CN112464366A - Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining - Google Patents

Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining Download PDF

Info

Publication number
CN112464366A
CN112464366A CN202011314168.1A CN202011314168A CN112464366A CN 112464366 A CN112464366 A CN 112464366A CN 202011314168 A CN202011314168 A CN 202011314168A CN 112464366 A CN112464366 A CN 112464366A
Authority
CN
China
Prior art keywords
fidelity
point
underwater vehicle
autonomous underwater
model
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
CN202011314168.1A
Other languages
Chinese (zh)
Other versions
CN112464366B (en
Inventor
王鹏
刘杰
宋保维
潘光
董华超
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.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical 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 Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202011314168.1A priority Critical patent/CN112464366B/en
Publication of CN112464366A publication Critical patent/CN112464366A/en
Application granted granted Critical
Publication of CN112464366B publication Critical patent/CN112464366B/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]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • 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
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Geometry (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Automation & Control Theory (AREA)
  • Algebra (AREA)
  • Computing Systems (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an autonomous underwater vehicle multi-fidelity shape optimization method based on data mining, which comprises the steps of firstly determining the design variable range, constraint conditions and objective functions of shape optimization of an underwater vehicle; then designing and sampling to establish initial agent models with high and low fidelity, preliminarily updating the agent models with low fidelity by using a method of improving expectation to the maximum extent, then carrying out data mining and knowledge extraction on the agent models with low fidelity, adding the obtained useful data into a high fidelity database for updating the agent models with high fidelity on one hand, and building a local confidence domain on the other hand, and inducing the agent models with high fidelity to strengthen the local exploration strength near the confidence domain, thereby enabling the agent models to be converged in the global optimal solution quickly. Meanwhile, a method for improving expectation in a maximized mode is adopted, the high-fidelity proxy model is updated in a point mode to conduct global exploration, the situation that the target performance is trapped in local optimization is avoided, meanwhile, the calculation cost is reduced, and the optimization efficiency is improved.

Description

Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining
Technical Field
The invention belongs to the technical field of underwater vehicles, and particularly relates to an underwater vehicle shape optimization method.
Background
An Autonomous Underwater Vehicle (AUV) refers to a small Autonomous navigation Vehicle which can be used for submarine exploration, remote mine laying and Underwater operations, is generally a recoverable Underwater unmanned operation platform which can navigate Underwater for a long time, and plays an important role in civil use and military use. The design of the AUV hydrodynamic appearance is an important link, and the hydrodynamic performance of the AUV hydrodynamic appearance is the basis of other technologies and plays a decisive role in the overall performance of an aircraft.
The appearance design of the traditional autonomous underwater vehicle mainly selects a master model with similar appearance types to carry out preliminary design by virtue of experience of designers, and then, appearance parameters are continuously and manually modified until the hydrodynamic performance meets the requirement of a design task.
According to the traditional autonomous underwater vehicle shape optimization method based on gradient optimization, on one hand, the gradient solving difficulty of a high-dimensional nonlinear non-smooth objective function is high, even the gradient cannot be solved, on the other hand, the optimization iteration needs to call a large amount of computational fluid force simulation, so that the calculation cost of the optimization is very expensive, even the optimization is difficult to carry out.
In order to overcome the defect of gradient Optimization, in recent years, a proxy model Optimization (SBO) method is applied, and the main idea of the proxy model is to construct the proxy model for complex and time-consuming numerical simulation or emulation in the Optimization process, so as to replace the calling of the complex numerical simulation in the sub-Optimization process, thereby greatly improving the design efficiency of Optimization and reducing the Optimization difficulty.
The proxy model optimization method reduces the calling times of the fluid simulation calculation to a certain extent, but still needs to call more high-precision numerical simulation to generate more sample point data to ensure the precision of the constructed proxy model. The main idea of Multi-fidelity Optimization is to reduce the number of calls to a high-fidelity proxy model with high precision but expensive calculation cost by adopting a corresponding Multi-fidelity Optimization strategy and calling a large number of low-fidelity proxy models with low precision but low calculation cost, so that the calculation precision can still meet the requirement of Optimization design while the calculation cost is reduced.
In the traditional Bridge Function (Bridge Function) multi-fidelity optimization, proxy models are respectively established for low-fidelity data and high-fidelity data, then the high-fidelity proxy models and the low-fidelity proxy models are fused, the result of the low-fidelity proxy models is mapped to the result of the high-fidelity proxy models, and one Bridge Function multi-fidelity proxy model is established for optimization. The method can meet the precision requirement and reduce the times of high-precision numerical simulation calling in practical application, but the gradient information of the high-and-low-fidelity proxy model needs to be calculated when the bridge function is constructed by the bridge function multi-fidelity proxy model method, and the error of the gradient is solved and the error of constructing the high-and-low-fidelity proxy model is added, so that the error between the approximate value of the multi-fidelity model to the high-fidelity proxy model and the real value of the high-fidelity model is larger.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an autonomous underwater vehicle multi-fidelity shape optimization method based on data mining, which comprises the steps of firstly determining the design variable range, constraint conditions and objective functions of the shape optimization of an underwater vehicle; then designing and sampling to establish initial agent models with high and low fidelity, preliminarily updating the agent models with low fidelity by using a method of improving expectation to the maximum extent, then carrying out data mining and knowledge extraction on the agent models with low fidelity, adding the obtained useful data into a high fidelity database for updating the agent models with high fidelity on one hand, and building a local confidence domain on the other hand, and inducing the agent models with high fidelity to strengthen the local exploration strength near the confidence domain, thereby enabling the agent models to be converged in the global optimal solution quickly. Meanwhile, a method for improving expectation in a maximized mode is adopted, the high-fidelity proxy model is updated in a point mode to conduct global exploration, the situation that the target performance is trapped in local optimization is avoided, meanwhile, the calculation cost is reduced, and the optimization efficiency is improved.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1: setting an optimized design variable and range, a constraint condition and an objective function of the autonomous underwater vehicle;
with adjustable parameters q of autonomous underwater vehicle headh1、qh2And tail curve segment linear adjustable parameter qt1、qt2For design variables, the head fullness, the caudal vertebra half angle and the diameter of the caudal facet are taken as constraint conditions, the minimum zero lift resistance of the autonomous underwater vehicle is taken as an optimization target, and an optimized mathematical model is as follows:
Figure BDA0002790788350000021
wherein psiHThe constraint condition of the head fullness coefficient of the autonomous underwater vehicle, alpha is the constraint condition of the caudal vertebra half angle of the autonomous underwater vehicle, DEThe tail end face diameter of the autonomous underwater vehicle, Ub and Lb are respectively the upper limit and the lower limit of the whole design space, psi*、α*、D*Respectively representing a head fullness coefficient threshold value of the autonomous underwater vehicle, a caudal vertebra half angle of the autonomous underwater vehicle and a tail end face diameter threshold value of the autonomous underwater vehicle;
respectively completing parametric geometric modeling, grid division and flow field analysis and calculation of the autonomous underwater vehicle, and realizing automatic appearance and resistance calculation flow creation of the autonomous underwater vehicle;
step 2: respectively evaluating high-fidelity sample points and low-fidelity sample points of the autonomous underwater vehicle by adopting an optimized Latin hypercube method, and respectively constructing high-fidelity agent model databases and low-fidelity agent model databases of the autonomous underwater vehicle according to design variables and resistance response values of the sample points;
and step 3: adopting a genetic algorithm to maximize the improvement expectation value of the low-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the evaluated new sample point into a low-fidelity proxy model database, and updating the low-fidelity proxy model;
and 4, step 4: performing surface data mining on the low-fidelity proxy model by adopting a multi-starting-point optimization method to find out all local optimal point positions in the low-fidelity proxy model;
and 5: removing points with the distance smaller than a set threshold value and completely repeated points from the data acquired in the mining process of the layer data in the step 4, and realizing the noise reduction of the local optimal point data;
step 6: screening out expected points in the local optimal point data subjected to noise reduction in the step 5 through a useful data evaluation algorithm, and removing repeated data;
and 7: carrying out fluid resistance evaluation on the expected points obtained in the step 6, adding the expected points passing the evaluation into a high fidelity proxy model database, and updating the high fidelity proxy model;
and 8: adopting a genetic algorithm to maximize the improvement expectation value of the high-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the new sample point which passes the evaluation into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
and step 9: further screening out a point with the minimum resistance response value as a most promising point from the evaluated promising points in the step 7, constructing a local confidence domain around the most promising point, minimizing a predicted value of the high-fidelity proxy model in the local confidence domain by adopting a genetic algorithm so as to obtain a new sample point and evaluate the fluid resistance, adding the evaluated new sample point into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
step 10: optimizing the maximum improvement expectation and the minimum agent model prediction simultaneously by adopting a genetic algorithm for the low-fidelity agent model, performing a de-duplication data processing process on the obtained new sample point and the most promising point obtained in the step 9, evaluating the fluid resistance, adding the evaluated point into a low-fidelity agent model database, and updating the low-fidelity agent model;
step 11: judging whether the current optimal value of the multi-fidelity shape optimization of the autonomous underwater vehicle meets an iteration termination condition formula (2), and if so, performing step 12; if not, returning to the step 4;
Figure BDA0002790788350000041
wherein x isi+1、xiRespectively, the optimal point of the current iteration and the optimal point of the last iteration, yhighIs a resistance response value;
step 12: and outputting the multi-fidelity shape optimization value of the autonomous underwater vehicle meeting the iteration termination condition, and finishing the optimization.
Preferably, the method for eliminating the points with the distance smaller than the set threshold in the local optimal points in step 5 is as follows:
point X of the local optimum pointsiAnd point XjDistance between themijIf the distance is smaller than the set Threshold, the distance is considered to be the adjacent point or the repeated point, and the distance judgment expression is as follows:
Figure BDA0002790788350000042
preferably, the useful data evaluation algorithm in step 6 is:
respectively using the point with the maximum improvement expectation value of the high-fidelity proxy model, the point with the maximum improvement expectation value of the low-fidelity proxy model and the point with the minimum prediction function as the desired points;
the expression for the expected improvement value is:
Figure BDA0002790788350000043
wherein
Figure BDA0002790788350000044
Predicting minimum value for proxy model, s is variance of proxy model, yminPhi () is the cumulative distribution function of the standard normal distribution function for the optimal value of the current sample point database,
Figure BDA0002790788350000045
is a probability density function of a standard normal distribution function.
Preferably, the method for constructing the local confidence domain in step 9 is as follows:
a local search range is constructed near the most promising point mined and screened from the low-fidelity agent model, the high-fidelity model is induced to quickly approach the global true optimal solution, and the confidence domain expression is as follows:
Figure BDA0002790788350000046
where delta is the confidence radius, where,
Figure BDA0002790788350000047
for the most promising point, w is the scaling factor, and Tr _ ub, Tr _ lb are the upper and lower limits of the confidence interval, respectively.
Preferably, the optimization expression of the maximum improvement expectation and the minimum agent model prediction is as follows:
Figure BDA0002790788350000051
Figure BDA0002790788350000052
wherein
Figure BDA0002790788350000053
A prediction function of the proxy model.
The invention has the beneficial effects that: according to the multi-fidelity shape optimization method of the autonomous underwater vehicle based on data mining, gradient information does not need to be calculated, high-fidelity data or high-fidelity.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a schematic diagram of the low-fidelity proxy model data mining process of the present invention.
FIG. 3 is a schematic view of the shape parameters of an underwater vehicle according to an embodiment of the invention.
Fig. 4 is a line drawing of an optimized underwater vehicle according to an embodiment of the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
The method directly carries out data mining on the low-fidelity proxy model, and directly extracts and screens useful information from the low-fidelity model through surface mining, data preprocessing and deep mining, wherein the useful information is used for updating a high-fidelity model database on one hand, and is used for constructing a local search region on the other hand, so that the high-fidelity model is induced to be converged near the global true optimal solution of the optimization model more quickly. And meanwhile, in order to avoid misleading of the low-fidelity proxy model to the high-fidelity proxy model and minimize the optimization process of the proxy model prediction function to enable the high-fidelity model to be in local optimization, the data mining of the low-fidelity proxy model, the global optimization of the proxy model with the maximum improvement expectation value and the local optimization process of the proxy model with the minimum prediction function are alternately carried out.
The invention aims to solve the defects of low precision of the shape optimization design of the traditional autonomous underwater vehicle, high cost and low efficiency of the shape calculation of the traditional gradient optimization and agent model optimization autonomous underwater vehicle.
As shown in fig. 1, the invention provides an autonomous underwater vehicle multi-fidelity shape optimization method based on data mining, comprising the following steps:
step 1: setting an optimized design variable and range, a constraint condition and an objective function of the autonomous underwater vehicle;
with adjustable parameters q of autonomous underwater vehicle headh1、qh2And tail curve segment linear adjustable parameter qt1、qt2For design variables, the head fullness, the caudal vertebra half angle and the diameter of the caudal facet are taken as constraint conditions, the minimum zero lift resistance of the autonomous underwater vehicle is taken as an optimization target, and an optimized mathematical model is as follows:
Figure BDA0002790788350000061
wherein psiHThe constraint condition of the head fullness coefficient of the autonomous underwater vehicle, alpha is the constraint condition of the caudal vertebra half angle of the autonomous underwater vehicle, DEThe tail end face diameter of the autonomous underwater vehicle, Ub and Lb are respectively the upper limit and the lower limit of the whole design space, psi*、α*、D*Respectively representing a head fullness coefficient threshold value of the autonomous underwater vehicle, a caudal vertebra half angle of the autonomous underwater vehicle and a tail end face diameter threshold value of the autonomous underwater vehicle;
respectively completing parametric geometric modeling, grid division and flow field analysis and calculation of the autonomous underwater vehicle, and realizing automatic appearance and resistance calculation flow creation of the autonomous underwater vehicle;
step 2: respectively evaluating high-fidelity sample points and low-fidelity sample points of the autonomous underwater vehicle by adopting an optimized Latin hypercube method, and respectively constructing high-fidelity agent model databases and low-fidelity agent model databases of the autonomous underwater vehicle according to design variables and resistance response values of the sample points;
and step 3: adopting a genetic algorithm to maximize the improvement expectation value of the low-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the evaluated new sample point into a low-fidelity proxy model database, and updating the low-fidelity proxy model;
and 4, step 4: performing surface data mining on the low-fidelity proxy model by adopting a multi-starting-point optimization method to find out all local optimal point positions in the low-fidelity proxy model;
and 5: removing points with the distance smaller than a set threshold value and completely repeated points from the data acquired in the mining process of the layer data in the step 4, and realizing the noise reduction of the local optimal point data;
step 6: screening out expected points in the local optimal point data subjected to noise reduction in the step 5 through a useful data evaluation algorithm, and removing repeated data;
and 7: carrying out fluid resistance evaluation on the expected points obtained in the step 6, adding the expected points passing the evaluation into a high fidelity proxy model database, and updating the high fidelity proxy model;
and 8: adopting a genetic algorithm to maximize the improvement expectation value of the high-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the new sample point which passes the evaluation into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
and step 9: further screening out a point with the minimum resistance response value as a most promising point from the evaluated promising points in the step 7, constructing a local confidence domain around the most promising point, minimizing a predicted value of the high-fidelity proxy model in the local confidence domain by adopting a genetic algorithm so as to obtain a new sample point and evaluate the fluid resistance, adding the evaluated new sample point into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
step 10: optimizing the maximum improvement expectation and the minimum agent model prediction simultaneously by adopting a genetic algorithm for the low-fidelity agent model, performing a de-duplication data processing process on the obtained new sample point and the most promising point obtained in the step 9, evaluating the fluid resistance, adding the evaluated point into a low-fidelity agent model database, and updating the low-fidelity agent model;
step 11: judging whether the current optimal value of the multi-fidelity shape optimization of the autonomous underwater vehicle meets an iteration termination condition formula (2), and if so, performing step 12; if not, returning to the step 4;
Figure BDA0002790788350000071
wherein x isi+1、xiRespectively, the optimal point of the current iteration and the optimal point of the last iteration, yhighIs a resistance response value;
step 12: and outputting the multi-fidelity shape optimization value of the autonomous underwater vehicle meeting the iteration termination condition, and finishing the optimization.
Preferably, the method for eliminating the points with the distance smaller than the set threshold in the local optimal points in step 5 is as follows:
point X of the local optimum pointsiAnd point XjDistance between themijIf the distance is smaller than the set Threshold, the distance is considered to be the adjacent point or the repeated point, and the distance judgment expression is as follows:
Figure BDA0002790788350000072
preferably, the useful data evaluation algorithm in step 6 is:
respectively using the point with the maximum improvement expectation value of the high-fidelity proxy model, the point with the maximum improvement expectation value of the low-fidelity proxy model and the point with the minimum prediction function as the desired points;
the expression for the expected improvement value is:
Figure BDA0002790788350000081
wherein
Figure BDA0002790788350000082
Predicting minimum value for proxy model, s is variance of proxy model, yminPhi () is the cumulative distribution function of the standard normal distribution function for the optimal value of the current sample point database,
Figure BDA0002790788350000083
is a probability density function of a standard normal distribution function.
Preferably, the method for constructing the local confidence domain in step 9 is as follows:
a local search range is constructed near the most promising point mined and screened from the low-fidelity agent model, the high-fidelity model is induced to quickly approach the global true optimal solution, and the confidence domain expression is as follows:
Figure BDA0002790788350000084
where delta is the confidence radius, where,
Figure BDA0002790788350000085
for the most promising point, w is the scaling factor, and Tr _ ub, Tr _ lb are the upper and lower limits of the confidence interval, respectively.
Preferably, the optimization expression of the maximum improvement expectation and the minimum agent model prediction is as follows:
Figure BDA0002790788350000086
Figure BDA0002790788350000087
wherein
Figure BDA0002790788350000088
A prediction function of the proxy model.
The specific embodiment is as follows:
1. determining design variables and ranges of an optimized model of the autonomous underwater vehicle, constraint conditions and an objective function:
the shape of the underwater vehicle of the embodiment adopts a Glan Ville curve family, a head curve adopts a biparametric square root polynomial flat-headed line type, a tail curve adopts a biparametric general polynomial sharp-tail line type, and the design variable is an adjustable parameter q of the head curveh1、qh2And adjustable parameter q of tail curvet1、qt2The value ranges are respectively as follows: q. q.sh1∈[0,4]，qh2∈[0,14]，qt1∈[3,20]，qt2∈[30,36]；
In order to meet the overall requirement of volume, the shape of the underwater vehicle of the embodiment has the constraint condition of the head fullness coefficient: psiH≥0.6；
In order to delay or prevent the separation of the tail boundary layer, the shape of the underwater vehicle of the embodiment has the following tail half-angle constraint conditions: alpha is less than or equal to 12 DEG
The shape of the underwater vehicle of the embodiment is that in order to reduce the bottom resistance and simultaneously increase the wake effect of the propeller, the diameter of the tail end face is as follows: dE≥0.1m
The optimization target of the underwater vehicle is that the underwater vehicle with the fin rudder has the smallest navigation resistance by taking the largest cross-sectional area as the characteristic area when the attack angle is zero, so that the energy is saved and the range is increased.
2. And integrating UG, ICEM CFD and FLUENT software by adopting an ISIGHT software platform to realize data stream transmission among the modules, respectively completing parameterized geometric modeling, grid division and flow field analysis and calculation of the underwater autonomous vehicle, realizing the flow of automatically creating the shape and flow calculation of the autonomous underwater vehicle, and completing the evaluation of the resistance value at the design point.
And respectively carrying out experimental design sampling on the high-fidelity agent model and the low-fidelity agent model by adopting an optimized Latin hypercube method in the whole design variable space, wherein the number of sampling points of the low-fidelity agent model is 20, and the number of sampling points of the high-fidelity agent model is 9.
And then, respectively calling CFD simulation with high and low precision to the sampling points of the high and low fidelity proxy models for resistance evaluation, and respectively constructing the high and low fidelity proxy models according to the sampling design points and the corresponding resistance response values.
3. And maximizing the improvement expectation value of the low-fidelity proxy model in the whole design space by adopting a particle swarm genetic algorithm (PSO), thus obtaining a new sample point, calling CFD simulation with low precision to evaluate the fluid resistance, adding the fluid resistance into a low-fidelity proxy model database, and updating the low-fidelity proxy model.
4. Data mining is performed on the low fidelity model, and the data mining process is shown in fig. 2. And mining the data of the surface layer of the low-fidelity agent model by adopting a multi-starting-point optimization method, wherein the number of the multi-starting points is 20, finding the positions of all local optimal points in the low-fidelity agent model closest to the multi-starting points from the multi-starting points through a Sequence Quadratic Programming (SQP) algorithm, and recording the surface layer mined data as Xlocal＝(X1,X2,...,X20)。
5. And performing data preprocessing on a large amount of data acquired in the surface layer mining process, and eliminating points with very close distances and completely repeated points in local optimal points through a preprocessing algorithm in the data preprocessing so as to realize the noise reduction process of the large amount of local optimal point data.
First calculate X before data preprocessinglocalTwo arbitrary points in XiAnd XjIs marked as the Euclidean distance between
Figure BDA0002790788350000091
Then compare DistanceijAnd a set distance threshold
Figure BDA0002790788350000092
Where Ub and Lb are the upper limit and the lower limit of the design space, respectively, if Distanceij≤Threshold, then the two local optimum points are considered to be similar or identical, and then the database X is mined from the surface layerlocalTo select any one of the two points to be saved to XusefulIn a database; otherwise, the two points are considered to be far away from each other, and the two points are stored in XusefulIn a database.
6. Data X after data preprocessingusefulCarrying out deep knowledge mining, and further extracting and screening to obtain promising data points X through a data evaluation algorithmpromising。
The data evaluation algorithm respectively uses the improved expectation function and the prediction function of the high fidelity agent model and the low fidelity agent model to correspondingly evaluate the four evaluation functions, and extracts XusefulThe data with the maximum expected improvement value and the data with the minimum prediction function value of the medium-high and low proxy models are taken as promising data, and then the promising data are stored in X after data preprocessingpromisingA database.
The improved expectation function expression of the proxy model is as follows:
Figure BDA0002790788350000101
wherein
Figure BDA0002790788350000102
Predicting minimum value for proxy model, s is variance of proxy model, yminThe optimal value is stored for the current sample point data. Phi (x) is the cumulative distribution function of the standard normal distribution function,
Figure BDA0002790788350000103
is a probability density function of a standard normal distribution function.
7. Selecting XpromisingMost promising point in database
Figure BDA0002790788350000104
The high fidelity proxy model is updated simultaneously.
To have a hope ofData X ofpromisingCalling high-precision CFD simulation to evaluate resistance, sorting and screening out the point with the minimum resistance value as the most promising point and recording the point as the most promising point
Figure BDA0002790788350000105
Will hopefully data XpromisingAnd adding the sample points into a high-fidelity proxy model database as new sample points, and updating the high-fidelity proxy model.
8. Constructing local confidence domain and inducing high fidelity model at most promising point
Figure BDA0002790788350000106
Local exploration and optimization are carried out nearby.
To the most promising point
Figure BDA0002790788350000107
Centered, in the vicinity of which a confidence radius δ for each dimension is constructedtThen will be
Figure BDA0002790788350000108
After the confidence radius is respectively added and subtracted, the intersection is taken with the upper limit Ub and the lower limit Lb of the overall design space, and then the intersection is taken as the confidence upper limit and the confidence lower limit of a confidence domain, the confidence upper limit and the confidence lower limit are respectively marked as Tr _ Ub and Tr _ Lb, the local confidence domain is prevented from crossing the boundary, and the expression of the confidence domain is as follows:
Figure BDA0002790788350000109
where w is the scaling factor that controls the confidence radius.
9. Global search of the high-fidelity proxy model: and maximizing the improvement expectation value of the high-fidelity agent model in the whole design space by adopting a PSO genetic algorithm, thereby obtaining a new sample point, carrying out fluid resistance evaluation, adding the new sample point into a high-fidelity agent model database, and updating the high-fidelity agent model.
10. Local search of the high-fidelity agent model: and minimizing the predicted value of the high-fidelity proxy model in a local confidence domain by adopting a PSO genetic algorithm so as to obtain a new sample point, carrying out fluid resistance evaluation, adding the new sample point into a high-fidelity proxy model database, and updating the high-fidelity proxy model.
11. Updating the low fidelity agent model: optimizing the low-fidelity agent model by adopting a genetic algorithm to maximize improvement expectation and minimize agent model prediction to obtain new sample points and most promising points
Figure BDA0002790788350000111
And after the repeated data processing process is carried out, the fluid resistance evaluation is carried out in parallel, and then the fluid resistance evaluation is added into the low-fidelity database to update the low-fidelity proxy model.
12. Judging whether the current optimal value meets the iteration termination condition, if so, performing the step 12; if not, returning to the step 4.
13. And outputting an optimized value meeting the iteration termination condition, and finishing the optimization.
According to the steps, the high-fidelity agent model is subjected to 8-suboptimal iteration, high-fidelity CFD simulation calculation is called for 20 times in total, low-fidelity is subjected to 28-suboptimal iteration, low-fidelity CFD simulation calculation is called for 48 times in total, optimization iteration termination conditions are met, and the optimal design point obtained by optimization is as follows: q. q.sh1＝1.2，qh2＝11.55，qt1＝3.0495，qt2The coefficient of the sailing resistance at the optimum design point is 34.8: cd0.094396, head fullness is: psiH0.6235, the caudal half angle is: alpha is 11.6669 degrees, and the diameter of the tail end face is as follows: dE0.18m, all meet the constraint requirement. The profile of the optimized autonomous underwater vehicle is shown in fig. 4.
The invention adopts a multi-fidelity shape optimization method based on data mining, calls high-fidelity CFD simulation for 20 times and low-fidelity CFD simulation for 48 times, reduces the total calculation cost to call high-fidelity CFD for 25 times, and compares the optimized shape parameters with the optimized results of test design sampling and calling high-fidelity CFD for 35 times as shown in a table below, wherein the optimized shape parameters are calculated as shown in a figure 3.
TABLE 1 comparison of the results of optimization of the profile drag of an underwater vehicle
Figure BDA0002790788350000112
The detailed steps and the optimization results of the optimized embodiment show that the resistance coefficient of the multi-fidelity shape optimization method based on data mining is reduced by 16.857%, the energy consumption is saved, the voyage is increased, the head fullness is increased by 2.55%, and the internal volume is larger; the caudal vertebra half angle has increased 4.49%, but still is less than 12 °, can avoid the separation of boundary layer, and the tail end face diameter has reduced 33.77%, and the tail end face diameter reduces and is favorable to reducing the bottom resistance, has increased the wake flow benefit of propeller simultaneously, and holistic performance parameter is superior to the result of experimental design sampling, and it reduces greatly to call high fidelity CFD calculation number of times simultaneously, has reduced the calculation cost when improving optimization accuracy.

Claims (5)

1. An autonomous underwater vehicle multi-fidelity shape optimization method based on data mining is characterized by comprising the following steps:
step 1: setting an optimized design variable and range, a constraint condition and an objective function of the autonomous underwater vehicle;
with adjustable parameters q of autonomous underwater vehicle headh1、qh2And tail curve segment linear adjustable parameter qt1、qt2For design variables, the head fullness, the caudal vertebra half angle and the diameter of the caudal facet are taken as constraint conditions, the minimum zero lift resistance of the autonomous underwater vehicle is taken as an optimization target, and an optimized mathematical model is as follows:
Figure FDA0002790788340000011
wherein psiHThe constraint condition of the head fullness coefficient of the autonomous underwater vehicle, alpha is the constraint condition of the caudal vertebra half angle of the autonomous underwater vehicle, DEThe tail end face diameter of the autonomous underwater vehicle is shown, and Ub and Lb are respectively the upper limit and the lower limit of the whole design space，ψ*、α*、D*Respectively representing a head fullness coefficient threshold value of the autonomous underwater vehicle, a caudal vertebra half angle of the autonomous underwater vehicle and a tail end face diameter threshold value of the autonomous underwater vehicle;
respectively completing parametric geometric modeling, grid division and flow field analysis and calculation of the autonomous underwater vehicle, and realizing automatic appearance and resistance calculation flow creation of the autonomous underwater vehicle;
step 2: respectively evaluating high-fidelity sample points and low-fidelity sample points of the autonomous underwater vehicle by adopting an optimized Latin hypercube method, and respectively constructing high-fidelity agent model databases and low-fidelity agent model databases of the autonomous underwater vehicle according to design variables and resistance response values of the sample points;
and step 3: adopting a genetic algorithm to maximize the improvement expectation value of the low-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the evaluated new sample point into a low-fidelity proxy model database, and updating the low-fidelity proxy model;
and 4, step 4: performing surface data mining on the low-fidelity proxy model by adopting a multi-starting-point optimization method to find out all local optimal point positions in the low-fidelity proxy model;
and 5: removing points with the distance smaller than a set threshold value and completely repeated points from the data acquired in the mining process of the layer data in the step 4, and realizing the noise reduction of the local optimal point data;
step 6: screening out expected points in the local optimal point data subjected to noise reduction in the step 5 through a useful data evaluation algorithm, and removing repeated data;
and 7: carrying out fluid resistance evaluation on the expected points obtained in the step 6, adding the expected points passing the evaluation into a high fidelity proxy model database, and updating the high fidelity proxy model;
and 8: adopting a genetic algorithm to maximize the improvement expectation value of the high-fidelity proxy model, carrying out fluid resistance evaluation after obtaining a new sample point, adding the new sample point which passes the evaluation into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
and step 9: further screening out a point with the minimum resistance response value as a most promising point from the evaluated promising points in the step 7, constructing a local confidence domain around the most promising point, minimizing a predicted value of the high-fidelity proxy model in the local confidence domain by adopting a genetic algorithm so as to obtain a new sample point and evaluate the fluid resistance, adding the evaluated new sample point into a high-fidelity proxy model database, and updating the high-fidelity proxy model;
step 10: optimizing the maximum improvement expectation and the minimum agent model prediction simultaneously by adopting a genetic algorithm for the low-fidelity agent model, performing a de-duplication data processing process on the obtained new sample point and the most promising point obtained in the step 9, evaluating the fluid resistance, adding the evaluated point into a low-fidelity agent model database, and updating the low-fidelity agent model;
step 11: judging whether the current optimal value of the multi-fidelity shape optimization of the autonomous underwater vehicle meets an iteration termination condition formula (2), and if so, performing step 12; if not, returning to the step 4;
Figure FDA0002790788340000021
wherein x isi+1、xiRespectively, the optimal point of the current iteration and the optimal point of the last iteration, yhighIs a resistance response value;
step 12: and outputting the multi-fidelity shape optimization value of the autonomous underwater vehicle meeting the iteration termination condition, and finishing the optimization.
2. The method for optimizing the multi-fidelity profile of the autonomous underwater vehicle based on data mining as claimed in claim 1, wherein the points with the distance smaller than the set threshold value in the local optimal points are removed in the step 5 by the following method:
point X of the local optimum pointsiAnd point XjDistance between themijIf the distance is smaller than the set Threshold, the distance is considered to be the adjacent point or the repeated point, and the distance judgment expression is as follows:
Figure FDA0002790788340000022
3. the method for optimizing the multi-fidelity profile of the autonomous underwater vehicle based on data mining of claim 1, wherein the useful data evaluation algorithm in the step 6 is as follows:
respectively using the point with the maximum improvement expectation value of the high-fidelity proxy model, the point with the maximum improvement expectation value of the low-fidelity proxy model and the point with the minimum prediction function as the desired points;
the expression for the expected improvement value is:
Figure FDA0002790788340000031
wherein
Figure FDA0002790788340000032
Predicting minimum value for proxy model, s is variance of proxy model, yminPhi () is the cumulative distribution function of the standard normal distribution function for the optimal value of the current sample point database,
Figure FDA0002790788340000033
is a probability density function of a standard normal distribution function.
4. The method for optimizing the multi-fidelity shape of the autonomous underwater vehicle based on data mining of claim 1, wherein the method for constructing the local confidence domain in the step 9 is as follows:
a local search range is constructed near the most promising point mined and screened from the low-fidelity agent model, the high-fidelity model is induced to quickly approach the global true optimal solution, and the confidence domain expression is as follows:
Figure FDA0002790788340000034
where delta is the confidence radius, where,
Figure FDA0002790788340000035
for the most promising point, w is the scaling factor, and Tr _ ub, Tr _ lb are the upper and lower limits of the confidence interval, respectively.
5. The method of optimizing the multi-fidelity profile of an autonomous underwater vehicle based on data mining of claim 1, wherein the optimization expressions of the maximum improvement expectation and the minimum agent model prediction are:
Figure FDA0002790788340000036
Figure FDA0002790788340000037
wherein
Figure FDA0002790788340000038
A prediction function of the proxy model.
CN202011314168.1A 2020-11-21 2020-11-21 Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining Active CN112464366B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011314168.1A CN112464366B (en) 2020-11-21 2020-11-21 Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011314168.1A CN112464366B (en) 2020-11-21 2020-11-21 Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining

Publications (2)

Publication Number Publication Date
CN112464366A true CN112464366A (en) 2021-03-09
CN112464366B CN112464366B (en) 2021-11-09

Family

ID=74798188

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011314168.1A Active CN112464366B (en) 2020-11-21 2020-11-21 Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining

Country Status (1)

Country Link
CN (1) CN112464366B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113591383A (en) * 2021-08-02 2021-11-02 大连理工大学 Multi-fidelity data-oriented digital twin modeling method and prediction system
CN113606991A (en) * 2021-07-21 2021-11-05 西北工业大学 High-pressure chamber structure for underwater boat speed launching and design method
CN114386342A (en) * 2022-01-19 2022-04-22 江西理工大学 Underwater ore collector structure optimization design method based on combined optimization algorithm
CN114580085A (en) * 2022-04-29 2022-06-03 北京理工大学 Multi-time underwater vehicle head shape optimization method based on proxy model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107515994A (en) * 2017-09-11 2017-12-26 西北工业大学 A kind of more fidelity design optimization methods of the shell structure of Autonomous Underwater Vehicle
CN107531317A (en) * 2015-03-03 2018-01-02 麻省理工学院 Submarine navigation device designs and control method
WO2018119104A1 (en) * 2016-12-21 2018-06-28 Northwestern University Systems and methods for computational simulation of self-propelling vehicles for aerodynamic design
US20200318602A1 (en) * 2018-01-17 2020-10-08 Lone Gull Holdings, Ltd. Self-powered, self-propelled computer grid with loop topology

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107531317A (en) * 2015-03-03 2018-01-02 麻省理工学院 Submarine navigation device designs and control method
WO2018119104A1 (en) * 2016-12-21 2018-06-28 Northwestern University Systems and methods for computational simulation of self-propelling vehicles for aerodynamic design
CN107515994A (en) * 2017-09-11 2017-12-26 西北工业大学 A kind of more fidelity design optimization methods of the shell structure of Autonomous Underwater Vehicle
US20200318602A1 (en) * 2018-01-17 2020-10-08 Lone Gull Holdings, Ltd. Self-powered, self-propelled computer grid with loop topology

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113606991A (en) * 2021-07-21 2021-11-05 西北工业大学 High-pressure chamber structure for underwater boat speed launching and design method
CN113606991B (en) * 2021-07-21 2022-07-26 西北工业大学 High-pressure chamber structure for underwater boat speed launching and design method
CN113591383A (en) * 2021-08-02 2021-11-02 大连理工大学 Multi-fidelity data-oriented digital twin modeling method and prediction system
CN113591383B (en) * 2021-08-02 2024-05-24 大连理工大学 Digital twin modeling method and prediction system for multi-fidelity data
CN114386342A (en) * 2022-01-19 2022-04-22 江西理工大学 Underwater ore collector structure optimization design method based on combined optimization algorithm
CN114386342B (en) * 2022-01-19 2024-03-29 江西理工大学 Underwater ore collector structure optimization design method based on combination optimization algorithm
CN114580085A (en) * 2022-04-29 2022-06-03 北京理工大学 Multi-time underwater vehicle head shape optimization method based on proxy model

Also Published As

Publication number Publication date
CN112464366B (en) 2021-11-09

Similar Documents

Publication Publication Date Title
CN112464366B (en) Multi-fidelity shape optimization method of autonomous underwater vehicle based on data mining
Jeong et al. Data mining for aerodynamic design space
Biancolini et al. Sails trim optimisation using CFD and RBF mesh morphing
CN108319799A (en) A kind of more fidelity optimum design methods of the shape of Autonomous Underwater Vehicle
Koziel et al. Knowledge-based airfoil shape optimization using space mapping
US20060058985A1 (en) Adjoint-based design variable adaptation
CN106934074B (en) Global optimal turbofan engine air inlet channel noise reduction design method
CN110795816B (en) Yacht molded line optimization method based on resistance performance
CN110309571B (en) Wing body fusion underwater glider external shape optimization method based on radial basis function model
CN111295657A (en) Shape optimization via gradient descent technology device using convolutional neural network proxy
Huang et al. Aerodynamic multi-objective integrated optimization based on principal component analysis
CN111191785A (en) Structure searching method based on expanded search space
CN114036650B (en) Reverse design method of non-axisymmetrical natural laminar flow nacelle and natural laminar flow nacelle
Gaggero et al. A marine propeller design method based on two-fidelity data levels
CN115146412A (en) Ship type performance forecasting and optimizing method and system
Serani et al. An adaptive N-fidelity metamodel for design and operational-uncertainty space exploration of complex industrial problems
Kirz Surrogate based shape optimization of a low boom fuselage wing configuration
CN112464367B (en) Appearance and structure two-layer design optimization method of autonomous underwater vehicle
Gaggero et al. A two-fidelity level approach for marine propeller design
CN114818128A (en) Ship body local curved surface optimization neural network modeling method and ship body local curved surface optimization method
CN113657045A (en) Complex aircraft model reduced order representation method based on multilayer cooperative Gaussian process
Tran et al. A method for optimizing the hull form of fishing vessels
Chen et al. Multi-objective shape optimization of underwater vehicles based on an adaptive sampling algorithm
Choo et al. Ship Design with a Morphing Evolutionary Algorithm
Epstein et al. Aerodynamic Shape Optimization by Automatic Hybrid Genetic Tool OPTIMENGA_AERO

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