period of ship - Google Patents
Method for forecasting movement rest period of ship Download PDFInfo
- Publication number
- CN114330828A CN114330828A CN202111431342.5A CN202111431342A CN114330828A CN 114330828 A CN114330828 A CN 114330828A CN 202111431342 A CN202111431342 A CN 202111431342A CN 114330828 A CN114330828 A CN 114330828A
- Authority
- CN
- China
- Prior art keywords
- ship
- wave
- sea
- steps
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention discloses a method for forecasting the movement resting period of a ship, which comprises the following steps: s1, data acquisition: sea surface image data are obtained through a stereoscopic vision camera, sea wave radar images are obtained through a ship-borne marine radar, the sea surface image data are converted into three-dimensional sea surface topography through stereo photogrammetry software, and sea wave motion characteristic data are obtained through inversion and correction; s2, processing data sequence: combining the wave motion characteristic data and the ship motion characteristic parameters into a ship-wave motion characteristic time-duration data sequence according to the time sequence, and performing normalization processing by combining with zoning management; s3, establishing a 2-layer long-short-term memory (LSTM) neural network model; s4, training and verifying the LSTM neural network model based on a sliding window segmentation method; and S5, predicting the ship motion resting period in a specific sea state by combining an LSTM neural network model with a ship-sea wave real-time motion characteristic data sequence. The method can realize the forecast of the ship motion resting period window.
Description
Technical Field
The invention relates to the technical field of advanced prediction of ship motion in open sea areas, in particular to a method for predicting the resting period of ship motion.
Background
The Period of rest (QP) refers to the "time interval during which all motions of the ship are within acceptable limits to perform the required activities". The "ship motion tolerance limits for performing vertical and short take-off and vertical landing operations are specified in the NATO standard protocol STANAG 4154 as: roll angle (roll)2.5 °, pitch angle (pitch)1.5 °, vertical velocity (Vs)1.0 m/s "; when the ship attitude is maintained within the motion range, the ship attitude is in a resting state meeting the requirement of aircraft landing. The association of the ship platform with other systems requires the duration of the quiet period. For example, the QP is required to reach at least 30s for tasks such as fixed-wing aircraft landing, helicopter take-off, missile launching, delivery and the like; fire control, general landing, maintenance, small-sized manned operation recovery, material supplement and the like, and the QP is required to be about 1 min.
The ultra-short-term prediction (6-11 s) technology of ship motion is sufficiently researched, and the traditional method comprises the following steps: a time domain model with wave sensor input data most consistent with ship response output data is searched by utilizing a statistical (regression analysis) method, so that the output prediction of the ship motion attitude within a plurality of seconds in the future is realized; the method has better effect on a linear system, and has low accuracy on time series modeling and forecasting of a nonlinear system; an improved strategy is to design a neural network model to realize the extremely short-term prediction of ship motion based on past time sequence data by utilizing the characteristic that a dynamic neural network can approximate a nonlinear function with any precision through learning.
The prediction of the ship motion resting period window is the QP detection and prediction for a relatively long time (> 30s) of the ship motion situation. However, in open sea areas, the wave force encountered by ships is very complex due to various factors such as ocean currents, weather, other navigation tools, etc., and even experienced captain sometimes cannot accurately capture the resting period of ship motion.
Disclosure of Invention
The invention aims to overcome the problems in the prior art, and provides a method for forecasting the movement resting period of a ship, which can realize the forecasting of a ship movement resting period window.
The invention provides a method for forecasting the movement resting period of a ship, which comprises the following steps:
s1, data acquisition: acquiring sea surface image data through a stereoscopic vision camera, acquiring sea wave radar images through a ship-borne marine radar, converting the sea surface image data into three-dimensional sea surface topography through stereo photogrammetry software, and acquiring a sea wave motion characteristic parameter array through inversion and correction;
s2, processing data sequence: combining the wave motion characteristic parameter array and the ship motion characteristic parameter array into a ship-wave motion characteristic parameter array according to time sequence, and carrying out normalization processing on the ship-wave motion characteristic parameter array by combining with zone management to obtain one-dimensional serialized data;
s3, establishing a long-term and short-term memory neural network prediction model;
s4, training and verifying the long-time memory neural network model based on a sliding window segmentation method in combination with the normalized ship-sea wave motion characteristic parameter array;
s5, real-time predicting the ship movement resting period in a specific sea state by combining the verified long-time memory neural network prediction model with the duration data of the ship-sea wave movement characteristics and processing the result in real time.
The method for forecasting the movement rest period of the ship, disclosed by the invention, is characterized in that the step S1 further comprises the following steps:
s11, acquiring sea surface image data within 300m near a ship through three-dimensional vision cameras, wherein the three-dimensional vision cameras are 3 groups and are respectively arranged on a bow, a left board and a right board, and two images obtained during synchronous exposure have an overlapping range of at least 60%;
s12, analyzing sea surface image data into a sea wave propagation phase and a sea wave propagation amplitude through stereo photogrammetry software, and calculating a sea wave propagation speed, wave steepness eta and a wave inclination angle gamma when meeting a ship to generate a three-dimensional sea surface topography;
s13, inverting the three-dimensional sea surface topography and performing slope and steepness correction by using a homonymy point matching technology and a rear intersection method of a stereopair image overlapping area to obtain a wave front amplitude A, a frequency f and a surface flow velocity B;
s14, obtaining a wave radar image of 240-3000 m through the carrier-borne marine radar, and obtaining the wave height H of the sense wave of the wave through inversionSPeak period TPThe main wave direction thetaPAnd a dominant wavelength LP；
S15, combining the wave steepness eta, the wave inclination angle gamma, the wave front amplitude A, the frequency f, the surface flow speed B and the wave height H of the wave senseSPeak period TPThe main wave direction thetaPAnd a dominant wavelength LPForm an array MCt of sea wave motion characteristic parameters.
The radar antenna of the X-band marine radar emits electromagnetic waves to the sea surface at a low glancing angle, small-scale rough waves and capillary waves generated by wind on the sea surface and the electromagnetic waves generate Bragg resonance scattering, and the scattering is modulated by long gravity waves on the sea surface, so that sea waves are imaged (displayed as sea clutter on a radar image); pixel values on the X-band echo image represent echo values of the radar in distance and scanning angle; the classical wave inversion algorithm assumes that the wave field and the flow field in an inversion region have stationarity, applies three-dimensional Fourier transform to a radar image sequence to obtain three-dimensional wave number-frequency radar image spectrum distribution of sea waves, and inverts the information of the sea waves and the sea currents according to the 'dispersion relation' existing in the sea wave frequency, the sea wave number and the sea currents of the gravity waves. Ideally, radar averagesThe image intensity should be given by r(-7/4)The form of (c) attenuates, but the actual echo situation of the radar is often quite complex. For example, sometimes under the influence of weather, the echo signal tends to decay more quickly in rainy days; because of the wind direction, the pixel values of the image at the scanning angle are different to different degrees. Therefore, the inversion result of the radar image is corrected in real time by adopting the wave inversion result of the optical stereo camera, so that the quality of wave flow data acquisition is ensured. When fog occurs and in the dark, the optical camera is limited in use, and the system is based on the X-band marine radar alone to invert the wave space-time characteristics.
According to the method for forecasting the movement resting period of the ship, as a preferred mode, in step S1, the inversion result of the sea wave radar image is corrected in real time through the inversion result of the optical stereo camera.
The method for forecasting the movement rest period of the ship, disclosed by the invention, is characterized in that the step S2 further comprises the following steps:
s21, acquiring the angular speed of the ship through a gyroscope, and obtaining the rotation angles in three directions after primary integration by combining with the geometric and physical parameters of the ship: roll thetaCAnd pitch psiCAnd bow KC；
S22, acquiring motion acceleration of the ship through an accelerometer, and obtaining translational speeds in three directions after primary integration by combining with geometric and physical parameters of the ship: swaying HD, swaying ZD and dangling CD;
s23, obtaining a ship motion parameter array M at the moment tt，Mt＝[HD，ZD，CD，KC，ΨC，θC]；
S24, combining ship motion parameter array M at time ttAnd sea wave motion characteristic parameter array MCtObtaining ship-sea wave linkage motion characteristic parameter array XtWherein X ist＝[Mt，MCt]，MC＝t[A，B，η，TP，ΘP，λP]；
S25, taking the ship course as 0 degree, and taking a circular sea area with the radius of P in the effective detection range of the radarDivided into N circular ring sectors, each sector having a central angle alphai-1360 °/N, where i is 1, 2, …, N; the sailing speed of the ship is Vm/s, the radius L of each ring is defined to be 2Vm, the wave flow field of the surrounding sea area is divided into P/L multiplied by N fan-shaped rings, and Q consistent with the course is obtained0The region weight is not less than Q0Left and right regions Q1，QN-1The weight of the ship is more than or equal to the weight of the rest areas, and when the ship is static, all the weights are 1;
s26, performing one-dimensional serialization on the sea area wave flow field weight of the two-dimensional circular ring area to obtain a one-dimensional weight sequence WQ ═ Q0，Q1，QN-1，，，QN，，，]。
The method for forecasting the movement rest period of the ship, which is disclosed by the invention, is used as a preferred mode, and the geometric and physical parameters of the ship in the steps S21 and S22 comprise the following steps: length, width of the profile, length of the bottom, depth of the deck, total tonnage, design hourly speed and design draft.
The method for forecasting the movement rest period of the ship is characterized in that as an optimal mode, the long-time memory neural network model in the step S3 comprises N time steps and 2 LSTM network layers (the number of the network layers is determined according to the complexity of the problem, and 2 layers are used in the method); the LSTM network layer comprises a work cell, and the work cell realizes the persistence and simulation of information through the structures of a forgetting gate f, an input gate i and a control gate o. The long-time memory neural network LSTM is a circulating neural network in a specific form; the method not only establishes the right connection between layers, but also establishes the right connection between the same layers, thereby effectively solving the problem that the gradient disappears when the cyclic neural network processes long sequence data. The gate is composed of a sigmoid network layer and a bitwise multiplication operation. When a message enters the LSTM network, only the message which accords with the algorithm authentication is left, and the unmatched message is left through the forgetting door. With fixed network model parameters, the weight scales at different times can be changed dynamically.
The invention relates to a method for forecasting ship movement resting period, which is a preferable mode, wherein a cell in an LSTM can update the state of an LSTM unit at each moment through calculation of the following four steps:
ft＝σ(Wf[ht-1，xt]+bf)
wherein: f. oftTo forget the door; σ () is an activation function, WfIs a weight; bfA bias term for a forget gate;
it＝σ(Wi[ht-1，xt]+bi)，
wherein: i.e. itIs an input gate; wiAnd WcIs a weight; biAnd bcBias terms for the input gate and the input node; h ist-1Is the output at time t-1; x is the number oftA new variable value input for the time t; tan h is a hyperbolic tangent function;
step 3, updating the cell state, multiplying the old state by the forgetting gate, discarding the information which is determined to be discarded, and adding the screened new information to obtain the state of the current working unit;
wherein: ctCell state at time t; ct-1Cell state at time t-1;
is the input state of the memory unit;
step 4, the layer processing value is processed by the tanh function to determine the final output value,
Ot＝σ(Wo[ht-1，xt]+bo)，ht＝Ot*tanh(Ct)
wherein: o istIs an output gate; woIs a weight; boIs the bias term of the output gate; h istIs the output at time t.
The method for forecasting the movement rest period of the ship is characterized in that as an optimal mode, the method for training the long-time memory neural network learning model in the step S4 comprises the following steps:
the number of the ship-sea wave time-lapse observation motion sequences is T, and the data sequence is [ X ] after the step S2i](i is more than or equal to 1 and less than or equal to T); selecting a sliding window with the length of N, limiting the length of an input training sequence by the length of the sliding window, and updating samples in the sliding window by adopting an iteration method; will [ X ]i](i is more than or equal to 1 and less than or equal to T) carrying out sliding window grouping, wherein each group comprises N +1 data, the first N data are used as input data, and the N +1 data value is predicted
Comparing errors with the true values of the (N + 1) th data, and iteratively adjusting network parameters;
defining a loss function
And (5) performing model optimization by adopting a gradient descent method, and finishing training when the value of the loss function is less than 0.00005.
The method for forecasting the movement rest period of the ship, disclosed by the invention, is characterized in that the step S5 further comprises the following steps:
s51, setting the motion state of the ship needing to be predicted in N steps, replacing the oldest data in the sliding window with the predicted value in the ith step through iteration, wherein i is 1, 2, 3, … and N, and enabling the LSTM network to perform new learning once every time the LSTM network is replaced, updating the network structure and performing next prediction by using the new network structure;
s52, taking three parameters of output roll angle, pitch angle and heave speed of the LSTM model as joint judgment variables, namely
S53, defining judgment threshold YτJudging that the variable Y is less than the predetermined threshold Y when 15 consecutive outputsτThe system initiates the forecast of the time when the ship enters the rest period.
The method for forecasting the ship movement resting period is used as an optimal mode, and when the sea state changes, the steps S1-S5 are repeated to complete forecasting of the ship movement resting period in the new sea state.
Compared with the prior art, the invention has the beneficial effects that:
(1) the method comprises the steps of extracting sea wave characteristics around a ship from a range of 240-3000 m by using an X-band marine radar, acquiring wave characteristic data flapping to the bow within 300m near the ship by using a stereo photogrammetric system, and recording ship attitude motion data by using a ship attitude sensor arranged on the ship; the data contain the characteristic rule that the interaction between the wave flow motion and the ship motion reaches 'rest' balance for a moment; the data are well used, and the problem of predicting the movement resting period of the ship can be solved.
(2) The prediction system of the invention applies an artificial intelligence deep learning technology, trains a long-term memory neural network by adopting a 'sea wave-ship' linkage duration data sequence, enables the system to continuously recognize the motion characteristics of waves around a ship, continuously learns the response condition of a specific ship to the wave action force under a specific sea condition, gradually grasps the characteristic rule of interaction between wave motion and ship motion, and finally captures the time window of the ship motion resting period.
(3) The system provided by the invention constructs a large motion characteristic data sequence processing mechanism of ships and sea waves on the basis of carrying out full factor analysis on the motion situation of the ship, and carrying out professional sensor data acquisition and professional inversion and correction processing, so that abundant and accurate training data input is provided for memorizing a neural network model at long and short time, and the effectiveness of model establishment and training is ensured.
Drawings
FIG. 1 is a flow chart of a method for forecasting a ship motion resting period;
fig. 2 is a flowchart of a method step S1 for forecasting a movement resting period of a ship;
FIG. 3 is a schematic diagram of a ship motion and sea wave motion data acquisition method;
FIG. 4 is a flow chart of marine radar image sequence inversion wave features;
fig. 5 is a flowchart of a method step S2 for forecasting a movement resting period of a ship;
fig. 6 is a schematic diagram of zone management;
FIG. 7 is a diagram of the internal structure of the LSTM neural network;
FIG. 8 is a two-layer LSTM neural network learning model framework;
FIG. 9 is a schematic diagram of a sliding window based LSTM model training and validation method;
fig. 10 is a flowchart of a method step S5 for forecasting a rest period of ship motion.
Detailed Description
Example 1
As shown in fig. 1, a method for forecasting a movement rest period of a ship comprises the following steps:
s1, data acquisition: acquiring sea surface image data through a stereoscopic vision camera, acquiring a sea wave radar image through a ship-borne marine radar, converting the sea surface image data into a three-dimensional sea surface topography through stereo photogrammetry software, acquiring sea wave motion characteristic data through inversion, and correcting the inversion result of the sea wave radar image in real time through the inversion result of an optical stereo camera; as shown in fig. 2, step S1 further includes the following steps:
s11, acquiring sea surface image data within 300m near a ship through three-dimensional vision cameras, wherein the three-dimensional vision cameras are 3 groups and are respectively arranged on a bow, a left board and a right board, and two images obtained during synchronous exposure have an overlapping range of at least 60%;
s12, analyzing sea surface image data into a sea wave propagation phase and a sea wave propagation amplitude through stereo photogrammetry software, and calculating a sea wave propagation speed, wave steepness eta and a wave inclination angle gamma when meeting a ship to generate a three-dimensional sea surface topography; a schematic diagram of a sea surface image data acquisition method is shown in FIG. 3;
s13, as shown in the figure 4, inverting the three-dimensional sea surface map and carrying out slope and steepness correction by using the homonymous point matching technology and the rear intersection method of the stereo image pair image overlapping area to obtain a wave front amplitude A, a frequency f and a surface flow velocity B;
s14, obtaining a wave radar image of 240-3000 m through the carrier-borne marine radar, and obtaining the wave height H of the sense wave of the wave through inversionSPeak period TPThe main wave direction thetaPAnd a dominant wavelength LP; the radar antenna is erected near the center of the ship body, the height of the antenna is 10-45 m, and a schematic diagram of an ocean wave radar image acquisition method is shown in FIG. 3; the main technical characteristics of the X-band marine radar are shown in the following table:
s15, combining the wave steepness eta, the wave inclination angle Y, the wave front amplitude A, the frequency f, the surface flow speed B and the wave height H of the wave senseSPeak period TPThe main wave direction thetaPAnd a dominant wavelength LPForming an array MCt of sea wave motion characteristic parameters;
s2, processing data sequence: combining the wave motion characteristic parameter array and the ship motion characteristic parameter array into a ship-wave motion characteristic parameter array, and carrying out normalization processing on ship-wave motion characteristic data by combining with region management to obtain one-dimensional serialized data; as shown in fig. 5, step S2 further includes the steps of:
s21, acquiring the angular speed of the ship through a gyroscope, and obtaining the rotation angles in three directions after primary integration by combining with the geometric and physical parameters of the ship: roll thetaCAnd pitch psiCAnd bow KC(ii) a The ship geometric and physical parameters comprise: length, width of the profile, length of the bottom, depth of the deck, total tonnage, design hourly speed and design draft;
s22, acquiring motion acceleration of the ship through an accelerometer, and obtaining translational speeds in three directions after primary integration by combining with geometric and physical parameters of the ship: swaying HD, swaying ZD and dangling CD;
s23, obtaining a ship motion parameter array M at the moment tt，Mt＝[HD，ZD，CD，KC，ΨC，θc]；
S24 Combined ship motion parameter array MtAnd sea wave motion characteristic parameter array MCtObtaining ship-sea wave linkage motion characteristic parameter array XtWherein X ist＝[Mt，MCt]，MCt＝[A，B，η，TP，ΘP，λP]；
S25, as shown in figure 6, dividing N circular ring sectors by taking a circular sea area with the radius of P in the effective detection range of the radar with the ship course of 0 degree, wherein the central angle alpha of each sectori-1360 °/N, where i is 1, 2, …, N; the sailing speed of the ship is Vm/s, the radius L of each ring is defined to be 2Vm, the wave flow field of the surrounding sea area is divided into P/L multiplied by N fan-shaped rings, and Q consistent with the course is obtained0The region weight is not less than Q0Left and right regions Q1，QN-1The weight of the ship is more than or equal to the weight of the rest areas, and when the ship is static, all the weights are 1;
s26, performing one-dimensional serialization on the sea area wave flow field weight of the two-dimensional circular ring area to obtain a one-dimensional weight sequence WQ ═ Q0，Q1，QN-1，，，QN，，，]；
S3, establishing a long-time memory neural network model; as shown in fig. 7 to 8, the long-and-short-term memory neural network learning model includes N time steps and 2 LSTM network layers; the LSTM network layer comprises a work cell, and the work cell realizes the persistence and simulation of information through the structures of a forgetting gate f, an input gate i and a control gate o;
the cell in the LSTM updates the state of the LSTM cell at each time by the following four steps of calculation:
ft＝σ(Wf[ht-1，xt]+bf)
wherein: f. oftTo forget the door; σ () is an activation function, WfIs a weight; bfA bias term for a forget gate;
it＝σ(Wi[ht-1，xt]+bi)，
wherein: i.e. itIs an input gate; wiAnd WcIs a weight; biAnd bcBias terms for the input gate and the input node; h ist-1Is the output at time t-1; x is the number oftA new variable value input for the time t; tan h is a hyperbolic tangent function;
step 3, updating the cell state, multiplying the old state by the forgetting gate, discarding the information which is determined to be discarded, and adding the screened new information to obtain the state of the current working unit;
wherein: ctCell state at time t; ct-1Cell state at time t-1;
is the input state of the memory unit;
step 4, the layer processing value is processed by the tanh function to determine the final output value,
Ot＝σ(Wo[ht-1，xt]+bo)，ht＝Ot*tanh(Ct)
wherein: o istIs an output gate; woIs a weight; boIs the bias term of the output gate; h istIs the output at time t;
s4, training and verifying the long-time memory neural network model based on a sliding window segmentation method in combination with the normalized ship-sea wave motion characteristic parameter array; as shown in fig. 9, the method for training the long-term and short-term memory neural network learning model includes:
the number of the ship-sea wave time-lapse observation motion sequences is T, and the data sequence is [ X ] after the step S2i](i is more than or equal to 1 and less than or equal to T); selecting a sliding window with the length of N, limiting the length of an input training sequence by the length of the sliding window, and updating samples in the sliding window by adopting an iteration method; will [ X ]i](i is more than or equal to 1 and less than or equal to T) carrying out sliding window grouping, wherein each group comprises N +1 data, the first N data are used as input data, and the N +1 data value is predicted
Comparing errors with the true values of the (N + 1) th data, and iteratively adjusting network parameters;
defining a loss function
Performing model optimization by adopting a gradient descent method, and finishing training when the value of the loss function is less than 0.00005;
s5, real-time predicting the ship movement resting period in a specific sea state by combining the verified long-time memory neural network learning model with the ship-sea wave movement characteristic duration data real-time processing result; as shown in fig. 10, step S5 further includes the steps of:
s51, setting the motion state of the ship needing to be predicted in N steps, replacing the oldest data in the sliding window with the predicted value in the ith step through iteration, wherein i is 1, 2, 3, … and N, and enabling the LSTM network to perform new learning once every time the LSTM network is replaced, updating the network structure and performing next prediction by using the new network structure;
s52, taking three parameters of roll angle, pitch angle and heave output by the LSTM model as joint judgment variables, namely
S53, defining judgment threshold YτWhen connecting toThe judgment variable Y of the next 15 outputs is less than the predetermined threshold value YτThe system initiates the forecast of the time when the ship enters the resting period;
and S6, when the sea state changes, repeating the steps S1-S5 to finish the forecast of the ship movement resting period in the new sea state.
The foregoing description is intended to be illustrative rather than limiting, and it will be appreciated by those skilled in the art that various modifications, changes, and equivalents may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A method for forecasting the movement rest period of a ship is characterized by comprising the following steps: the method comprises the following steps:
s1, data acquisition: acquiring sea surface image data through a stereoscopic vision camera, acquiring a sea wave radar image through a ship-borne marine radar, converting the sea surface image data into a three-dimensional sea surface topography through stereo photogrammetry software, and acquiring a sea wave motion characteristic parameter array through inversion and correction;
s2, processing data sequence: combining the wave motion characteristic parameter array and the ship motion characteristic parameter array into a ship-wave motion characteristic parameter array according to a time sequence, and performing normalization processing on the ship-wave motion characteristic parameter array by combining with zoning management to obtain one-dimensional serialized data;
s3, establishing a long-term and short-term memory neural network prediction model;
s4, training and verifying the long-time memory neural network model based on a sliding window segmentation method in combination with the normalized ship-sea wave motion characteristic parameter array;
and S5, instantly predicting the motion resting period of the ship in a specific sea state by combining the verified long-time memory neural network prediction model with the duration data real-time processing result of the ship-sea wave motion characteristic.
2. The method for forecasting the motional rest period of the ship according to claim 1, wherein the method comprises the following steps: step S1 further includes the steps of:
s11, acquiring sea surface image data within 300m near a ship through three-dimensional vision cameras, wherein the three-dimensional vision cameras are 3 groups and are respectively arranged on the bow, the left side and the right side of the ship;
s12, analyzing the sea surface image data into a sea wave propagation phase and a sea wave propagation amplitude through stereo photogrammetry software, and calculating a sea wave propagation speed, wave steepness eta and a wave inclination angle Y when meeting a ship to generate a three-dimensional sea surface topography;
s13, inverting the three-dimensional sea surface topography and performing slope and steepness correction by using a homonymy point matching technology and a rear intersection method of a stereopair image overlapping area to obtain a wave front amplitude A, a frequency f and a surface flow velocity B;
s14, obtaining a wave radar image of 240-3000 m through the carrier-borne marine radar, and obtaining the sense wave height H of the wave through inversionSPeak period TPThe main wave direction thetaPAnd a dominant wavelength LP；
S15, setting the wave steepness eta, the wave inclination angle gamma, the wave front amplitude A, the frequency f, the surface flow speed B and the wave height H of the sense wave of the sea waveSThe peak period TPAnd the main wave direction thetaPAnd said dominant wavelength LPForm the sea wave motion characteristic parameter array MCt。
3. The method for forecasting the motional rest period of the ship according to claim 2, wherein the method comprises the following steps: in step S1, the inversion result of the sea wave radar image is corrected in real time according to the inversion result of the optical stereo camera.
4. The method for forecasting the motional rest period of the ship according to claim 1, wherein the method comprises the following steps: step S2 further includes the steps of:
s21, acquiring the angular speed of the ship through a gyroscope, and obtaining the rotation angles in three directions after primary integration by combining with the geometric and physical parameters of the ship: roll thetaCAnd pitch psiCAnd bow KC；
S22, acquiring motion acceleration of the ship through an accelerometer, and obtaining translational speeds in three directions after primary integration by combining with geometric and physical parameters of the ship: swaying HD, swaying ZD and dangling CD;
s23, obtaining a ship motion parameter array M at the moment tt，Mt＝[HD，ZD，CD，KC，ΨC，θC]；
S24, combining the ship motion parameter arrays M at the t momenttAnd the sea wave motion characteristic parameter array MCtObtaining the ship-sea wave linkage motion characteristic parameter array XtWherein X ist＝[Mt，MCt]，MCt＝[A，B，η，TP，ΘP，λP]；
S25, dividing a circular sea area with the radius of P into N circular ring sectors within the effective detection range of the radar with the ship course of 0 degree, wherein the central angle alpha of each sectori-1360 °/N, where i is 1, 2, …, N; the sailing speed of the ship is Vm/s, the radius L of each ring is defined to be 2Vm, the wave flow field of the surrounding sea area is divided into P/L multiplied by N fan-shaped rings, and Q consistent with the course is obtained0The region weight is not less than Q0Left and right regions Q1，QN-1The weight of the ship is more than or equal to the weight of the rest areas, and when the ship is static, all the weights are 1;
s26, performing one-dimensional serialization processing on the sea area wave flow field weight in the two-dimensional annular area to obtain a one-dimensional weight sequence WQ ═ Q0，Q1，QN-1，，，QN，，，]。
5. The method for forecasting the motional rest period of the ship according to claim 4, wherein the method comprises the following steps: the ship geometric and physical parameters in the steps S21 and S22 include: length, width of the profile, length of the bottom, depth of the deck, total tonnage, design hourly speed and design draft.
6. The method for forecasting the motional rest period of the ship according to claim 1, wherein the method comprises the following steps: the long-time memory neural network model in the step S3 includes N time steps and 2 LSTM network layers; the LSTM network layer comprises a work cell, and the work cell realizes information persistence and simulation through the structures of a forgetting gate f, an input gate i and a control gate o.
7. The method for forecasting the motional rest period of a ship according to claim 6, wherein the method comprises the following steps: the cell in the LSTM updates the state of the LSTM cell at each time by the following four steps of calculation:
step 1, screening h of forgetting gate f through activation functiont-1And xtInformation, deciding which information to discard from the last state,
ft＝σ(Wf[ht-1，xt]+bf)
wherein: f. oftTo forget the door; σ () is an activation function, WfIs a weight; bfA bias term for a forget gate;
step 2, the input gate selects which information to keep through the sigmoid function, generates candidate vector values through the activation function tanh,
it＝σ(Wi[ht-1，xt]+bi)，
wherein: i.e. itIs an input gate; wiAnd WcIs a weight; biAnd bcBias terms for the input gate and the input node; h ist-1Is the output at time t-1; x is the number oftA new variable value input for the time t; tan h is a hyperbolic tangent function;
step 3, updating the cell state, multiplying the old state by the forgetting gate, discarding the information which is determined to be discarded, and adding the screened new information to obtain the state of the current working unit;
wherein: ctCell state at time t; ct-1Cell state at time t-1;
is the input state of the memory unit;
step 4, the layer processing value is processed by the tanh function to determine the final output value,
Ot＝σ(Wo[ht-1，xt]+bo)，ht＝Ot*tanh(Ct)
wherein: o istIs an output gate; woIs a weight; boIs the bias term of the output gate; h istIs the output at time t.
8. The method for forecasting the motional rest period of the ship according to claim 1, wherein the method comprises the following steps: in step S4, the method for training the long-term and short-term memory neural network model includes:
the number of the ship-sea wave time-lapse observation motion sequences is T, and the data sequence is [ X ] after the step S2i](i is more than or equal to 1 and less than or equal to T); selecting a sliding window with the length of N, limiting the length of an input training sequence by the length of the sliding window, and updating samples in the sliding window by adopting an iteration method; will [ X ]i](i is more than or equal to 1 and less than or equal to T) carrying out sliding window grouping, wherein each group comprises N +1 data, the first N data are used as input data, and the N +1 data value is predicted
Comparing errors with the true values of the (N + 1) th data, and iteratively adjusting network parameters;
defining a loss function
Model optimization is carried out by adopting a gradient descent method when the value of a loss functionAnd (5) finishing training if the frequency is less than 0.00005.
9. The method for forecasting the motional rest period of the ship according to claim 1, wherein the method comprises the following steps: step S5 further includes the steps of:
s51, setting the motion state of the ship needing to be predicted in N steps, replacing the oldest data in the sliding window with the predicted value in the ith step through iteration, wherein i is 1, 2, 3, … and N, and enabling the LSTM network to perform new learning once every time the LSTM network is replaced, updating the network structure and performing next prediction by using the new network structure;
s52, taking three parameters of output roll angle, pitch angle and heave of the LSTM model as joint judgment variables, namely
S53, defining judgment threshold YτWhen the judgment variable Y of the continuous 15 outputs is less than the predetermined threshold value YτAnd then the system initiates the forecast of the time when the ship enters the rest period.
10. The method for forecasting the motional rest period of a ship according to any one of claims 1 to 9, wherein the method comprises the following steps: and when the sea state changes, repeating the steps S1-S5 to complete the forecast of the ship movement resting period in the new sea state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111431342.5A CN114330828A (en) | 2021-11-29 | 2021-11-29 | Method for forecasting movement rest period of ship |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111431342.5A CN114330828A (en) | 2021-11-29 | 2021-11-29 | Method for forecasting movement rest period of ship |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114330828A true CN114330828A (en) | 2022-04-12 |
Family
ID=81046271
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111431342.5A Pending CN114330828A (en) | 2021-11-29 | 2021-11-29 | Method for forecasting movement rest period of ship |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330828A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116842381A (en) * | 2023-06-13 | 2023-10-03 | 青岛哈尔滨工程大学创新发展中心 | Ship motion extremely-short-term prediction model generalization optimization method based on data fusion |
| CN117048802A (en) * | 2023-04-26 | 2023-11-14 | 哈尔滨工业大学（威海） | Ship future motion attitude prediction method and system based on real sea state strong adaptation |
-
2021
- 2021-11-29 CN CN202111431342.5A patent/CN114330828A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117048802A (en) * | 2023-04-26 | 2023-11-14 | 哈尔滨工业大学（威海） | Ship future motion attitude prediction method and system based on real sea state strong adaptation |
| CN116842381A (en) * | 2023-06-13 | 2023-10-03 | 青岛哈尔滨工程大学创新发展中心 | Ship motion extremely-short-term prediction model generalization optimization method based on data fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10936907B2 (en) | Training a deep learning system for maritime applications | |
| CN110796037B (en) | Satellite-borne optical remote sensing image ship target detection method based on lightweight receptive field pyramid | |
| CN110782481B (en) | Unmanned ship intelligent decision-making method and system | |
| CN110222612B (en) | Dynamic target identification and tracking method for autonomous landing of unmanned aerial vehicle | |
| CN110866887A (en) | Target situation fusion sensing method and system based on multiple sensors | |
| CN114330828A (en) | Method for forecasting movement rest period of ship | |
| KR20220155559A (en) | Autonomous navigation method using image segmentation | |
| US20210334651A1 (en) | Learning point cloud augmentation policies | |
| CN109596078A (en) | Multi-information fusion spectrum of road surface roughness real-time testing system and test method | |
| CN113627473B (en) | Multi-mode sensor-based water surface unmanned ship environment information fusion sensing method | |
| JP2017522548A (en) | Data fusion analysis for automatic maritime target recognition | |
| US11967103B2 (en) | Multi-modal 3-D pose estimation | |
| KR102466804B1 (en) | Autonomous navigation method using image segmentation | |
| KR20200094644A (en) | Learning method and learning device for updating hd map by reconstructing 3d space by using depth estimation information and class information on each object, which have been acquired through v2x information integration technique, and testing method and testing device using the same | |
| CN113408340A (en) | Dual-polarization SAR small ship detection method based on enhanced feature pyramid | |
| Trunov | Criteria for the evaluation of model's error for a hybrid architecture DSS in the underwater technology ACS | |
| CN116563466A (en) | Deep learning-based three-dimensional Shan Mudian cloud completion method | |
| CN116027349A (en) | Coral reef substrate classification method based on laser radar and side scan sonar data fusion | |
| Almanza-Medina et al. | Deep learning architectures for navigation using forward looking sonar images | |
| CN113933828A (en) | Unmanned ship environment self-adaptive multi-scale target detection method and system | |
| Thompson | Maritime object detection, tracking, and classification using lidar and vision-based sensor fusion | |
| CN116184376A (en) | Underwater three-dimensional terrain and multi-beam image sonar data simulation system and method | |
| CN104615987A (en) | Method and system for intelligently recognizing aircraft wreckage based on error back propagation neural network | |
| CN111339836A (en) | SAR image ship target detection method based on transfer learning | |
| Shakirzyanov et al. | Method for unmanned vehicles automatic positioning based on signal radially symmetric markers recognition of underwater targets |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |