CN114936413A - Ship body shape optimization neural network modeling method and ship body shape optimization method - Google Patents

Ship body shape optimization neural network modeling method and ship body shape optimization method Download PDF

Info

Publication number
CN114936413A
CN114936413A CN202210421252.6A CN202210421252A CN114936413A CN 114936413 A CN114936413 A CN 114936413A CN 202210421252 A CN202210421252 A CN 202210421252A CN 114936413 A CN114936413 A CN 114936413A
Authority
CN
China
Prior art keywords
neural network
ship
training
sample data
control points
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
CN202210421252.6A
Other languages
Chinese (zh)
Other versions
CN114936413B (en
Inventor
唐滨
董韬
李宝君
黄礼敏
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.)
Harbin Engineering University
Original Assignee
Harbin Engineering 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 Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202210421252.6A priority Critical patent/CN114936413B/en
Publication of CN114936413A publication Critical patent/CN114936413A/en
Application granted granted Critical
Publication of CN114936413B publication Critical patent/CN114936413B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/15—Vehicle, aircraft or watercraft design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/08—Fluids
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Fluid Mechanics (AREA)
  • Algebra (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Automation & Control Theory (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Computational Mathematics (AREA)
  • Feedback Control In General (AREA)

Abstract

A hull shape optimization neural network modeling method and a hull shape optimization method relate to the field of ship design. Aiming at the problems that the agent model established by the traditional BP neural algorithm is established based on the most basic neural network structure, so that the convergence speed is low and the convergence can be realized only by iterating for many times, the invention provides the following scheme: the modeling method of the ship hull shape optimization mental network comprises the following steps: selecting a plurality of control points representing the characteristics of the ship body in the ship body to be optimized; changing the coordinates of the control points to obtain new control points as sample data; acquiring the resistance of the ship corresponding to the control point; and (3) constructing a network model, using the control point coordinate group committee input layer neuron and the corresponding ship resistance as training data, sampling and extracting part of sample data for training, extracting the rest of sample data for training, training all the sample data, and finishing the training. The method is suitable for optimizing the shape of the ship body in the ship manufacturing process.

Description

Ship body shape optimization neural network modeling method and ship body shape optimization method
Technical Field
Relate to the ship design field, concretely relates to hull appearance design technical field of boats and ships.
Background
The ship resistance is an important factor influencing the design work of the ship and is directly related to the technical and economic indexes of the ship. The existing common design method is mainly used for carrying out performance prediction on a ship body remodeling scheme by applying a CFD numerical simulation technology according to ship model test data, but the CFD numerical simulation calculation amount is extremely large, the calculation speed is slow, and aiming at the problem, a BP neural network can be adopted to establish an agent model, so that the calculation amount is reduced.
With the continuous development of the computer field in the years, the method of computer-based machine learning has also developed very rapidly. While the traditional empirical formula has low calculation precision and low required workload, the calculation amount of the CFD simulation with high precision in the optimization field is too large.
At present, the traditional BP neural network model is commonly used in the field of ship type design to replace CFD simulation numerical calculation, and a proxy model established by the traditional BP neural network algorithm is established on the basis of the most basic neural network structure, so that the problem of low convergence speed generally exists, and the convergence can be realized only by iterating for many times. In the field of optimizing the shape of a ship hull, the convergence speed of a neural network model is also one of the key problems to be solved.
Disclosure of Invention
The invention provides an optimized neural network, which is used for optimizing the appearance of a ship body, and specifically comprises the following steps of (1) establishing an agent model based on the most basic neural network structure aiming at the traditional BP neural network algorithm, so that the problems of low convergence speed and convergence only after iteration for many times generally exist:
a hull form optimization neural network modeling method comprises the following steps:
the method comprises the following steps: selecting a plurality of control points representing the characteristics of the ship body in the original ship shape;
step two: changing the coordinates of the control points for multiple times to obtain multiple groups of new control points which respectively correspond to the multiple ship types and serve as sample data;
step three: acquiring the ship resistance of each ship type by adopting a CFD (computational fluid dynamics) technology;
step four: constructing a BP neural network model, taking the control point coordinates of each group as input layer neurons, corresponding the ship resistance acquired in the third step for each group of control points, and taking the control point coordinates of each group and the corresponding ship resistance as training data;
step five: setting an error function for the BP neural network model;
step six: initializing parameters in the BP neural network model;
step seven: adopting Latin hypercube sampling to extract part of sample data;
step eight: training part of the sample data extracted in the seventh step by adopting a BP neural network model to obtain a training result, and calculating an error;
step nine: judging whether the error of the training result obtained in the step eight meets the requirement, if not, returning to the step eight, and if so, performing the step ten;
step ten: adopting Latin hypercube sampling to extract partial sample data of the residual sample data extracted in the seventh step;
step eleven: training part of the sample data extracted in the step ten by adopting a BP neural network model, and calculating an error;
step twelve: judging whether the error of the training result obtained in the step eleven meets the requirement, if not, returning to the step eleven, and if so, performing the step thirteen;
step thirteen: and training all sample data by adopting a BP neural network model until the error meets the requirement, and finishing the training.
Preferably, the selection mode of the control point in the first step is as follows: selecting coordinate points uniformly distributed in the original ship type as control points, wherein the control points comprise three pieces of coordinate information: coordinate information in the ship length direction, coordinate information in the ship width direction, and coordinate information in the ship depth direction.
Preferably, the number of the groups of the control points obtained in the second step is 100.
Preferably, in the seventh step, the number of the extracted partial sample data accounts for 20% -30% of the total number of the samples.
Preferably, in the step eight, the training is specifically performed in the following manner: and iterating the data extracted in the step eight for 50-100 times by adopting the BP neural network.
Preferably, in the step ten, the percentage of the extracted sample data in the remaining sample data extracted in the step seven is:
(30+70×(2-2×Sigmoid(|10×E/E MAX |)))％，
wherein E represents the error calculated in step eight, E MAX Sigmoid represents the activation function for a preset error.
Preferably, the activation function of the BP neural network is a Sigmoid function.
Preferably, the error function of the BP neural network adopts an mse function.
Based on the same inventive concept, the invention also provides a ship hull shape optimization method, which comprises the following steps:
training: establishing a neural network model;
the training steps are specifically as follows: the hull appearance optimization neural network modeling method;
optimizing: and searching for the optimal ship model through the neural network model established in the training step.
Preferably, the optimizing step specifically comprises:
and optimizing the trained BP neural network by adopting an artificial bee colony algorithm to find out the optimal ship type.
The invention has the advantages that:
the input of the neural network is optimized, the traditional input mode is changed, the input quantity of the sample is controlled, the purpose of accelerating the ship type optimization speed is finally achieved, and the efficiency is improved.
The improved neural network algorithm further accelerates the calculation speed, replaces CFD calculation with overlarge calculation amount, accelerates the speed of ship type optimization modeling, and improves the efficiency.
In the process of solving the ship model design problem, the training neural network model is improved, the artificial bee colony algorithm is applied to search the optimal ship model corresponding to the optimal result, the efficient artificial bee colony algorithm is adopted to search the optimal solution, and the speed of the whole optimization process is further improved.
The method is suitable for optimizing the appearance of the ship body in the ship manufacturing process.
Drawings
FIG. 1 is a flow chart of a hull form optimization neural network modeling method according to an embodiment;
fig. 2 is a flowchart of a hull form optimization method according to a ninth embodiment;
fig. 3 is a flowchart of an artificial bee colony algorithm provided in the tenth embodiment.
Detailed Description
The first embodiment is described with reference to fig. 1, and the first embodiment provides a hull contour optimization neural network modeling method, which includes:
the method comprises the following steps: selecting a plurality of control points representing the characteristics of the ship body in the original ship shape;
step two: changing the coordinates of the control points for multiple times to obtain multiple groups of new control points which respectively correspond to the multiple ship types and serve as sample data;
step three: acquiring the ship resistance of each ship type by adopting a CFD (computational fluid dynamics) technology;
step four: constructing a BP neural network model, taking the control point coordinates of each group as input layer neurons, corresponding the ship resistance acquired in the third step for each group of control points, and taking the control point coordinates of each group and the corresponding ship resistance as training data;
step five: setting an error function for the BP neural network model;
step six: initializing parameters in the BP neural network model;
step seven: adopting Latin hypercube sampling to extract part of sample data;
step eight: training part of the sample data extracted in the seventh step by adopting a BP neural network model to obtain a training result, and calculating an error;
step nine: judging whether the error of the training result obtained in the step eight meets the requirement, if not, returning to the step eight, and if so, performing the step ten;
step ten: adopting Latin hypercube sampling to extract partial sample data of the residual sample data extracted in the seventh step;
step eleven: training part of the sample data extracted in the step ten by adopting a BP neural network model, and calculating errors;
step twelve: judging whether the error of the training result obtained in the step eleven meets the requirement, if not, returning to the step eleven, and if so, performing the step thirteen;
step thirteen: and training all sample data by adopting a BP neural network model until the error meets the requirement, and finishing the training.
It is noted here that: the first to thirteenth steps mentioned in the present embodiment are merely names of steps, and are not used to limit the actual operation sequence of the present embodiment.
In particular, the method comprises the following steps of,
firstly, selecting coordinate points uniformly distributed on an original ship model as control points to represent the whole ship model, and recording a control point set as S ═ S (S) 1 ,s 2 ,s 3 ,...,s n ) Wherein each element in S contains three location information: s i ＝(x 1i ,x 2i ,x 3i ) Wherein x is 1i Represents the coordinates of the ith point in the ship length direction,x 2i represents the coordinate of the ith point in the width direction of the ship, wherein x 3i Represents the coordinates of the ith point along the molding depth direction.
The control points are distributed on the original ship type, the hull profile of the original ship type can be accurately obtained through all the control points, the hull profile can be changed by changing all or control point coordinates, and a plurality of groups of control points are obtained to represent different ship types.
A series of changes are made to the hull profile of the original ship shape through commercial software or a curved surface deformation technology, the steps mainly include that the profiles of the bow and the stern are changed to a certain extent, and a Latin hypercube sampling method is adopted to guarantee the stability of data. After n-1 new ship body shapes are obtained, a series of new control point sets S can be obtained 2 ,S 3 ,S 4 ,...,S n 。
Resistance was calculated for the newly formed profile using CFD techniques: calculating ship resistance under different ship types by using software such as NSYS Fluent, STAR-CCM, comsol, OpenFOAM and the like by using CFD technology for a series of ship types generated newly, wherein the resistance data set of each ship type is recorded as L ═ L (L 1 ,l 2 ,l 3 ,...,l n )。
Packaging data: and recording the data of the resistance and the data of the hull appearance in a csv file as training data of the neural network.
Constructing a bp neural network, wherein the number of neurons in an input layer of the bp neural network is set to be 3, and the bp neural network corresponds to three position information x contained in all control points of a ship 1 ，x 2 ，x 3 And a resistance value l corresponding to the ship type. The hidden layer is set to be 3 layers, and yi is the output corresponding to the ith hidden layer.
The activation function of the network is set as Sigmoid function:
Figure RE-GDA0003764145160000051
where σ denotes the number of neurons calculated by the activation function, x ij J input representing i hidden layerThe value is obtained.
The error function is the mse function:
Figure RE-GDA0003764145160000052
wherein L is the accurate resistance value obtained by CFD calculation, G (S) is the resistance value predicted by the neural network, C represents the variance between L and G (S), and E represents the minimum value of the error obtained by the calculation of the mes function.
Initializing parameters: the learning rate is set to lr which is 0-1]Random number of (a), weight matrix and bias matrix w ij Is set to [0,1 ]]I represents the number of hidden layers, and j represents the jth neuron corresponding to the hidden layer.
And randomly extracting 20-30% of data in the total data set by adopting Latin hypercube sampling, and recording as a first extraction step.
And iterating the extracted data for 50-100 times in the neural network, wherein the formula for updating the weight w and the bias b in each step is as follows:
Figure RE-GDA0003764145160000053
w 'represents the first-order momentum of the weight, b' represents the first-order momentum of the bias, t represents the iteration number, t is 0 initially, t is t +1 after each cycle is finished, and the error E is calculated.
And judging whether the error meets the requirement, if not, returning to the first extraction step, and if so, continuing to operate.
Randomly extracting a total data set by adopting Latin hypercube sampling:
(30+70×(2-2×Sigmoid(|10×E/E MAX |)))％，
inheriting the trained parameters of the neural network, inputting a new sample and continuing training as a second extraction step;
wherein, E MAX Representing the maximum value of the error in the neural network calculation process.
And judging whether the error reaches the specified error, if not, returning to the second extraction step, and if so, continuing to operate.
All samples are input for training, and the calculation can be stopped until the error precision requirement is met or the maximum iteration number is reached.
The maximum iteration number is set according to actual conditions, and usually, training is not completed when the maximum iteration number is reached, so that the set error precision is proved to be too high, and the error precision needs to be adjusted.
In the second embodiment, the method for modeling the hull form optimization neural network provided in the first embodiment is further defined, and the selection mode of the control points in the first step is as follows: selecting coordinate points uniformly distributed in the original ship type as control points, wherein the control points comprise three pieces of coordinate information: coordinate information in the ship length direction, coordinate information in the ship width direction, and coordinate information in the ship depth direction.
In a third embodiment, the number of the sets of control points obtained in the second step is 100.
In a fourth embodiment, the number of the extracted partial sample data accounts for 20% to 30% of the total number of the samples in the seventh step.
In the fifth embodiment, the hull form optimization neural network modeling method provided in the first embodiment is further limited, and in the eighth step, the training is specifically performed in the following manner: and iterating the data extracted in the step eight for 50-100 times by adopting the BP neural network.
In a sixth embodiment, the present embodiment is further limited to the method for modeling a ship hull contour optimization neural network provided in the first embodiment, in the tenth step, the percentage of the extracted sample data in the remaining sample data extracted in the seventh step is as follows:
(30+70×(2-2×Sigmoid(|10×E/E MAX |)))％，
wherein E represents the error calculated in step eight, E MAX Sigmoid represents the activation function for a preset error.
Seventh, in this embodiment, the hull form optimization neural network modeling method provided in the first embodiment is further limited, and the activation function of the BP neural network is a Sigmoid function.
In an eighth embodiment, the modeling method for the hull form optimization neural network provided in the first embodiment is further limited, and the mse function is used as the error function of the BP neural network.
Ninth, the present embodiment is described with reference to fig. 2, and provides a hull form optimizing method, including:
training: establishing a neural network model;
the training steps are specifically as follows: the method comprises the following steps of (1) implementing a ship hull shape optimization neural network modeling method;
optimizing: and searching for the optimal ship model through the neural network model established in the training step.
Tenth embodiment, the present embodiment is described with reference to fig. 3, and the present embodiment is a ship hull shape optimization method provided in the ninth embodiment, where the optimizing steps specifically include:
and optimizing the trained BP neural network by adopting an artificial bee colony algorithm to find out the optimal ship type.
The optimal ship type is the ship type with the minimum resistance.
In particular, the method comprises the following steps of,
and after the BP neural network training is finished, obtaining a neural network model f (S), wherein S is a ship control point set and is used for representing the predicted ship type. Calculating the optimal solution of the BP neural network model by using an artificial bee colony algorithm, and initializing a honey source X ═ X 1 ,X 2 ,L,X N ]，X N ∈(L d ,U d ) Wherein, L d Denotes the lower bound, U, of the search space d Representing the upper bound of the search space, each honey source represents a feasible solution, and each element within X is a feasible solution. Is provided withThe maximum search times of the honey source is limit, and the maximum iteration times of the algorithm is t max . Honey source X N Initial position X of N ＝L d +rand(0,1)(U d -L d ) The number of iterations t is 1.
Starting a searching step: and (4) allocating a leading bee for each honey source, and searching near the honey source.
And judging the fitness of the area near the honey source, wherein the fitness fit is 1/(1+ f (S)), and determining the reserved area as a new honey source according to a greedy selection method.
The probability of following the peak is determined by the fitness, and the higher the fitness is, the higher the following probability is. When the follower bees choose to follow, honey sources are searched near the followed leading bees. The non-selection is kept still when following.
And if the maximum iteration times are reached and no new honey source is found, searching the new honey source by adopting the scout bee with large step length and converting the scout bee into the leading bee for searching.
When all bees finish the action t as t +1, judging whether t reaches the maximum iteration time t max If the optimal ship type and the resistance of the ship are reached, stopping the operation, and outputting the optimal solution X, namely the optimal ship type and the resistance of the ship, otherwise, turning to the step of starting searching.
In an eleventh embodiment, the present invention provides a computer storage medium for storing a computer program, wherein when the computer runs the computer, the computer executes the hull contour optimization neural network modeling method provided in any one of the first to sixth embodiments.
A twelfth embodiment provides a computer storage medium storing a computer program, wherein when the computer runs the storage medium, the computer executes the hull form optimizing method according to any one of the ninth to the tenth embodiments.
In a thirteenth embodiment, the present embodiment provides a specific practice for the hull form optimization method provided in the ninth embodiment, for comparing with the prior art, specifically:
the Ferry boat type was used, and the boat type parameters are as follows:
TABLE 1 Ship major dimension information Table
Figure RE-GDA0003764145160000081
Selecting 400 control points on the ship, sorting coordinates, and then transforming to form 119 new ship types;
after the resistance value corresponding to the ship is calculated by CFD software, 100 samples are selected, the Ferry ship model is optimized by respectively adopting the existing neural network and the neural network mentioned in the technical scheme provided by the invention, and the training time of the obtained neural network is as follows:
TABLE 2 comparison of training times for different neural networks
Number of training samples Training time(s)
Primitive neural network 100 1921.52
Improved neural network 100 1637.47
According to the table, under the condition that the requirements of computer configuration, calculation software, calculation precision and the like are the same, the time consumed by the neural network model trained by the technical scheme provided by the invention is 14.8 percent less than that consumed by the prior art.
The final ship types obtained after optimization are compared as follows:
TABLE 3 comparison of different neural network predictions
Resistance value of original ship Speed of flight Predicting minimum resistance value
Primitive neural network 198KN 16 section 189.67KN
Improved neural network 198KN 16 section 189.65KN
The trained neural network is optimized by using an artificial bee colony algorithm, so that the predicted resistance values of the two neural networks are very close to each other under the condition of a certain navigational speed, and the convergence capability of the improved neural network model is proved to be that the calculation speed of the neural network is improved and the time is saved on the premise of ensuring the effect of the prior art.
The technical solutions provided by the present invention are further described in detail through several specific embodiments, so that the advantages and benefits of the technical solutions provided by the present invention can be embodied more specifically, but the above embodiments are not intended to limit the present invention, and any modifications, combinations, improvements, equivalents, and the like of the present invention based on the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The ship hull shape optimization neural network modeling method is characterized by comprising the following steps:
the method comprises the following steps: selecting a plurality of control points representing the characteristics of the ship body in the original ship shape;
step two: changing the coordinates of the control points for multiple times to obtain multiple groups of new control points which respectively correspond to the multiple ship types and serve as sample data;
step three: acquiring the ship resistance of each ship type by adopting a CFD (computational fluid dynamics) technology;
step four: constructing a BP neural network model, taking the control point coordinates of each group as input layer neurons, corresponding the ship resistance acquired in the third step for each group of control points, and taking the control point coordinates of each group and the corresponding ship resistance as training data;
step five: setting an error function for the BP neural network model;
step six: initializing parameters in the BP neural network model;
step seven: sampling part of the sample data by adopting Latin hypercube sampling;
step eight: training part of the sample data extracted in the seventh step by adopting a BP neural network model to obtain a training result, and calculating an error;
step nine: judging whether the error of the training result obtained in the step eight meets the requirement, if not, returning to the step eight, and if so, performing the step ten;
step ten: adopting Latin hypercube sampling to extract partial sample data of the residual sample data extracted in the seventh step;
step eleven: training part of the sample data extracted in the step ten by adopting a BP neural network model, and calculating errors;
step twelve: judging whether the error of the training result obtained in the step eleven meets the requirement, if not, returning to the step eleven, and if so, performing the step thirteen;
step thirteen: and training all sample data by adopting a BP neural network model until the error meets the requirement, and finishing the training.
2. The hull form optimization neural network modeling method of claim 1, wherein the control points in step one are selected in a manner that: selecting coordinate points uniformly distributed in the original ship form as control points, wherein the control points comprise three pieces of coordinate information: coordinate information in the ship length direction, coordinate information in the ship width direction, and coordinate information in the ship depth direction.
3. The hull form optimization neural network modeling method of claim 1, wherein the number of the groups of control points obtained in the second step is 100.
4. The hull form optimization neural network modeling method of claim 1, wherein in step seven, the number of the extracted partial sample data accounts for 20% -30% of the total number of the sample data.
5. The hull form optimization neural network modeling method according to claim 1, wherein in the step eight, the training is specifically performed in a manner that: and iterating the data extracted in the step eight for 50-100 times by adopting the BP neural network.
6. The hull form optimization neural network modeling method according to claim 1, wherein in the tenth step, the percentage of the extracted sample data to the remaining sample data extracted in the seventh step is:
(30+70×(2-2×Sigmoid(|10×E/E MAX |)))％，
wherein E representsError calculated in step eight, E MAX Sigmoid represents the activation function for a preset error.
7. A computer storage medium storing a computer program, wherein when the storage medium is executed by a computer, the computer executes the hull form optimization neural network modeling method of any one of claims 1-6.
8. The hull shape optimization method is characterized by comprising the following steps:
training: establishing a neural network model;
the training steps are specifically as follows: the hull form optimizing neural network modeling method of claim 1;
optimizing: and searching for the optimal ship model through the neural network model established in the training step.
9. The hull form optimizing method according to claim 9, characterized in that the optimizing step is specifically:
and optimizing the trained BP neural network by adopting an artificial bee colony algorithm to find out the optimal ship type.
10. A computer storage medium for storing a computer program, wherein when the storage medium is executed by a computer, the computer executes the hull form optimizing method according to any one of claims 8 to 9.
CN202210421252.6A 2022-04-21 2022-04-21 Ship body appearance optimization neural network modeling method and ship body appearance optimization method Active CN114936413B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210421252.6A CN114936413B (en) 2022-04-21 2022-04-21 Ship body appearance optimization neural network modeling method and ship body appearance optimization method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210421252.6A CN114936413B (en) 2022-04-21 2022-04-21 Ship body appearance optimization neural network modeling method and ship body appearance optimization method

Publications (2)

Publication Number Publication Date
CN114936413A true CN114936413A (en) 2022-08-23
CN114936413B CN114936413B (en) 2023-06-06

Family

ID=82861608

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210421252.6A Active CN114936413B (en) 2022-04-21 2022-04-21 Ship body appearance optimization neural network modeling method and ship body appearance optimization method

Country Status (1)

Country Link
CN (1) CN114936413B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115859485A (en) * 2023-02-27 2023-03-28 青岛哈尔滨工程大学创新发展中心 Streamline seed point selection method based on ship appearance characteristics

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107633301A (en) * 2017-08-28 2018-01-26 广东工业大学 The training method of testing and its application system of a kind of BP neural network regression model
US20200160061A1 (en) * 2017-12-11 2020-05-21 Zhuhai Da Hengqin Technology Development Co., Ltd. Automatic ship tracking method and system based on deep learning network and mean shift
CN111506968A (en) * 2020-04-21 2020-08-07 常熟理工学院 Ship type optimization method based on BP neural network algorithm
CN111619755A (en) * 2020-06-09 2020-09-04 中国船舶科学研究中心 Hull profile design method based on convolutional neural network
US20210157962A1 (en) * 2017-09-08 2021-05-27 Ecole Polytechnique Fédérale De Lausanne Epfl-Tto Shape optimisation of technical devices via gradient descent using convolutional neural network proxies

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107633301A (en) * 2017-08-28 2018-01-26 广东工业大学 The training method of testing and its application system of a kind of BP neural network regression model
US20210157962A1 (en) * 2017-09-08 2021-05-27 Ecole Polytechnique Fédérale De Lausanne Epfl-Tto Shape optimisation of technical devices via gradient descent using convolutional neural network proxies
US20200160061A1 (en) * 2017-12-11 2020-05-21 Zhuhai Da Hengqin Technology Development Co., Ltd. Automatic ship tracking method and system based on deep learning network and mean shift
CN111506968A (en) * 2020-04-21 2020-08-07 常熟理工学院 Ship type optimization method based on BP neural network algorithm
CN111619755A (en) * 2020-06-09 2020-09-04 中国船舶科学研究中心 Hull profile design method based on convolutional neural network

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115859485A (en) * 2023-02-27 2023-03-28 青岛哈尔滨工程大学创新发展中心 Streamline seed point selection method based on ship appearance characteristics

Also Published As

Publication number Publication date
CN114936413B (en) 2023-06-06

Similar Documents

Publication Publication Date Title
CN107239825B (en) Deep neural network compression method considering load balance
CN110245741A (en) Optimization and methods for using them, device and the storage medium of multilayer neural network model
CN104751842B (en) The optimization method and system of deep neural network
CN111859790B (en) Intelligent design method for curve reinforcement structure layout based on image feature learning
CN112000772B (en) Sentence-to-semantic matching method based on semantic feature cube and oriented to intelligent question and answer
CN109918663A (en) A kind of semantic matching method, device and storage medium
CN103729694B (en) The method that improvement GA based on polychromatic sets hierarchical structure solves Flexible workshop scheduling
CN112947300A (en) Virtual measuring method, system, medium and equipment for processing quality
CN110688502A (en) Image retrieval method and storage medium based on depth hash and quantization
CN112000770A (en) Intelligent question and answer oriented sentence-to-sentence matching method based on semantic feature map
CN114492191A (en) Heat station equipment residual life evaluation method based on DBN-SVR
CN111832839B (en) Energy consumption prediction method based on sufficient incremental learning
CN113269312B (en) Model compression method and system combining quantization and pruning search
CN114880806A (en) New energy automobile sales prediction model parameter optimization method based on particle swarm optimization
CN114936413A (en) Ship body shape optimization neural network modeling method and ship body shape optimization method
CN113537365A (en) Multitask learning self-adaptive balancing method based on information entropy dynamic weighting
CN109284388B (en) Text classification method and storage medium for character-number unique translatable depth model
CN113255235B (en) Approximate modeling method, device, equipment and medium for complex structure of aircraft
GB2617741A (en) Multi-level multi-objective automated machine learning
CN113868765A (en) Ship main scale parameter optimization method based on approximate model
CN114818128B (en) Modeling method and optimizing method for ship body local curved surface optimizing neural network
CN113807496A (en) Method, apparatus, device, medium and program product for constructing neural network model
CN117421989A (en) Agent model-assisted parallel collaboration method for high-dimensional expensive optimization problem
CN110399619B (en) Position coding method for neural machine translation and computer storage medium
CN109740221B (en) Intelligent industrial design algorithm based on search tree

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