CN113837454A - Hybrid neural network model prediction method and system for three degrees of freedom of ship - Google Patents
Hybrid neural network model prediction method and system for three degrees of freedom of ship Download PDFInfo
- Publication number
- CN113837454A CN113837454A CN202111053868.4A CN202111053868A CN113837454A CN 113837454 A CN113837454 A CN 113837454A CN 202111053868 A CN202111053868 A CN 202111053868A CN 113837454 A CN113837454 A CN 113837454A
- Authority
- CN
- China
- Prior art keywords
- ship
- short term
- term memory
- bidirectional long
- memory network
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000003062 neural network model Methods 0.000 title claims abstract description 27
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 102
- 230000006403 short-term memory Effects 0.000 claims abstract description 70
- 230000007787 long-term memory Effects 0.000 claims abstract description 69
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 32
- 239000002245 particle Substances 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 238000005457 optimization Methods 0.000 claims abstract description 26
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 23
- 238000012795 verification Methods 0.000 claims abstract description 20
- 230000005484 gravity Effects 0.000 claims abstract description 18
- 238000010845 search algorithm Methods 0.000 claims abstract description 15
- 238000012952 Resampling Methods 0.000 claims abstract description 7
- 230000015654 memory Effects 0.000 claims description 46
- 230000006870 function Effects 0.000 claims description 22
- 210000002569 neuron Anatomy 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 13
- 238000001228 spectrum Methods 0.000 claims description 13
- 238000004458 analytical method Methods 0.000 claims description 9
- 230000036544 posture Effects 0.000 description 10
- 230000003044 adaptive effect Effects 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 9
- 239000000126 substance Substances 0.000 description 6
- 230000001133 acceleration Effects 0.000 description 5
- 238000004590 computer program Methods 0.000 description 5
- 238000005096 rolling process Methods 0.000 description 5
- 241000282461 Canis lupus Species 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013277 forecasting method Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000002441 reversible effect Effects 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000003116 impacting effect Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000004873 anchoring Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000011664 nicotinic acid Substances 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B63—SHIPS OR OTHER WATERBORNE VESSELS; RELATED EQUIPMENT
- B63B—SHIPS OR OTHER WATERBORNE VESSELS; EQUIPMENT FOR SHIPPING
- B63B71/00—Designing vessels; Predicting their performance
- B63B71/10—Designing vessels; Predicting their performance using computer simulation, e.g. finite element method [FEM] or computational fluid dynamics [CFD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C5/00—Registering or indicating the working of vehicles
- G07C5/08—Registering or indicating performance data other than driving, working, idle, or waiting time, with or without registering driving, working, idle or waiting time
- G07C5/0808—Diagnosing performance data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Molecular Biology (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Business, Economics & Management (AREA)
- Evolutionary Biology (AREA)
- Physiology (AREA)
- Development Economics (AREA)
- Ocean & Marine Engineering (AREA)
- Game Theory and Decision Science (AREA)
- Mechanical Engineering (AREA)
- Combustion & Propulsion (AREA)
- Entrepreneurship & Innovation (AREA)
- Chemical & Material Sciences (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Fluid Mechanics (AREA)
- Computer Vision & Pattern Recognition (AREA)
Abstract
The invention provides a method and a system for predicting a three-degree-of-freedom hybrid neural network model of a ship, which comprises the steps of acquiring original ship shaking attitude data by an attitude sensor arranged on the ship, and decoding the original ship shaking attitude data by resampling to obtain a ship shaking attitude time sequence; carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed subsequences, forming a subsequence matrix to reduce the nonlinear non-stationarity of three-degree-of-freedom ship motion, and dividing the subsequence into a training set, a verification set and a test set; the bidirectional long and short term memory network is introduced to learn the time characteristics of three degrees of freedom of the ship from the past and the future, the hybrid optimization algorithm combining the gravity search algorithm and the particle swarm algorithm is used for optimizing the number of nodes of the hidden layer and the learning rate, and the results are predicted and output based on the optimized bidirectional long and short term memory network. The method improves the stability and accuracy of the three-degree-of-freedom prediction of the ship.
Description
Technical Field
The invention relates to a three-degree-of-freedom prediction method for a ship, in particular to a three-degree-of-freedom hybrid prediction method and a three-degree-of-freedom hybrid prediction system for the ship based on a bidirectional long and short-term memory network.
Background
When a ship sails on the sea, the ship is influenced by uncertain sea conditions such as sea wind, sea waves, ocean currents and the like, and the three-degree-of-freedom swinging motion inevitably occurs. The analysis was mainly performed from two angles: from the civil angle analysis, the dynamic effects generated by the violent swaying motion of the ship, such as slamming, wave rising, stalling and even the phenomenon of rotary drift, can all give ship surface operation in the big stormy waves, such as cargo moving, dynamic positioning, anchoring and the like, great potential safety hazards are generated, and the risks of ship sinking and anchor-walking accidents caused by improper operation time exist all the time; from the military perspective, when a carrier-based aircraft or a carrier-based helicopter takes off and lands under severe sea conditions, the transient additional speed of the aircraft body contacting the deck and the rolling of the deck increase the risk of the aircraft body impacting the deck or impacting the ship body, and the visible signal is poor and is easily restricted by night weather, so that the landing safety of the landing aircraft is seriously threatened.
At the moment, if the information of the rolling attitude of the future ship can be accurately predicted in a short time, the deck operator of the civil ship can avoid the dangerous time period of offshore operation, and the pilot of the carrier-based aircraft can also predict the body rolling state of the aircraft body when contacting the deck, plan the landing plan in advance, reduce the landing risk of the carrier-based aircraft, and ensure the operational efficiency of the ship and the carrier-based weapon. Therefore, the prediction of the ship motion is the key for ensuring the operation safety of the offshore ship surface and improving the operation efficiency.
In the early stage of the ship motion prediction research, a forecasting method based on a linear hydrodynamic motion equation is firstly proposed. The method can be mainly divided into a convolution forecasting method and a Kalman filtering forecasting method. However, this kind of method is susceptible to noise, and is not highly stable in practical application, and it is difficult to achieve the desired result. With the development of artificial intelligence and nonlinear theory, some research methods based on nonlinear theory are gradually applied to ship attitude prediction, such as support vector machines, neural networks and the like. However, for the non-linearity and non-stationarity of a ship motion sequence under the condition of medium and high sea, the high-precision prediction of three degrees of freedom of a ship can be expected to be realized by a certain neural network mixed model prediction method.
Disclosure of Invention
The invention aims to provide a three-degree-of-freedom hybrid bidirectional long and short term memory network model prediction scheme for a ship based on adaptive empirical wavelet transform and hybrid hyper-parameter optimization, aiming at the problem of low prediction precision in the three degrees of freedom of the ship at present.
The invention provides a hybrid neural network model prediction method of three degrees of freedom of a ship, which comprises the following steps:
1) acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
2) carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed subsequences to form a subsequence matrix so as to reduce the nonlinear non-stationarity of the ship three-degree-of-freedom motion;
3) dividing the subsequence matrix into a training set, a verification set and a test set based on the subsequence matrix obtained by decomposing in the step 2);
4) the method comprises the steps of sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, obtaining a ship attitude sequence at a future moment through prediction of the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at a real time, and further performing optimization training on neuron weight and hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node number and learning rate, so that the neuron weight and hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
5) inputting the verification set obtained in the step 3) into the initial bidirectional long and short term memory network obtained in the step 4), optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into an initial bidirectional long and short term memory network model to obtain the optimized bidirectional long and short term memory network;
6) and evaluating the optimized bidirectional long and short term memory network in the step 5), wherein the test set obtained in the step 3) is input into the optimized bidirectional long and short term memory network obtained in the step 5 to carry out robustness test, error analysis is carried out on the prediction result, whether the prediction result is reliable or not is evaluated, and if the prediction result is reliable, the result is predicted and output based on the optimized bidirectional long and short term memory network.
And in the step 2), Fourier transform is carried out on the ship attitude time sequence according to the shaking rule of the ship three-degree-of-freedom data.
And in the step 2), the second derivative is solved for the signal spectrum, the number of maximum points of the signal spectrum is obtained, each frequency band interval is divided according to the number of maximum points, and each frequency band interval corresponds to each subsequence, so that the process of self-adaptive decomposition is realized.
And in the step 2), performing Hilbert transform on the empirical mode component to obtain a decomposed sequence matrix.
And in step 3), dividing the decomposed subsequence matrix into a training set, a verification set and a test set according to a ratio of 8:1:1 by combining the length L of the rows.
And in the step 5), setting the number range of the hidden layer nodes to be 10:100 and the learning rate range to be 0.0001-0.1.
On the other hand, the invention also provides a ship three-degree-of-freedom hybrid neural network model prediction system, which is used for realizing the ship three-degree-of-freedom hybrid neural network model prediction method.
And, including the following modules,
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
the second module is used for carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed sub-sequences and form a sub-sequence matrix so as to reduce the nonlinear non-stationarity of the ship three-degree-of-freedom motion;
the third module is used for dividing the subsequence matrix obtained by the decomposition of the second module into a training set, a verification set and a test set;
the fourth module is used for introducing a bidirectional long and short term memory network to learn the time characteristics of three degrees of freedom of the ship from the past and the future, and comprises the steps of sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, predicting and obtaining a ship attitude sequence at the future moment through the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at the real time moment, further performing optimization training on the neuron weight and the hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node numbers and a learning rate, and the neuron weight and the hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
the fifth module is used for inputting the verification set obtained by the third module into the initial bidirectional long-short term memory network obtained by the fourth module, optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into the initial bidirectional long-short term memory network model to obtain the optimized bidirectional long-short term memory network;
and a sixth module for evaluating the optimized bidirectional long-short term memory network of the fifth module, which comprises inputting the test set obtained by the third module into the optimized bidirectional long-short term memory network obtained by the fifth module for robustness test, performing error analysis on the prediction result, evaluating whether the prediction result is reliable, and predicting and outputting the result based on the optimized bidirectional long-short term memory network if the prediction result is reliable.
Or the three-degree-of-freedom hybrid neural network model prediction method comprises a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute the three-degree-of-freedom hybrid neural network model prediction method for the ship.
The invention adopts self-adaptive empirical wavelet decomposition to decompose a data set into a limited number of stable subsequences, and then combines each subsequence with a bidirectional long-short term memory network model respectively, thereby realizing high-precision prediction of the actual three-degree-of-freedom nonlinear data of a ship;
and optimizing the bidirectional long and short term memory network by adopting a hybrid hyper-parameter optimization mode to obtain the optimal hyper-parameter of the bidirectional long and short term memory network and show the optimal performance of the bidirectional long and short term memory network.
Compared with the prior art, the invention has the following advantages and beneficial effects:
the decomposition and reconstruction process based on the self-adaptive EWT algorithm can well reduce the nonlinear non-stationarity of three-degree-of-freedom motion of the ship, and different from the prior art, the nonlinear problem of three-degree-of-freedom data of the ship can be solved;
inputting the decomposed subsequence into a bidirectional long-short term memory network, the time characteristics of three degrees of freedom of the ship can be learned from the past direction and the future direction at the same time, the prediction accuracy is improved to a certain extent, and different from the prior art, the characteristics of original data can be learned in the forward direction and the reverse direction;
the hybrid particle swarm and the gravity search algorithm introduce the interaction force among the particles into the particle optimization process, guide the next motion direction of the particles, can more quickly search the number of hidden layer nodes and the learning rate of the optimal bidirectional long and short term memory, further improve the prediction performance of the bidirectional long and short term memory network, and different from the prior art, the performance of the bidirectional long and short term memory network can be greatly improved by using the hybrid particle swarm and the gravity search algorithm.
Drawings
FIG. 1 is a schematic diagram of the overall flow structure of the embodiment of the present invention.
Fig. 2 is a simplified flow chart of adaptive empirical wavelet transform according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of bi-directional long short term memory network prediction according to an embodiment of the present invention.
FIG. 4 is a schematic view of a process for optimizing the hyperparameter according to an embodiment of the present invention.
Detailed Description
The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and the accompanying drawings.
Referring to fig. 1, the three-degree-of-freedom neural network hybrid model prediction method for the ship based on adaptive empirical wavelet transform and hybrid hyper-parameter optimization in the embodiment of the present invention includes the following steps:
step 1: acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
when a ship sails on the sea, the ship is influenced by uncertain sea conditions such as sea wind, sea waves, ocean currents and the like, and the three-degree-of-freedom swinging motion inevitably occurs, and the three-degree-of-freedom swinging motion comprises the following steps: roll, pitch, and heave. Wherein, the rolling is the reciprocating rotation of the ship around an x axis, and the unit is degree; pitching is the reciprocating rotation of the ship around the y axis, and the unit is degree; heave is the reciprocating motion of the ship from side to side along the y-axis direction, and the unit is meter.
In specific implementation, corresponding ship shaking posture time sequences can be respectively generated for the three degrees of freedom, and then subsequent steps are respectively carried out.
Step 1, the time sequence of the ship shaking posture is as follows:
D＝[data1,data2,...,dataL]
l∈[1,L]
wherein D represents a time sequence of the ship shaking postures, datalThe ship shaking attitude data at the ith moment in the ship shaking attitude time sequence is a roll/pitch/heave degree of freedom value at the ith moment. In the embodiment, L is 1200, and L represents the number of acquisition moments;
step 2: carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed subsequences and form a subsequence matrix;
there are many methods for decomposing time series signals, such as Wavelet Transform (WT), Empirical Mode Decomposition (EMD), and Empirical Wavelet Transform (EWT). Among them, WT is a data decomposition and reconstruction method that decomposes original data into low-frequency wavelet coefficients and high-frequency wavelet coefficients using a low-pass filter and a high-pass filter, but since the size of the number of components needs to be manually set, the number of components cannot be optimally selected according to the data. With the development of decomposition algorithm, Huang et al proposed an EMD algorithm which can perform adaptive decomposition according to the original sequence, but the decomposition will have an end effect, i.e. the end of the signal cannot be at the maximum or minimum at the same time, so the upper and lower envelopes will diverge at both ends of the data sequence, and this divergence will gradually go inwards as the operation proceeds, so that the whole data sequence will be affected.
The invention has the core idea that an appropriate filter bank is selected according to the result of signal division, and an amplitude modulation-frequency modulation component with Fourier spectrum characteristics is extracted, so that the method not only has the advantage of WT enhanced local signal performance, but also has the advantage of EMD self-adaptive decomposition. Therefore, the invention proposes to preferably use the ship shaking posture time sequence to carry out the adaptive decomposition by an adaptive empirical wavelet transform method. Further, the invention provides an improvement on the existing adaptive empirical wavelet transform method: the variation rule of the signal frequency spectrum can be better captured by utilizing the maximum value to point and divide the frequency band interval, and the empirical mode components are subjected to Hilbert transform to obtain a decomposed sequence matrix. See, in particular, step 2.2 and step 2.5 below.
Referring to fig. 2, the adaptive decomposition by the adaptive empirical wavelet transform method described in step 2 in the embodiment is:
step 2.1: fourier transform is carried out on the ship shaking posture time sequence D to obtain a signal frequency spectrum F (omega), and the F (omega) represents a signal frequency spectrum with the angular frequency omega;
wherein the content of the first and second substances,
the Fourier transformation is carried out on the ship shaking posture time series D;
according to the shaking rule of the three-degree-of-freedom data of the ship, the invention further provides an optimal scheme: the ship attitude time sequence is subjected to Fourier transform, the sampling rate is set to be 1Hz, and the number of sampling points is set to be 1024 points.
Step 2.2: solving the second derivative of the signal frequency spectrum F (omega) to obtain the number P of maximum points of the F (omega), and dividing the frequency band interval according to each maximum point, wherein the division method is a one-to-one correspondence relationship, one maximum point corresponds to one frequency band interval and is marked as Λm，ΛmRepresents the m-th band segment;
the step utilizes the maximum value to dot-and-dash the sub-band interval, so that the change rule of the signal spectrum can be better captured. And solving a second derivative of the signal spectrum to obtain the number of maximum value points of the signal spectrum, and dividing each frequency band interval according to the number of maximum value points, wherein each frequency band interval corresponds to each subsequence, so that the process of self-adaptive decomposition is realized.
Step 2.3: frequency band interval lambda after signal divisionmConstructing an orthogonal wavelet filter bank, wherein the orthogonal wavelet filter bank comprises an empirical scale function and an empirical wavelet function;
said empirical scale function
And said empirical wavelet function
The definition is as follows:
wherein the content of the first and second substances,
an empirical scale function representing the frequency point omega in the mth frequency band interval,
an empirical wavelet function, ω, representing frequency points at ω within the mth band intervalmThe angular frequency of the mth frequency band interval is represented, β (x) represents a polynomial function, γ represents an empirical mode frequency band coefficient, and the empirical mode constraint condition is given as follows:
wherein x represents an independent variable, τmPolynomial function, min, representing the mth band intervalmThe minimum value of the m-th frequency band interval is taken;
step 2.4: extracting the signal frequency spectrum F (omega) through the wavelet filter bank obtained in the step 2.3 to obtain low-frequency components and high-frequency components, namely empirical mode components, wherein the empirical mode components comprise empirical scale components F0(t) and a finite number of empirical wavelet components fk(t)；
The extraction process is as follows:
wherein denotes a convolution, f0(t) represents the empirical scale component at time t; f. ofk(t) represents the kth empirical wavelet component at time t, detail coefficient Wf(k, t) and approximation coefficient Wf(0, t) is defined as follows:
wherein, Wf(k, t) represents the detail coefficient of the kth empirical wavelet component at time t, Wf(0, t) represents the approximation coefficient of the empirical scale component at time t;
f represents an empirical wavelet component;
τ represents an independent variable;
ψmrepresenting an empirical wavelet function within the mth band bin;
f (τ) represents the empirical wavelet component at the argument τ;
Step 2.5: performing Hilbert transform on the empirical mode component to obtain a decomposed sequence matrix;
by means of Hilbert transform, the narrow-band signal is subjected to envelope elimination, the instantaneous frequency of the signal is solved, and the phase of all frequency components of the signal is delayed by 90 degrees. The advantages of such a process are: the subsequence after the empirical wavelet decomposition is a narrow-band signal, and characteristic information contained in the signal can be more effectively and truly acquired by performing Hilbert transform on the subsequence.
The obtained decomposed sequence matrix by performing hilbert transform on the empirical mode component is as follows:
i＝1,2,…,N
l＝1,2,…,L
wherein S represents a decomposed sequence matrix, Si,lRepresenting the ith signal in the ith subsequence, N representing the number of subsequences, and L representing the length of each subsequence; f. ofi(t) represents the ith empirical wavelet component at time t;
indicating that the hilbert transform is performed on the ith empirical wavelet component at time t.
Step 3, decomposing the subsequence matrix Si,1,Si,2,...,Si,LDividing the length L of the combined row into a training set, a verification set and a test set according to the ratio of 8:1: 1;
the preferred scheme principle adopted by the embodiment of the invention is as follows:
training set: training an initial network requires a large amount of data, and the set amount is 0.8L;
and (4) verification set: optimizing the initial network, and setting the number of verification sets to be 0.1L;
and (3) test set: and evaluating the performance of the whole prediction model method, and setting the number of the test sets to be 0.1L.
In an embodiment, the training set in step 3 is defined as:
i∈[1,0.8L]
l1≤li≤l0.8L
1≤na≤N
wherein the content of the first and second substances,
represents the nth in the training setaThe number of sub-sequence samples is,
represents the nth in the training setaL in the sub-sequence sampleiThe freedom data of ship shaking, 0.8L represents the number of subsequence samples in the training set;
step 3 the verification set is defined as:
i∈[0.8L,0.9L]
l0.8L＜li≤l0.9L
1≤nb≤N
wherein the content of the first and second substances,
representing the nth of the authentication setbThe number of sub-sequence samples is,
representing the nth of the authentication setbL in the sub-sequence sampleiData of degree of freedom of ship shaking, (0.9L-0.8L) represents the number of subsequence samples in the verification set;
step 3, the test set is defined as:
i∈[0.9L,L]
l0.9L≤li≤lL
1≤nc≤N
wherein the content of the first and second substances,
indicating nth in test setcThe number of sub-sequence samples is,
indicating nth in test setcL in the sub-sequence sampleiData of degree of freedom of individual vessel rolling, (L-0.9L) represents the number of subsequence samples in the test set;
step 3, the training set is used for training an initial network model;
step 3, the verification set is used for hyper-parameter optimization;
and 3, the test set is used for testing the robustness of the model.
And 4, step 4: introducing a bidirectional long and short term memory network, sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, predicting to obtain a ship attitude sequence at a future moment through the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at a real time moment, and further performing optimization training on the neuron weight and the hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node numbers and a learning rate, so that the neuron weight and the hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
in the early stage of the ship motion prediction research, a forecasting method based on a linear hydrodynamic motion equation is firstly proposed. However, the method has low calculation efficiency, is complex, has low stability in practical application, and has difficulty in meeting practical requirements on prediction accuracy. According to the invention, a large number of experiments and comparisons are carried out, and the bidirectional long-short term memory network model trains the forward propagation long-short term memory network and the backward propagation bidirectional long-short term primary network together for three degrees of freedom of the ship, and extracts the past time characteristic and the future time characteristic of sample data, so that the time information can be processed in two opposite directions at the same time, and a better prediction effect can be obtained.
The bidirectional long-short term memory network trains the forward propagation long-short term memory network and the backward propagation long-short term memory network together, and extracts the past time characteristic and the future time characteristic of the sample data. The implementation can adopt the existing bidirectional long and short term memory network structure, and can be seen in the literature: M.Schuster and K.K.Paliwal, "Bidirective recovery neural networks," in IEEE Transactions on Signal Processing, vol.45, No.11, pp.2673-2681, Nov.1997.
Referring to fig. 3, the specific steps in step 4 in the embodiment are:
step 4.1: will train sample SnaInputting sample data from a moment a-1 to a moment a-T into the bidirectional long-short term memory network model, wherein T represents the number of iterations, and calculating the output value of each neuron of the forward LSTM, and recording the output value as
h represents a neuron output value;
step 4.2: will train the sample
Inputting sample data from a moment a to T to a moment a to 1 into a bidirectional long-short term memory network model, wherein T represents iteration times, and calculating the output value of each neuron of the inverse LSTM
Step 4.3: forward and reverse LSTM neuron output values obtained according to steps 4.1 and 4.2
And
the output value of the two-way long-short term memory network can be obtained and recorded as
Means that the forward and reverse output values of each neuron are added correspondingly;
step 4.4: calculating a loss function according to the output value of the bidirectional long and short term memory network obtained in the step 4.3, and recording the loss function as loss;
step 4.4 the calculation process is:
wherein, glThe first true value is represented by the first true value,
the output value of the first bidirectional long-short term memory network is shown, and L represents the sequence length.
Step 4.5: calculating according to the step 4.4 to obtain a loss function, and updating all weights based on a gradient descent algorithm;
in particular, the minimum value of the objective function may be found through iteration, or the minimum value may be converged (in an embodiment, the minimum value of the objective function is found).
The gradient descent algorithm is repeatedly utilized, the gradient is repeatedly solved, and finally the local minimum value can be reached.
Step 4.6: judging whether a preset iteration maximum value is reached, if not, turning to the step 4.1 to continue calculation, and if so, ending iteration and entering the step 4.7;
in specific implementation, the iteration maximum value can be preset by a user, and an empirical value can be adopted. The preferred embodiment is set to 100.
Step 4.7: finally, the predicted value of the bidirectional long-short term memory network can be obtained.
And 5: optimizing the super parameters of the initial bidirectional long-short term memory network model obtained in the step 4, wherein the super parameters comprise the number of hidden layer nodes and the learning rate, and the method comprises the step 3 of verifying the set
Inputting the data into a bidirectional long and short term memory network, optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into an initial bidirectional long and short term memory network model to obtain the optimized bidirectional long and short term memory network;
the traditional neural network also has some problems in parameter setting, for example, how to update network parameters in the training iteration process of the network so as to avoid the iteration from falling into local optimization. Therefore, many applications of the heuristic bionic optimization algorithm in the neural network become widespread, for example, a particle swarm-long and short term memory network hybrid prediction algorithm is used, but the local optimization capability of the particle swarm algorithm is poor, and the particle swarm algorithm is easy to fall into local minimum. The gravitation search algorithm utilizes the universal gravitation among objects to enable the particles with small inertial mass to move towards the particles with large inertial mass, and can be optimized to reach the optimal position, but the gravitation search algorithm is easy to generate premature and is not high in search accuracy. The gray wolf algorithm achieves the optimization purpose by simulating the predation behavior of the gray wolf colony and based on a wolf colony cooperation mechanism, and the algorithm has poor population diversity, low later convergence speed and easy trapping in local optimization.
Through a large amount of research and comparison, the optimization algorithm combining the gravity search algorithm and the particle swarm algorithm proposed by the reverse Mirjalii et al introduces the mechanism of the gravity search algorithm into the particle swarm algorithm, so that particles with large inertial mass move towards the globally optimal position, the problem that the particle swarm algorithm is easy to fall into the locally optimal position is solved, the defect that the gravity search algorithm is stagnant sometimes is overcome, and compared with the particle swarm algorithm, the gravity search algorithm and the Grey wolf algorithm, the hybrid optimization algorithm is high in convergence speed and high in optimization precision, and is particularly suitable for three-degree-of-freedom prediction of the ship.
Referring to fig. 4, the specific steps in step 5 in the embodiment are:
step 5.1: initializing a population, wherein the population quantity N and the particle dimension d are set, the position and the speed of the particle population are randomly generated, and then a gravity constant G, an inertial mass M and an acceleration A are initialized;
initializing a population, wherein the specific initialization process comprises the following steps: setting a population number N, wherein N is 10 in the embodiment; setting a particle dimension d, wherein d is 2 in the embodiment; randomly generating the position and the speed of the particle population; setting a gravitational constant G, wherein G is 1 in the embodiment; setting an inertial mass M, wherein M is 0 in the embodiment; setting an acceleration A, wherein A is 0 in the embodiment;
step 5.2: determining the searching range of the number of hidden layer nodes and the learning rate of the neural network;
in the invention, the number of hidden layer nodes represents the number of neurons of the bidirectional long-short term memory network. The selection of the number of hidden nodes is very important, the selection not only has great influence on the performance of the established neural network model, but also is a direct reason for the occurrence of 'overfitting' during training, but at present, no scientific and universal determination method exists in theory. The performance of the bidirectional long-term and short-term memory network can be improved from two aspects by optimizing the number of nodes of the hidden layer and the learning rate, which are respectively as follows: number of hidden layer nodes-improving prediction accuracy; learning rate-improving prediction efficiency (shortening prediction time); it is desirable to satisfy both high prediction accuracy and less prediction time.
Therefore, the invention breaks through the conventional technology and provides two super parameters of optimizing the number of hidden layer nodes and the learning rate.
Further, in the embodiment, it is preferable to provide:
implicit tier node number range: 10-100
Learning rate range: 0.0001-0.1.
Step 5.3: verification set through step 3
Sequentially inputting the prediction results obtained by the bidirectional long-short term memory network and recording the prediction results as
Calculating the fitness of each particle each time, recording the fitness as fitness, and obtaining the global optimal fitness and the worst fitness through size comparison, wherein the optimal fitness is fbestAs shown, best is best represented by best, and worst fitness is represented by fworstExpressed, worst, the fitness function is calculated as:
wherein x isjThe true value at the j-th time instant is indicated,
indicates the predicted value at the j-th time, j indicates the time, and M indicates the length of the predicted data.
Step 5.4: according to the optimal fitness fbestAnd worst fitness fworstThe gravity coefficient G, the inertial mass M and the acceleration A of the population are updated according to the positions, and the updating process is as follows:
wherein G is0The gravity coefficient at the initial time 0 is represented, e represents a natural index, alpha represents a descent coefficient, T represents the maximum iteration number, and G (T) represents the gravity coefficient at the time T; f. ofj(t) denotes the fitness value of the jth data at time t, Mi(t) represents the inertial mass of the jth datum at time t;
represents the Newton attraction of the j-th data at the time t in the d-dimension case,
represents the acceleration of the jth data at the time of t in the case of d dimension.
Step 5.5: calculating the position and the speed of the updated particles according to the updated gravity coefficient G, the inertial mass M and the acceleration A, wherein the specific calculation process comprises the following steps:
wherein the content of the first and second substances,
representing the particle velocity of the jth data at time t in d-dimension, c1Denotes the gravitational constant, r, of the previous moment1A random number indicating a previous time, 1 indicating a previous time, 2 indicating a current time, c2A gravitational constant, r, representing the current time2A random number indicating a current time;
represents the particle position of the jth data at the time of t under the d-dimension condition,
representing the individual extremum of the jth data at time t in the case of d dimension,
the global extreme value at the moment t under the d-dimension condition is represented, and g refers to the global; w tableShowing the inertia factor, wmaxRepresents the maximum inertia factor, wminRepresenting the minimum inertia factor.
Step 5.6: judging whether the convergence termination condition is met, if not, turning to the step 5.3 to continue searching, if so, ending iteration, and entering the step 5.7;
step 5.7: returning the optimal hidden layer node number and the learning rate of the bidirectional long-short term memory network.
Step 6: and evaluating the optimized bidirectional long-short term memory network. Inputting the test set data into the optimized bidirectional long-short term memory network for robustness test, and respectively utilizing evaluation indexes of root mean square error, average absolute error and average absolute percentage error to carry out error analysis on the prediction result and evaluate whether the prediction result is reliable or not.
If the prediction method is reliable, the hybrid neural network prediction model has good performance, can predict and output results based on the optimized bidirectional long-short term memory network, and can be subsequently used for automatic alarm and the like; if the prediction method is unreliable, the performance of the hybrid neural network prediction model is proved to be invalid.
In order to facilitate understanding of the technical effects of the present invention, the corresponding analysis results of the examples are provided as follows:
in specific implementation, a person skilled in the art can implement the automatic operation process by using a computer software technology, and a system device for implementing the method, such as a computer-readable storage medium storing a corresponding computer program according to the technical solution of the present invention and a computer device including a corresponding computer program for operating the computer program, should also be within the scope of the present invention.
In some possible embodiments, a three-degree-of-freedom hybrid neural network model prediction system for a ship is provided, which includes the following modules,
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
the second module is used for carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed sub-sequences and form a sub-sequence matrix so as to reduce the nonlinear non-stationarity of the ship three-degree-of-freedom motion;
the third module is used for dividing the subsequence matrix obtained by the decomposition of the second module into a training set, a verification set and a test set;
the fourth module is used for introducing a bidirectional long and short term memory network to learn the time characteristics of three degrees of freedom of the ship from the past and the future, and comprises the steps of sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, predicting and obtaining a ship attitude sequence at the future moment through the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at the real time moment, further performing optimization training on the neuron weight and the hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node numbers and a learning rate, and the neuron weight and the hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
the fifth module is used for inputting the verification set obtained by the third module into the initial bidirectional long-short term memory network obtained by the fourth module, optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into the initial bidirectional long-short term memory network model to obtain the optimized bidirectional long-short term memory network;
and a sixth module for evaluating the optimized bidirectional long-short term memory network of the fifth module, which comprises inputting the test set obtained by the third module into the optimized bidirectional long-short term memory network obtained by the fifth module for robustness test, performing error analysis on the prediction result, evaluating whether the prediction result is reliable, and predicting and outputting the result based on the optimized bidirectional long-short term memory network if the prediction result is reliable.
In some possible embodiments, a three-degree-of-freedom hybrid neural network model prediction system for a ship is provided, which includes a processor and a memory, where the memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute a three-degree-of-freedom hybrid neural network model prediction method for a ship as described above.
In some possible embodiments, a three-degree-of-freedom hybrid neural network model prediction system for a ship is provided, which includes a readable storage medium, where a computer program is stored on the readable storage medium, and when the computer program is executed, the three-degree-of-freedom hybrid neural network model prediction method for a ship is implemented.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Claims (10)
1. A hybrid neural network model prediction method of three degrees of freedom of a ship is characterized by comprising the following steps:
1) acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
2) carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed subsequences to form a subsequence matrix so as to reduce the nonlinear non-stationarity of the ship three-degree-of-freedom motion;
3) dividing the subsequence matrix into a training set, a verification set and a test set based on the subsequence matrix obtained by decomposing in the step 2);
4) the method comprises the steps of sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, obtaining a ship attitude sequence at a future moment through prediction of the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at a real time, and further performing optimization training on neuron weight and hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node number and learning rate, so that the neuron weight and hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
5) inputting the verification set obtained in the step 3) into the initial bidirectional long and short term memory network obtained in the step 4), optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into an initial bidirectional long and short term memory network model to obtain the optimized bidirectional long and short term memory network;
6) and evaluating the optimized bidirectional long and short term memory network in the step 5), wherein the test set obtained in the step 3) is input into the optimized bidirectional long and short term memory network obtained in the step 5 to carry out robustness test, error analysis is carried out on the prediction result, whether the prediction result is reliable or not is evaluated, and if the prediction result is reliable, the result is predicted and output based on the optimized bidirectional long and short term memory network.
2. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 1, wherein the method comprises the following steps: and 2), performing Fourier transform on the ship attitude time sequence according to the shaking rule of the ship three-degree-of-freedom data.
3. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 1, wherein the method comprises the following steps: in the step 2), the second derivative is solved for the signal spectrum, the number of maximum points of the signal spectrum is obtained, each frequency band interval is divided according to the number of maximum points, each frequency band interval corresponds to each subsequence, and the process of self-adaptive decomposition is realized.
4. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 1, wherein the method comprises the following steps: in the step 2), performing Hilbert transform on the empirical mode component to obtain a decomposed sequence matrix.
5. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 1, 2, 3 or 4, wherein: in step 3), the decomposed subsequence matrix is divided into a training set, a verification set and a test set according to a ratio of 8:1:1 by combining the length L of the rows.
6. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 1, 2, 3 or 4, wherein: in the step 5), the number range of the hidden layer nodes is set to be 10:100, and the learning rate range is set to be 0.0001-0.1.
7. The method for predicting the three-degree-of-freedom hybrid neural network model of the ship according to claim 5, wherein the method comprises the following steps: in the step 5), the number range of the hidden layer nodes is set to be 10:100, and the learning rate range is set to be 0.0001-0.1.
8. A hybrid neural network model prediction system of three degrees of freedom of a ship is characterized in that: the hybrid neural network model prediction method for realizing three degrees of freedom of a ship according to any one of claims 1 to 7.
9. The three-degree-of-freedom hybrid neural network model prediction system for the ship of claim 8, wherein: comprises the following modules which are used for realizing the functions of the system,
the system comprises a first module, a second module and a third module, wherein the first module is used for acquiring original ship shaking attitude data by an attitude sensor arranged on a ship, and decoding the original ship shaking attitude data through resampling to obtain a ship shaking attitude time sequence;
the second module is used for carrying out self-adaptive decomposition on the ship shaking posture time sequence by a self-adaptive empirical wavelet transform method to obtain a plurality of decomposed sub-sequences and form a sub-sequence matrix so as to reduce the nonlinear non-stationarity of the ship three-degree-of-freedom motion;
the third module is used for dividing the subsequence matrix obtained by the decomposition of the second module into a training set, a verification set and a test set;
the fourth module is used for introducing a bidirectional long and short term memory network to learn the time characteristics of three degrees of freedom of the ship from the past and the future, and comprises the steps of sequentially inputting each subsequence sample in a training set into the bidirectional long and short term memory network, predicting and obtaining a ship attitude sequence at the future moment through the bidirectional long and short term memory network, constructing a loss function model by combining the ship attitude sequence at the real time moment, further performing optimization training on the neuron weight and the hyper-parameters of the bidirectional long and short term memory network in a gradient descent training mode, wherein the hyper-parameters comprise hidden layer node numbers and a learning rate, and the neuron weight and the hyper-parameters of the updated bidirectional long and short term memory network are obtained and are used for initializing the bidirectional long and short term memory network to obtain an initial bidirectional long and short term memory network model;
the fifth module is used for inputting the verification set obtained by the third module into the initial bidirectional long-short term memory network obtained by the fourth module, optimizing the number of hidden layer nodes and the learning rate by using a hybrid optimization algorithm combining a gravity search algorithm and a particle swarm algorithm to obtain the optimized number of hidden layer nodes and the optimized learning rate, and inputting the optimized number of hidden layer nodes and the optimized learning rate into the initial bidirectional long-short term memory network model to obtain the optimized bidirectional long-short term memory network;
and a sixth module for evaluating the optimized bidirectional long-short term memory network of the fifth module, which comprises inputting the test set obtained by the third module into the optimized bidirectional long-short term memory network obtained by the fifth module for robustness test, performing error analysis on the prediction result, evaluating whether the prediction result is reliable, and predicting and outputting the result based on the optimized bidirectional long-short term memory network if the prediction result is reliable.
10. The three-degree-of-freedom hybrid neural network model prediction system for the ship of claim 8, wherein: the ship three-degree-of-freedom hybrid neural network model prediction method comprises a processor and a memory, wherein the memory is used for storing program instructions, and the processor is used for calling the stored instructions in the memory to execute the ship three-degree-of-freedom hybrid neural network model prediction method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111053868.4A CN113837454B (en) | 2021-09-09 | 2021-09-09 | Ship three-degree-of-freedom hybrid neural network model prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111053868.4A CN113837454B (en) | 2021-09-09 | 2021-09-09 | Ship three-degree-of-freedom hybrid neural network model prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113837454A true CN113837454A (en) | 2021-12-24 |
| CN113837454B CN113837454B (en) | 2024-04-23 |
Family
ID=78958732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111053868.4A Active CN113837454B (en) | 2021-09-09 | 2021-09-09 | Ship three-degree-of-freedom hybrid neural network model prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113837454B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114936669A (en) * | 2022-04-06 | 2022-08-23 | 武汉大学 | Hybrid ship rolling prediction method based on data fusion |
| CN115079576A (en) * | 2022-07-20 | 2022-09-20 | 西南科技大学 | Amplitude-frequency characteristic parameter acquisition method based on unmanned aerial vehicle recovery landing under ship shaking |
| CN115294674A (en) * | 2022-10-09 | 2022-11-04 | 南京信息工程大学 | Unmanned ship navigation state monitoring and evaluating method |
| 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 |
| CN116842474A (en) * | 2023-06-14 | 2023-10-03 | 青岛哈尔滨工程大学创新发展中心 | Ship motion extremely short-term forecasting method and system based on TFT model |
| CN117909665A (en) * | 2024-03-18 | 2024-04-19 | 青岛哈尔滨工程大学创新发展中心 | Ship motion envelope forecast data processing method and system based on Fourier filtering |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022471A (en) * | 2016-05-12 | 2016-10-12 | 大连海事大学 | Wavelet neural network model ship rolling real-time prediction method based on particle swarm optimization algorithm |
| CN111160668A (en) * | 2020-01-02 | 2020-05-15 | 中国矿业大学 | Method for improving accuracy of coal seam geographic information system |
| CN112100841A (en) * | 2020-09-09 | 2020-12-18 | 中铁二十局集团有限公司 | Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium |
| KR102204724B1 (en) * | 2020-11-19 | 2021-01-19 | 한국해양과학기술원 | Apparatus and method for controlling OWC(Oscillating Water Column)-type wave power generator system using digital twin technology |
| CN112463987A (en) * | 2020-12-09 | 2021-03-09 | 中国园林博物馆北京筹备办公室 | Chinese classical garden knowledge graph completion and cognitive reasoning method |
| CN112684701A (en) * | 2020-12-01 | 2021-04-20 | 哈尔滨工程大学 | Ship motion forecasting method based on long-time and short-time memory network and Gaussian process regression |
| WO2021082809A1 (en) * | 2019-10-29 | 2021-05-06 | 山东科技大学 | Training optimization method for foreign exchange time series prediction |
| CN112948969A (en) * | 2021-03-01 | 2021-06-11 | 哈尔滨工程大学 | Ship rolling prediction method based on LSTMC hybrid network |
| CN113156815A (en) * | 2021-03-10 | 2021-07-23 | 广东海洋大学 | Data-driven marine ship motion attitude real-time forecasting method |
-
2021
- 2021-09-09 CN CN202111053868.4A patent/CN113837454B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022471A (en) * | 2016-05-12 | 2016-10-12 | 大连海事大学 | Wavelet neural network model ship rolling real-time prediction method based on particle swarm optimization algorithm |
| WO2021082809A1 (en) * | 2019-10-29 | 2021-05-06 | 山东科技大学 | Training optimization method for foreign exchange time series prediction |
| CN111160668A (en) * | 2020-01-02 | 2020-05-15 | 中国矿业大学 | Method for improving accuracy of coal seam geographic information system |
| CN112100841A (en) * | 2020-09-09 | 2020-12-18 | 中铁二十局集团有限公司 | Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium |
| KR102204724B1 (en) * | 2020-11-19 | 2021-01-19 | 한국해양과학기술원 | Apparatus and method for controlling OWC(Oscillating Water Column)-type wave power generator system using digital twin technology |
| CN112684701A (en) * | 2020-12-01 | 2021-04-20 | 哈尔滨工程大学 | Ship motion forecasting method based on long-time and short-time memory network and Gaussian process regression |
| US20210117796A1 (en) * | 2020-12-01 | 2021-04-22 | Harbin Engineering University | Ship Motion Prediction Method Based on Long Short-Term Memory Network and Gaussian Process Regression |
| CN112463987A (en) * | 2020-12-09 | 2021-03-09 | 中国园林博物馆北京筹备办公室 | Chinese classical garden knowledge graph completion and cognitive reasoning method |
| CN112948969A (en) * | 2021-03-01 | 2021-06-11 | 哈尔滨工程大学 | Ship rolling prediction method based on LSTMC hybrid network |
| CN113156815A (en) * | 2021-03-10 | 2021-07-23 | 广东海洋大学 | Data-driven marine ship motion attitude real-time forecasting method |
Non-Patent Citations (2)
| Title |
|---|
| RENQIANG DING.ETC: ""A Novel Hybrid Model of Wind Speed Forecasting Based on EWT, BiLSTM, SVR Optimized by BOA in Inner Mongolia, China"", 《SPRINGER NATURE SINGAPORE PTE LTD. 2020》, pages 183 - 191 * |
| 王巨彪;赵晨;: "南海航海保障中心：踏浪前行助力粤港澳大湾区腾飞", 中国海事, no. 09 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114936669A (en) * | 2022-04-06 | 2022-08-23 | 武汉大学 | Hybrid ship rolling prediction method based on data fusion |
| CN114936669B (en) * | 2022-04-06 | 2024-05-17 | 武汉大学 | Mixed ship rolling prediction method based on data fusion |
| CN115079576A (en) * | 2022-07-20 | 2022-09-20 | 西南科技大学 | Amplitude-frequency characteristic parameter acquisition method based on unmanned aerial vehicle recovery landing under ship shaking |
| CN115294674A (en) * | 2022-10-09 | 2022-11-04 | 南京信息工程大学 | Unmanned ship navigation state monitoring and evaluating method |
| CN115294674B (en) * | 2022-10-09 | 2022-12-20 | 南京信息工程大学 | Unmanned ship navigation state monitoring and evaluating method |
| 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 |
| CN116842474A (en) * | 2023-06-14 | 2023-10-03 | 青岛哈尔滨工程大学创新发展中心 | Ship motion extremely short-term forecasting method and system based on TFT model |
| CN117909665A (en) * | 2024-03-18 | 2024-04-19 | 青岛哈尔滨工程大学创新发展中心 | Ship motion envelope forecast data processing method and system based on Fourier filtering |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113837454B (en) | 2024-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113837454B (en) | Ship three-degree-of-freedom hybrid neural network model prediction method and system | |
| CN106355151A (en) | Recognition method, based on deep belief network, of three-dimensional SAR images | |
| Kawai et al. | Sea state estimation using monitoring data by convolutional neural network (CNN) | |
| CN109901598A (en) | Autonomous underwater robot path following method based on stochastic model Prediction and Control Technology | |
| CN110472738A (en) | A kind of unmanned boat Real Time Obstacle Avoiding algorithm based on deeply study | |
| US11828894B2 (en) | Multi-scale unsupervised seismic velocity inversion method based on autoencoder for observation data | |
| CN113156815B (en) | Data-driven marine ship motion attitude real-time forecasting method | |
| CN103743402B (en) | A kind of underwater intelligent self adaptation Approach of Terrain Matching of topographic information based amount | |
| CN111199270A (en) | Regional wave height forecasting method and terminal based on deep learning | |
| CN111291505A (en) | Wing-type icing shape prediction method and device based on depth confidence network | |
| CN110209152A (en) | The deeply learning control method that Intelligent Underwater Robot vertical plane path follows | |
| CN114445634A (en) | Sea wave height prediction method and system based on deep learning model | |
| CN111027627A (en) | Vibration information terrain classification and identification method based on multilayer perceptron | |
| Barkhatov et al. | Effect of solar dynamics parameters on the formation of substorm activity | |
| Zhou et al. | Nonparametric modeling of ship maneuvering motions in calm water and regular waves based on R-LSTM hybrid method | |
| CN117104452A (en) | Method and system for inverting waves by ship-following swinging motion based on artificial neural network | |
| CN116341612A (en) | AUV drift track prediction method based on ABiLSTM-QSOA network | |
| Xu et al. | Accurate tsunami wave prediction using long short-term memory based neural networks | |
| Kumar et al. | Track prediction of tropical cyclones using long short-term memory network | |
| CN112948969B (en) | Ship rolling prediction method based on LSTMC hybrid network | |
| CN114580301A (en) | Hypersonic aircraft trajectory online prediction method based on parameter extrapolation | |
| CN113421281A (en) | Pedestrian micromotion part separation method based on segmentation theory | |
| Antonov | Comparative analysis of algorithms for ship pitching forecasting | |
| CN114625151B (en) | Underwater robot obstacle avoidance path planning method based on reinforcement learning | |
| Wang et al. | Semisupervised semantic segmentation for seismic interpretation |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |