CN111141483A - Intelligent method for generating malformed waves in water pool based on neural network self-learning - Google Patents

Intelligent method for generating malformed waves in water pool based on neural network self-learning Download PDF

Info

Publication number
CN111141483A
CN111141483A CN202010016991.8A CN202010016991A CN111141483A CN 111141483 A CN111141483 A CN 111141483A CN 202010016991 A CN202010016991 A CN 202010016991A CN 111141483 A CN111141483 A CN 111141483A
Authority
CN
China
Prior art keywords
wave
malformed
waves
neural network
numerical
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
Application number
CN202010016991.8A
Other languages
Chinese (zh)
Other versions
CN111141483B (en
Inventor
李焱
谢芃
尹天畅
赵志娟
夏妮
祁宣博
魏高龙
刘晓毅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN202010016991.8A priority Critical patent/CN111141483B/en
Publication of CN111141483A publication Critical patent/CN111141483A/en
Application granted granted Critical
Publication of CN111141483B publication Critical patent/CN111141483B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M10/00—Hydrodynamic testing; Arrangements in or on ship-testing tanks or water tunnels
    • 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

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Fluid Mechanics (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses an intelligent method for generating malformation waves in a pool based on neural network self-learning, which comprises the following steps: generating a wave spectrum target spectrum and a wave making machine motion amplitude target spectrum; generating a numerical value malformed wave by adopting a numerical value wave generating module of the malformed wave, and obtaining a preset motion time history of the wave generator; the wave making machine generates an evolving malformed wave in a simulation mode according to a preset time history of the wave making machine; the error of the evolving malformed wave and the numerical malformed wave is detected by adopting a malformed wave detection module to obtain the evolving malformed wave meeting the experimental requirements and a preset time history of the wave making machine motion, and the actually measured malformed wave is made in the wave pool according to the time history; repeating the steps to record a plurality of groups of wave surface time histories of the numerical malformed waves and the actually measured malformed waves to form a training sample library; training and testing the intelligent neural network self-learning system based on the BP neural network algorithm, and using the intelligent neural network self-learning system for wave generation and prediction of malformed waves. The invention solves the limitation of abnormal wave generation of the wave generator in the current laboratory pool, and optimizes the wave generation effect by adopting the neural network autonomous learning technology.

Description

Intelligent method for generating malformed waves in water pool based on neural network self-learning
Technical Field
The invention relates to ship and ocean engineering experimental technology, in particular to an intelligent method for generating malformed waves in a water pool based on neural network self-learning.
Background
A freak wave is a wave in the ocean having an irregular wave form, also known as an inland wave, an extraordinary wave, usually having an ultra-high wave height, and can be said almost without rules. The malformed waves are often generated in areas with developed sea shipping or rich oil and gas storage, and are easy to cause great harm to offshore structures. For example, for most marine structures moored for long periods at operating sea, which are in varying marine environments at the time, there is also a risk of being affected by freak waves. It is therefore necessary to analyse the dynamic response of marine engineering structures under the action of teratogenic waves, the primary task being the simulation of teratogenic waves.
Compared with random waves, the malformed waves are formed by concentrating wave energy at a certain moment and a certain position at a certain moment, and if the generation method of the random waves is adopted, the generation efficiency of the malformed waves is very low, so that some special numerical methods are usually adopted to simulate the malformed waves. Commonly used methods include a phase modulation method, a focused wave method, and the like. However, these methods are generally used in numerical analysis and are relatively rarely used in laboratory basins. Compared with numerical simulation, experimental simulation is particularly irreplaceable in position and advantage, and is necessary for practical engineering.
The wave generator program commonly used in the current ship and ocean engineering water pool only comprises the wave generating function of regular waves and random waves. If the traditional function is adopted to make the malformed wave, the waveform time history of the malformed wave can be accidentally made only by carrying out a plurality of times of analog adjustment. Therefore, the wave generation process can be optimized by adopting a certain numerical algorithm, and the efficient simulation of the malformed waves is realized. However, in the actual wave-making process, it is found that the wave height does not reach the wave height value preset by the numerical algorithm, and therefore, the existing wave-making method for the abnormal wave experiment needs to be improved.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, provides an intelligent method for generating abnormal waves in a water pool based on neural network self-learning, solves the limitation of the abnormal waves generated by a wave generator in a water pool in a current laboratory, optimizes the wave generating effect by adopting a neural network autonomous learning technology, and provides experimental basis for researching the dynamic response of a marine structure under the action of the abnormal waves.
The technical scheme adopted by the invention is as follows: an intelligent method for generating malformed waves in a pool based on neural network self-learning, which adopts a device comprising a wave generator and a wave height meter, and comprises the following steps:
step 1, selecting the height H of a sense wave according to an experimental scheme and the built-in random wave spectrum and the form of a malformed wave of a wave making machinesPeriod T, generating a wave spectrum target spectrum S1(f) Target spectrum S of motion amplitude of sum wave generator2(f)；
Step 2, according to the wave spectrum target spectrum S1(f) Generating numerical value abnormal waves by adopting a numerical value wave generating module of the abnormal waves based on randomly selected wave elements and randomly generated phase angles, counting relevant parameters of the numerical value abnormal waves, and obtaining the wave surface duration η (t) of the numerical value abnormal waves;
step 3, adopting a numerical wave generating module of the malformed waves to generate a target spectrum S of the motion amplitude of the wave generator2(f) Inverting to preset time history a (t) for the motion of the wave making machine;
step 4, inputting the preset motion time history a (t) of the wave maker into a built-in control module of the wave maker, simulating and generating the evolutionary malformed wave by adopting a built-in evolution module of the wave maker, and obtaining the wave surface time history η of the evolutionary malformed waves(t)；
Step 5, a malformed wave inspection module is adopted to inspect errors of the evolving malformed waves and the numerical malformed waves, and the evolving malformed waves meeting the experimental requirements and the preset motion time history of the wave making machine are obtained;
step 6, inputting the wave making machine motion preset time history meeting the experimental requirements into a built-in control module of the wave making machine, controlling the wave making machine to make an actually measured malformed wave in a wave water pool, and recording η the wave surface time history of the actually measured malformed wave in real time*(t) counting relevant parameters of the actually measured malformation waves;
7, reselecting wave elements and regenerating phase angles, repeating the steps 2 to 6, recording a plurality of groups of wave surface time histories of the numerical value abnormal waves and the actually measured abnormal waves, and forming a learning training sample library;
step 8, training an intelligent neural network self-learning system based on a BP neural network algorithm;
step 9, testing the trained intelligent neural network self-learning system;
and step 10, using the tested intelligent neural network self-learning system for wave generation and prediction of the malformed waves.
Further, in step 2, the numerical wave generating module of the abnormal wave is adopted to generate the numerical abnormal wave, and the relevant parameters of the numerical abnormal wave are counted, including:
step 2-1, target spectrum S of wave spectrum1(f) Dispersing the frequency of (a) into M parts to form a dispersed frequency set F;
step 2-2, focusing time t of the malformed waves according to a preset numerical valuecFocal position xcRandomly selecting a part of wave elements corresponding to the frequency in the frequency set F, and setting an initial phase angle of the wave elements corresponding to the selected frequency by adopting a random frequency phase angle modulation method
Figure BDA0002359264520000021
The phase angle of the wave element corresponding to the other frequencies is from 0,2 pi]Randomly selecting and uniformly distributing the intervals to form a phase angle set phi which corresponds to the frequency elements in the frequency set F one by one;
step 2-3, based on the Longuet-Higgins wave model, adopting a Fourier series superposition method to target the wave spectrum S1(f) Inverse directionGenerating a numerical malformation wave, and counting relevant parameters of the numerical malformation wave according to the formula (1) to the formula (4):
a1＝Hmax/Hs(1)
a2＝Hmax/H1(2)
a3＝Hmax/H2(3)
a4＝ηmax/Hmax(4)
in the formula, a1、a2、a3And a4Four judgment parameters of the numerical value malformation wave; hmaxIs the instantaneous maximum wave height; h1And H2The wave height of two waves adjacent to the peak of the numerical malformed wave, wherein H1Before the numerical malformation wave, H2After numerical malformation, ηmaxIs the height of the numerical value of the peak of the malformed wave.
Further, in step 5, the step of checking the error of the evolving freak wave and the numerical freak wave by using the freak wave checking module includes:
step 5-1, calculating a wave surface time history η of the evolving malformed waves(t) a peak deviation from a wavefront of a numerical freak of η (t);
step 5-2, comparing the peak deviation of the two waves with a preset threshold error of 1%, if the peak deviation of the two waves exceeds an acceptable threshold range, repeating the steps 2 to 4 to regenerate the numerical value freak wave and the evolution freak wave, and calculating η according to the step 5-1 to obtain the wave surface time of the regenerated evolution freak wavesAnd (t) comparing the peak deviation of η (t) with the regenerated time history of the numerical malformed wave, comparing the peak deviation of the regenerated time history of η (t) with a preset threshold error of 1%, and if the peak deviation of the regenerated time history of η (t) and the regenerated time history of the numerical malformed wave are within an acceptable threshold range, considering that the obtained evolving malformed wave and the preset time history of wave making machine motion meet the experimental requirements.
Further, in step 6, the real-time recording η of the wavefront of the measured malformed wave*(t) comprises: at the focus position x of the measured malformed wavecArranging a wave height instrument, and recording the wave of the actually measured abnormal wave in real time by adopting the wave height instrumentCalendar η*(t)；
The statistics of the relevant parameters of the actually measured malformed waves are as follows:
Figure BDA0002359264520000031
Figure BDA0002359264520000032
Figure BDA0002359264520000041
Figure BDA0002359264520000042
in the formula (I), the compound is shown in the specification,
Figure BDA0002359264520000043
and
Figure BDA0002359264520000044
four judgment parameters of the actually measured malformed waves;
Figure BDA0002359264520000045
η real-time recorded wave surface time of actually measured malformed wave for wave height instrument*(t) the corresponding wave height;
Figure BDA0002359264520000046
and
Figure BDA0002359264520000047
the wave heights of two waves adjacent to the wave crest of the actually measured malformed wave are shown, wherein,
Figure BDA0002359264520000048
prior to the actual measurement of the anomalous wave,
Figure BDA0002359264520000049
after the measured malformed wave;
Figure BDA00023592645200000410
the height of the actually measured deformed wave crest is obtained.
Further, in step 8, the training of the intelligent neural network self-learning system based on the BP neural network algorithm includes:
η is performed on the wave surface time histories of the n numerical value abnormal waves in the training sample library obtained in the step 7iN wave spectrum target spectrums S corresponding to the wave surface of 1,2, …, n, n numerical value malformed waves1i(f) I-1, 2, …, n, n frequency sets FiI-1, 2, …, n and n sets of phase angles ΦiPreset time history a of movement of 1,2, …, n, n wave-making machinesi(t), i is 1,2, …, n is used as the estimated control parameter, n wave surface time courses of n measured malformed waves corresponding to n wave making machine motion preset time courses are ηi *And (t), i is 1,2, …, n is used as a label and is input into the intelligent neural network self-learning system, the least square result of the pre-estimated control parameters and the label is used as a loss function, and the intelligent neural network self-learning system is trained through a gradient descent method.
Further, in step 9, the testing of the trained intelligent neural network self-learning system includes:
step 9-1, subjecting the wave surface time of the numerical malformed wave for inspection to ηc(t) and a wavefront time of numerical malformed wave ηc(t) corresponding wave spectrum target spectrum S for inspection1c(f) Frequency set F for testingcPhase set for testing phicThe numerical wave generating module input into the malformed wave obtains the preset motion time history a of the wave generator for inspectionc(t)；
Step 9-2, target spectrum S of wave spectrum for inspection1c(f) η wave surface time calendar of numerical malformed wave for inspectionc(t) frequency set F for testingcPhase set for testing phicWave making machine movement presetting time calendar a for inspectionc(t) inputting the trained intelligent neural network self-learning system, and obtaining the wave of the predicted malformed wave through the nonlinear mapping of the neural networkCalendar ηαc *(t)；
Step 9-3, presetting a time history a of the wave maker movement for inspectionc(t) inputting the wave generator to generate wave, and recording the measured abnormal wave duration η for inspection by using the wave height meterc *(t)；
Step 9-4, calculating η the wave surface time history of the predicted malformed waveαc *(t) and measured malformed wave duration η for verificationc *And (t) comparing the peak deviation with a preset threshold value in the malformed wave inspection module, repeating the step 8 to continue training the intelligent neural network self-learning system if the peak deviation of the two peak deviations exceeds an acceptable threshold value range, and considering that the intelligent neural network self-learning system passes the test at the moment if the peak deviation of the two peak deviations is within the acceptable threshold value range.
Further, in step 10, the self-learning system of the tested intelligent neural network is used for wave generation and prediction of the malformed wave, and includes:
after the wave height, the period and the appearance time and the position of the abnormal wave are set at the beginning of an experiment, a wave spectrum target spectrum is selected, a numerical wave generating module of the abnormal wave is adopted to generate a plurality of groups of frequency sets, phase sets and numerical abnormal wave surfaces, the generated frequency sets, phase sets and numerical abnormal wave surfaces are input into an intelligent neural network self-learning system passing through the test, the wave generating result is predicted, abnormal wave surface time histories meeting the experimental conditions of a physical model are screened, and subsequent wave generation and experiments are carried out.
The invention has the beneficial effects that:
1. the intelligent method for generating the malformed waves in the water pool based on the neural network self-learning is simple, efficient, intelligent and feasible.
2. The intelligent method for generating the malformed waves in the water pool based on the neural network self-learning realizes the prediction of the wave surface through the neural network self-learning technology, has high intelligent degree, is simple and easy to operate, is convenient to operate, reduces the difficulty of experimental operation, and improves the accuracy of the experiment.
3. In the past, wave making waves in a laboratory need to be debugged on a wave making machine so as to make a preset wave surface, and the time and economic cost are higher. According to the intelligent method for generating the abnormal waves in the water pool based on the neural network self-learning, the trained neural network prediction is adopted, the wave adjusting process can be performed on any computer in a preposed mode, the operation efficiency is greatly improved, and the economic cost is reduced.
4. According to the intelligent method for generating the abnormal waves in the water pool based on the neural network self-learning, the result of each wave generation can be used as a training sample used in a subsequent experiment, the neural network trained based on a large data volume sample can continuously improve the prediction precision of the neural network through a self-learning technology, the experiment efficiency is improved, and the experiment error is reduced.
5. According to the intelligent method for generating the abnormal wave in the water pool based on the neural network self-learning, disclosed by the invention, the database formed by the training samples can store a large number of abnormal wave samples for direct calling in subsequent experiments, and the intelligence is high.
Drawings
FIG. 1: the invention relates to a schematic diagram of an intelligent method for generating malformation waves in a pool based on neural network self-learning;
FIG. 2 a: target spectrum S of wave spectrum target1(f) A schematic diagram;
FIG. 2 b: wave making machine motion amplitude spectrum S2(f) A schematic diagram;
FIG. 3a is a schematic view of the surface history η (t) of a numerical malformed wave;
FIG. 3 b: a related malformed wave definition parameter schematic diagram;
FIG. 4: a schematic diagram of a (t) of the wave making machine during preset movement time;
FIG. 5 shows the wave surface time history η of the evolving malformed waves(t) schematic drawing;
FIG. 6 shows η measured wave surface time of the measured malformed wave generated by the wave generator*(t) schematic drawing;
the attached drawings are marked as follows: 1 is a wave making machine; 2 is a numerical wave generating module of the malformed waves; 3, abnormal wave; 4, a built-in control module of the wave making machine; 5 is a wave height instrument; 6 is a wave pool; 7 is an intelligent neural network self-learning system; and 8, a malformed wave inspection module.
Detailed Description
In order to further understand the contents, features and effects of the present invention, the following embodiments are illustrated and described in detail with reference to the accompanying drawings:
the invention provides an intelligent method for generating abnormal waves in a water pool based on neural network self-learning based on ocean engineering hydrodynamics, which realizes the generation of abnormal waves in a laboratory water pool, expands the functions of a laboratory wave generator and has important reference significance for ensuring dynamic response experimental tests of ocean structures under the action of abnormal waves developed in a laboratory. And a neural network self-learning technology is adopted, and control parameters are optimized, so that the wave generation meets the requirements.
As shown in fig. 1, the intelligent method for generating the malformed waves in the pool based on the neural network self-learning is used for generating the malformed waves 3 in a wave pool 6, and the adopted devices comprise a wave generator 1 and a wave height gauge 5, wherein the wave generator 1 can adopt a wave generator manufactured by 702 institute or university of technology, and is provided with a built-in wave generator control module 4; the method comprises the following steps:
step 1, selecting proper sense wave height H according to an experimental scheme and the built-in random wave spectrum and the form of the malformed wave 3 of the wave making machine 1sPeriod T, generating a wave spectrum target spectrum S1(f) Target spectrum S of motion amplitude of sum wave generator2(f) FIG. 2a shows the target spectrum S of the generated wave spectrum1(f) FIG. 2b shows the produced wave generator motion amplitude spectrum S2(f) And the abscissa f in the graph represents the frequency.
Step 2, according to the selected wave spectrum target spectrum S1(f) The numerical value wave generating module 2 of the abnormal wave is adopted to generate the numerical value abnormal wave based on the randomly selected wave elements and the randomly generated phase angle, the relevant parameters of the numerical value abnormal wave are counted, and meanwhile, the wave surface duration η (t) of the numerical value abnormal wave is obtained.
The working steps of the numerical wave generating module 2 for the malformed waves are as follows: using random frequency phase angle modulation method (see the influence of second order wave load on tension leg platform dynamic response under abnormal wave action in random frequency phase angle modulation method) and dynamic analysis of turbine-Moored FPSOSystem in Freek Wave), as an example, a Wave spectrum target spectrum S1(f) Dispersing the frequency of (a) into M parts to form a dispersed frequency set F; focusing time t of abnormal wave according to preset numerical valuecFocal position xcRandomly selecting a part of wave elements corresponding to the frequency in the frequency set F, and setting an initial phase angle of the wave elements corresponding to the selected frequency by adopting a random frequency phase angle modulation method
Figure BDA0002359264520000071
The phase angle of the wave element corresponding to the other frequencies is from 0,2 pi]Randomly selecting and uniformly distributing the intervals, thus forming a phase angle set phi which corresponds to the frequency elements in the frequency set F one by one; based on the Longuet-Higgins wave model, a Fourier series superposition method is adopted to obtain a wave spectrum target spectrum S1(f) Inversion generates numerical value malformation waves, FIG. 3a is a curve η (t) of the wave surface time of the numerical value malformation waves, and relevant parameters of the numerical value malformation waves are counted according to the formula (1) to the formula (4):
a1＝Hmax/Hs(1)
a2＝Hmax/H1(2)
a3＝Hmax/H2(3)
a4＝ηmax/Hmax(4)
in the formula, a1、a2、a3And a4Four judgment parameters of the numerical value malformation wave; as shown in FIG. 3b, HmaxIs the instantaneous maximum wave height, H1And H2The wave height of two waves adjacent to the peak of the numerical malformed wave, wherein H1Before the numerical malformation wave, H2After numerical malformation, ηmaxIs the height of the numerical value of the peak of the malformed wave.
Step 3, adopting a numerical wave generating module 2 of the abnormal waves to generate a target spectrum S of the motion amplitude of the wave generator according to the discrete frequency set F and the phase angle set phi2(f) Inverting to preset time history a (t) for the motion of the wave making machine; fig. 4 shows the wave generator movement preset time period a (t).
Step 4, will obtainThe wave maker movement preset time history a (t) is input into the wave maker built-in control module 4, an evolution module built in the wave maker 1 is adopted to generate an evolution malformed wave in an analog mode, and the wave surface time history η of the evolution malformed wave is obtaineds(t) as shown in FIG. 5.
And 5, adopting a malformed wave inspection module 8 to inspect errors of the evolving malformed waves and the numerical malformed waves to obtain the evolving malformed waves meeting the experimental requirements and the preset time history of the wave making machine motion.
Step 5-1, calculating a wave surface time history η of the evolving malformed waves(t) a peak deviation from a wavefront of a numerical freak of η (t);
step 5-2, comparing the peak deviation of the two waves with a preset threshold error of 1%, if the peak deviation of the two waves exceeds an acceptable threshold range, repeating the steps 2 to 4 to regenerate the numerical value freak wave and the evolution freak wave, and calculating η according to the step 5-1 to obtain the wave surface time of the regenerated evolution freak wavesAnd (t) comparing the peak deviation of the regenerated numerical value malformed wave with the peak deviation of η (t), and comparing the peak deviation of the regenerated numerical value malformed wave with a preset threshold error of 1%, wherein if the peak deviation of the regenerated numerical value malformed wave and the peak deviation of the regenerated numerical value malformed wave are within an acceptable threshold range, the evolutionary malformed wave and the wave making machine motion preset time a (t) obtained at the moment are considered to meet the experimental requirements.
Step 6, inputting the preset time a (t) of the wave maker motion meeting the experimental requirements into a built-in control module 4 of the wave maker to control the wave maker 1 to make actually measured malformed waves in a wave water pool 6; at the focus position x of the measured malformed wavecArranging a wave height instrument 5, recording the wave surface time of the actually measured malformed wave in real time η*(t), as shown in FIG. 6; and according to formulas (5) to (8), counting relevant parameters of the measured malformation waves:
Figure BDA0002359264520000081
Figure BDA0002359264520000082
Figure BDA0002359264520000083
Figure BDA0002359264520000084
in the formula (I), the compound is shown in the specification,
Figure BDA0002359264520000085
and
Figure BDA0002359264520000086
four judgment parameters of the actually measured malformed waves;
Figure BDA0002359264520000087
η real-time recorded wave surface time of actually measured malformed wave for wave height instrument 5*(t) the corresponding wave height;
Figure BDA0002359264520000088
and
Figure BDA0002359264520000089
the wave heights of two waves adjacent to the wave crest of the actually measured malformed wave are shown, wherein,
Figure BDA00023592645200000810
prior to the actual measurement of the anomalous wave,
Figure BDA00023592645200000811
after the measured malformed wave;
Figure BDA00023592645200000812
the height of the actually measured deformed wave crest is obtained.
Step 7, because the frequencies in the frequency set F in the step 2 are randomly selected, the frequencies selected each time are different, and the wave element and the phase angle are regenerated each time; and (6) repeating the steps 2 to 6, and recording a plurality of groups of wave surface time histories of the numerical malformed waves and the actually measured malformed waves to form a learning training sample library.
Step 8, based on BP neural network algorithmTraining an intelligent neural network self-learning system 7, wherein the intelligent neural network self-learning system 7 is an existing algorithm, and the wave surface time of the n numerical value abnormal waves in the training sample library obtained in the step 7 is ηiN wave spectrum target spectrums S corresponding to the wave surface of 1,2, …, n, n numerical value malformed waves1i(f) I-1, 2, …, n, n frequency sets FiI-1, 2, …, n and n sets of phase angles ΦiPreset time history a of movement of 1,2, …, n, n wave-making machinesi(t), i is 1,2, …, n is used as the estimated control parameter, n wave surface time courses of n measured malformed waves corresponding to n wave making machine motion preset time courses are ηi *And (t), i is 1,2, …, n is used as a label and is input into the intelligent neural network self-learning system 7, the least square result of the pre-estimated control parameters and the label is used as a loss function, and the intelligent neural network self-learning system 7 is trained through a gradient descent method. Therefore, the intelligent neural network self-learning system 7 is trained by adopting the method, and the wave-making structure with the given parameter can be predicted.
On the basis, the frequency set F and the phase set phi can be used as training labels, the actually measured wave surface profile is used as an estimated control parameter, the frequency set F and the phase set phi capable of creating the actually measured wave surface of the target abnormal wave can be predicted, the parameters can be given by the intelligent neural network self-learning system 7, and the required abnormal wave 3 can be directly created.
In addition, the result of each wave generation is stored in a training sample library to be used as a training sample used in a subsequent experiment, so that the neural network can be more fully learned and trained.
And 9, testing the trained intelligent neural network self-learning system 7, and checking the pre-estimation accuracy.
Step 9-1, subjecting the wave surface time of the numerical malformed wave for inspection to ηc(t) and a wavefront time of numerical malformed wave ηc(t) corresponding wave spectrum target spectrum S for inspection1c(f) Frequency set F for testingcPhase set for testing phicThe numerical value wave-making module 2 input into the malformed wave obtains the wave-making machine for inspectionDynamic preset time calendar ac(t)；
Step 9-2, target spectrum S of wave spectrum for inspection1c(f) η wave surface time calendar of numerical malformed wave for inspectionc(t) frequency set F for testingcPhase set for testing phicWave making machine movement presetting time calendar a for inspectionc(t) inputting the trained intelligent neural network self-learning system 7, and obtaining η wave surface time history of the predicted malformed waves through nonlinear mapping of the neural networkαc *(t)；
Step 9-3, presetting a time history a of the wave maker movement for inspectionc(t) inputting the wave maker 1 to make wave, and recording the measured abnormal wave duration η for inspection by the wave height meter 5c *(t)；
Step 9-4, calculating η the wave surface time history of the predicted malformed waveαc *(t) and measured malformed wave duration η for verificationc *And (t) comparing the peak deviation with a preset threshold value in the malformed wave inspection module 8, if the peak deviation of the two peak deviations exceeds an acceptable threshold value range, repeating the step 8 to continue training the intelligent neural network self-learning system 7, and if the peak deviation of the two peak deviations is within the acceptable threshold value range, considering that the intelligent neural network self-learning system 7 passes the test.
And step 10, after the intelligent neural network self-learning system 7 passes the test, the wave generation and prediction of the malformed wave 3 can be realized. After the wave height, the period and the appearance time and the position of the abnormal wave 3 are set at the beginning of an experiment, a wave spectrum target spectrum is selected, a numerical wave generation module 2 of the abnormal wave is adopted to generate a plurality of groups of frequency sets, phase sets and numerical abnormal wave surfaces, the generated frequency sets, phase sets and numerical abnormal wave surfaces are input into an intelligent neural network self-learning system 7 which passes a test, the wave generation result is predicted, abnormal wave surface time histories meeting the experimental conditions of a physical model (namely, a structural object model for the experiment, such as a scale model of a ship and an ocean platform) are screened, and subsequent wave generation and experiments are carried out.
Although the preferred embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and those skilled in the art can make many modifications without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims (7)

1. An intelligent method for generating malformed waves in a pool based on neural network self-learning is characterized in that the intelligent method comprises the following steps:
step 1, selecting the height H of a sense wave according to an experimental scheme and the forms of a random wave spectrum and a malformed wave (3) built in a wave making machine (1)sPeriod T, generating a wave spectrum target spectrum S1(f) Target spectrum S of motion amplitude of sum wave generator2(f)；
Step 2, according to the wave spectrum target spectrum S1(f) Adopting a numerical wave generation module (2) of the abnormal waves to generate numerical abnormal waves based on randomly selected wave elements and randomly generated phase angles, counting relevant parameters of the numerical abnormal waves, and obtaining a wave surface duration η (t) of the numerical abnormal waves;
step 3, adopting a numerical wave generating module (2) of the abnormal waves to generate a target spectrum S of the motion amplitude of the wave generator2(f) Inverting to preset time history a (t) for the motion of the wave making machine;
step 4, inputting the wave maker motion preset time history a (t) into a wave maker built-in control module (4), simulating and generating an evolution malformed wave by adopting an evolution module built in the wave maker (1), and obtaining a wave surface time history η of the evolution malformed waves(t)；
Step 5, a malformed wave inspection module (8) is adopted to inspect errors of the evolving malformed waves and the numerical malformed waves, so that the evolving malformed waves meeting the experimental requirements and the preset motion time history of the wave making machine are obtained;
step 6, inputting the wave making machine motion preset time history meeting the experimental requirements into a wave making machine built-in control module (4), controlling the wave making machine (1) to make an actually measured malformed wave in a wave water pool (6), and recording the wave surface time history of the actually measured malformed wave in real time η*(t) incorporationMeasuring relevant parameters of the measured malformed waves;
7, reselecting wave elements and regenerating phase angles, repeating the steps 2 to 6, recording a plurality of groups of wave surface time histories of the numerical value abnormal waves and the actually measured abnormal waves, and forming a learning training sample library;
step 8, training an intelligent neural network self-learning system (7) based on a BP neural network algorithm;
step 9, testing the trained intelligent neural network self-learning system (7);
and step 10, using the tested intelligent neural network self-learning system (7) for wave generation and prediction of the malformed waves (3).
2. The intelligent method for generating malformed waves in the pool based on the neural network self-learning as claimed in claim 1, wherein in the step 2, the numerical wave generating module (2) of the malformed waves is adopted to generate numerical malformed waves, and relevant parameters of the numerical malformed waves are counted, and the method comprises the following steps:
step 2-1, target spectrum S of wave spectrum1(f) Dispersing the frequency of (a) into M parts to form a dispersed frequency set F;
step 2-2, focusing time t of the malformed waves according to a preset numerical valuecFocal position xcRandomly selecting a part of wave elements corresponding to the frequency in the frequency set F, and setting an initial phase angle of the wave elements corresponding to the selected frequency by adopting a random frequency phase angle modulation method
Figure FDA0002359264510000021
The phase angle of the wave element corresponding to the other frequencies is from 0,2 pi]Randomly selecting and uniformly distributing the intervals to form a phase angle set phi which corresponds to the frequency elements in the frequency set F one by one;
step 2-3, based on the Longuet-Higgins wave model, adopting a Fourier series superposition method to target the wave spectrum S1(f) And (3) carrying out inversion to generate a numerical value malformation wave, and counting relevant parameters of the numerical value malformation wave according to the formula (1) to the formula (4):
a1＝Hmax/Hs(1)
a2＝Hmax/H1(2)
a3＝Hmax/H2(3)
a4＝ηmax/Hmax(4)
in the formula, a1、a2、a3And a4Four judgment parameters of the numerical value malformation wave; hmaxIs the instantaneous maximum wave height; h1And H2The wave height of two waves adjacent to the peak of the numerical malformed wave, wherein H1Before the numerical malformation wave, H2After numerical malformation, ηmaxIs the height of the numerical value of the peak of the malformed wave.
3. The intelligent pool malformation wave generation method based on neural network self-learning as claimed in claim 1, wherein in step 5, the error of the evolving malformation wave and the numerical malformation wave is verified by using the malformation wave verifying module (8), which comprises:
step 5-1, calculating a wave surface time history η of the evolving malformed waves(t) a peak deviation from a wavefront of a numerical freak of η (t);
step 5-2, comparing the peak deviation of the two waves with a preset threshold error of 1%, if the peak deviation of the two waves exceeds an acceptable threshold range, repeating the steps 2 to 4 to regenerate the numerical value freak wave and the evolution freak wave, and calculating η according to the step 5-1 to obtain the wave surface time of the regenerated evolution freak wavesAnd (t) comparing the peak deviation of η (t) with the regenerated time history of the numerical malformed wave, comparing the peak deviation of the regenerated time history of η (t) with a preset threshold error of 1%, and if the peak deviation of the regenerated time history of η (t) and the regenerated time history of the numerical malformed wave are within an acceptable threshold range, considering that the obtained evolving malformed wave and the preset time history of wave making machine motion meet the experimental requirements.
4. The intelligent method for generating malformed waves in the pool based on neural network self-learning as claimed in claim 1, wherein in step 6, the waves of the actually measured malformed waves are recorded in real timeCalendar η*(t) comprises: at the focus position x of the measured malformed wavecArranging a wave height instrument (5), and recording η the wave surface time of actually measured malformed waves in real time by adopting the wave height instrument (5)*(t)；
The statistics of the relevant parameters of the actually measured malformed waves are as follows:
Figure FDA0002359264510000031
Figure FDA0002359264510000032
Figure FDA0002359264510000033
Figure FDA0002359264510000034
in the formula (I), the compound is shown in the specification,
Figure FDA0002359264510000035
and
Figure FDA0002359264510000036
four judgment parameters of the actually measured malformed waves;
Figure FDA0002359264510000037
η real-time recorded wave surface time of actually measured malformed wave for wave height instrument (5)*(t) the corresponding wave height;
Figure FDA00023592645100000312
and
Figure FDA0002359264510000038
the wave heights of two waves adjacent to the wave crest of the actually measured malformed wave are shown, wherein,
Figure FDA0002359264510000039
prior to the actual measurement of the anomalous wave,
Figure FDA00023592645100000310
after the measured malformed wave;
Figure FDA00023592645100000311
the height of the actually measured deformed wave crest is obtained.
5. The intelligent pool malformation wave generation method based on neural network self-learning as claimed in claim 1, wherein in step 8, the intelligent neural network self-learning system is trained based on BP neural network algorithm, comprising:
η is performed on the wave surface time histories of the n numerical value abnormal waves in the training sample library obtained in the step 7iN wave spectrum target spectrums S corresponding to the wave surface of 1,2, …, n, n numerical value malformed waves1i(f) I-1, 2, …, n, n frequency sets FiI-1, 2, …, n and n sets of phase angles ΦiPreset time history a of movement of 1,2, …, n, n wave-making machinesi(t), i is 1,2, …, n is used as the estimated control parameter, n wave surface time courses of n measured malformed waves corresponding to n wave making machine motion preset time courses are ηi *And (t), i is 1,2, …, n is used as a label and is input into the intelligent neural network self-learning system (7), the least square result of the pre-estimated control parameters and the label is used as a loss function, and the intelligent neural network self-learning system (7) is trained through a gradient descent method.
6. The intelligent method for creating malformation waves in water pool based on neural network self-learning as claimed in claim 1, wherein in step 9, the testing of the trained intelligent neural network self-learning system comprises:
step 9-1, subjecting the wave surface time of the numerical malformed wave for inspection to ηc(t) and a wavefront time of numerical malformed wave ηc(t) corresponding wave spectrum target spectrum S for inspection1c(f) Frequency set F for testingcPhase set for testing phicThe numerical value wave generating module (2) input into the malformed wave obtains the preset motion time a of the wave generator for inspectionc(t)；
Step 9-2, target spectrum S of wave spectrum for inspection1c(f) η wave surface time calendar of numerical malformed wave for inspectionc(t) frequency set F for testingcPhase set for testing phicWave making machine movement presetting time calendar a for inspectionc(t) inputting the trained intelligent neural network self-learning system (7), and obtaining η wave surface time history of the predicted malformed wave through nonlinear mapping of the neural networkαc *(t)；
Step 9-3, presetting a time history a of the wave maker movement for inspectionc(t) inputting the wave generator (1) to generate waves, and recording an actually measured abnormal wave duration η for inspection by using a wave height instrument (5)c *(t)；
Step 9-4, calculating η the wave surface time history of the predicted malformed waveαc *(t) and measured malformed wave duration η for verificationc *And (t) comparing the peak deviation with a preset threshold value in the malformed wave inspection module (8), if the peak deviation of the two peak deviations exceeds an acceptable threshold value range, repeating the step 8 to continue training the intelligent neural network self-learning system (7), and if the peak deviation of the two peak deviations is within the acceptable threshold value range, considering that the intelligent neural network self-learning system (7) passes the test at the moment.
7. The intelligent method for generating malformed waves in a pool based on neural network self-learning as claimed in claim 1, wherein the step 10 of applying the tested intelligent neural network self-learning system for generating and predicting the malformed waves (3) comprises:
after the wave height, the period and the appearance time and the position of the abnormal wave (3) are set at the beginning of an experiment, a wave spectrum target spectrum is selected, a numerical wave generation module (2) of the abnormal wave is adopted to generate a plurality of groups of frequency sets, phase sets and numerical abnormal wave surfaces, the generated frequency sets, phase sets and numerical abnormal wave surfaces are input into an intelligent neural network self-learning system (7) passing the test, the wave generation result is predicted, abnormal wave surface histories meeting the experiment conditions of a physical model are screened, and subsequent wave generation and experiments are carried out.
CN202010016991.8A 2020-01-08 2020-01-08 Intelligent method for generating malformed waves in water pool based on neural network self-learning Active CN111141483B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010016991.8A CN111141483B (en) 2020-01-08 2020-01-08 Intelligent method for generating malformed waves in water pool based on neural network self-learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010016991.8A CN111141483B (en) 2020-01-08 2020-01-08 Intelligent method for generating malformed waves in water pool based on neural network self-learning

Publications (2)

Publication Number Publication Date
CN111141483A true CN111141483A (en) 2020-05-12
CN111141483B CN111141483B (en) 2020-12-01

Family

ID=70523940

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010016991.8A Active CN111141483B (en) 2020-01-08 2020-01-08 Intelligent method for generating malformed waves in water pool based on neural network self-learning

Country Status (1)

Country Link
CN (1) CN111141483B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114003847A (en) * 2021-09-24 2022-02-01 大连理工大学 Generation modulation method and spectrum correction method for malformed wave
CN114624002A (en) * 2022-03-15 2022-06-14 浙江大学 Intelligent wave generation method based on machine learning

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6608559B1 (en) * 1997-04-18 2003-08-19 Jerome H. Lemelson Danger warning and emergency response system and method
US20090070282A1 (en) * 2005-09-26 2009-03-12 Lockheed Martin Corporation Method and system of monitoring prognostics
CN103065044A (en) * 2012-12-20 2013-04-24 江苏科技大学 Method for simulating freak waves under fractal sea surface background
CN108805100A (en) * 2018-06-25 2018-11-13 大连理工大学 Based on the distorted wave short-term earthquake prediction method of wave group characteristic and application under two-dimentional sea situation
CN208688774U (en) * 2018-08-06 2019-04-02 宁波傲天阔自动化设备有限公司 A kind of multiple spot curve wave making epparatus
CN109696292A (en) * 2017-10-20 2019-04-30 交通运输部天津水运工程科学研究所 The wave simulation method, apparatus and wave making epparatus of pond wave maker
CN109799065A (en) * 2019-01-14 2019-05-24 哈尔滨工业大学 A method of wave theory is made based on second order and generates sequential focusing wave
CN109946763A (en) * 2019-03-27 2019-06-28 大连理工大学 A kind of distorted wave short-term earthquake prediction method based on wave group identification

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6608559B1 (en) * 1997-04-18 2003-08-19 Jerome H. Lemelson Danger warning and emergency response system and method
US20090070282A1 (en) * 2005-09-26 2009-03-12 Lockheed Martin Corporation Method and system of monitoring prognostics
CN103065044A (en) * 2012-12-20 2013-04-24 江苏科技大学 Method for simulating freak waves under fractal sea surface background
CN109696292A (en) * 2017-10-20 2019-04-30 交通运输部天津水运工程科学研究所 The wave simulation method, apparatus and wave making epparatus of pond wave maker
CN108805100A (en) * 2018-06-25 2018-11-13 大连理工大学 Based on the distorted wave short-term earthquake prediction method of wave group characteristic and application under two-dimentional sea situation
CN208688774U (en) * 2018-08-06 2019-04-02 宁波傲天阔自动化设备有限公司 A kind of multiple spot curve wave making epparatus
CN109799065A (en) * 2019-01-14 2019-05-24 哈尔滨工业大学 A method of wave theory is made based on second order and generates sequential focusing wave
CN109946763A (en) * 2019-03-27 2019-06-28 大连理工大学 A kind of distorted wave short-term earthquake prediction method based on wave group identification

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
DONG-JIING DOONG 等: "Development of a warning model for coastal freak wave occurrences using an artificial neural network", 《OCEAN ENGINEERING》 *
YANFEI DENG 等: "An experimental study on deterministic freak waves: Generation, propagation and local energy", 《OCEAN ENGINEERING》 *
薛亚东 等: "基于雷达测波的畸形波短时预报机制探究", 《舰船电子工程》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114003847A (en) * 2021-09-24 2022-02-01 大连理工大学 Generation modulation method and spectrum correction method for malformed wave
CN114003847B (en) * 2021-09-24 2024-04-30 大连理工大学 Method for generating and modulating malformed wave and method for correcting spectrum
CN114624002A (en) * 2022-03-15 2022-06-14 浙江大学 Intelligent wave generation method based on machine learning
CN114624002B (en) * 2022-03-15 2023-02-14 浙江大学 Intelligent wave generation method based on machine learning

Also Published As

Publication number Publication date
CN111141483B (en) 2020-12-01

Similar Documents

Publication Publication Date Title
CN111141483B (en) Intelligent method for generating malformed waves in water pool based on neural network self-learning
Quigley et al. Review and assessment of frequency-based fatigue damage models
KR102052289B1 (en) Method and apparatus for detecting equivalent load of wind turbine
CN110160789B (en) GA-ENN-based wind turbine generator bearing fault diagnosis method
Ducrozet et al. Influence of varying bathymetry in rogue wave occurrence within unidirectional and directional sea-states
Fonseca et al. Numerical and experimental analysis of extreme wave induced vertical bending moments on a FPSO
McGovern et al. Pneumatic long-wave generation of tsunami-length waveforms and their runup
Dezvareh Application of Soft Computing in the Design and Optimization of Tuned Liquid Column–Gas Damper for Use in Offshore Wind Turbines
Samadani et al. Model-based fault diagnostics of nonlinear systems using the features of the phase space response
Huang et al. Semi-empirical single realization and ensemble crest distributions of long-crest nonlinear waves
CN114386175B (en) Ship local fatigue hotspot damage rate forecasting method
Durante et al. Accurate experimental benchmark study of a catamaran in regular and irregular head waves including uncertainty quantification
WO2022187898A1 (en) Pipeline anomaly detection method and system
Sclavounos et al. Offshore wind turbine nonlinear wave loads and their statistics
CN114781709A (en) Wind measurement data interpolation method and device and computer readable storage medium
CN107335908A (en) The automation random device found for feature and its use in repetitive process
Thomsen et al. Experimental testing of moorings for large floating wave energy converters
Jha et al. Motions of a spar buoy in random seas: Comparing predictions and model test results
CN112465136A (en) Hydroelectric generating set degradation prediction method and system
CN114624002B (en) Intelligent wave generation method based on machine learning
He et al. Predicting the breaking onset of wave groups in finite water depths based on the Hilbert-Huang transform method
Klein et al. Application of machine learning for the generation of tailored wave sequences
Chen Hydrodynamic analysis and optimization of a hinged-type wave energy converter-SeaWEED
Canard et al. Experimental reproduction of an extreme sea state in two wave tanks at various generation scales
Skitt et al. Process monitoring using auto-associative, feed-forward artificial neural networks

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