CN113255243A - Bionic robot fish near-wall flow field identification method and system based on artificial lateral line - Google Patents

Bionic robot fish near-wall flow field identification method and system based on artificial lateral line Download PDF

Info

Publication number
CN113255243A
CN113255243A CN202110511911.0A CN202110511911A CN113255243A CN 113255243 A CN113255243 A CN 113255243A CN 202110511911 A CN202110511911 A CN 202110511911A CN 113255243 A CN113255243 A CN 113255243A
Authority
CN
China
Prior art keywords
wall
pressure
regression model
prediction regression
bionic robot
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
CN202110511911.0A
Other languages
Chinese (zh)
Other versions
CN113255243B (en
Inventor
谢鸥
姚吉
葛飞飞
孙兆光
牛雪梅
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.)
Suzhou University of Science and Technology
Original Assignee
Suzhou University of Science and Technology
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 Suzhou University of Science and Technology filed Critical Suzhou University of Science and Technology
Priority to CN202110511911.0A priority Critical patent/CN113255243B/en
Publication of CN113255243A publication Critical patent/CN113255243A/en
Application granted granted Critical
Publication of CN113255243B publication Critical patent/CN113255243B/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]
    • G06F30/20—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
    • 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
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Farming Of Fish And Shellfish (AREA)

Abstract

The invention relates to a method and a system for recognizing a near-wall flow field of a bionic robot fish based on an artificial lateral line, which comprises the following steps: a plurality of pressure sensors are arranged on the bionic robot fish, and pressure data are acquired through the pressure sensors; according to the pressure data, a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance are constructed by adopting a multilayer feedforward neural network; evaluating the prediction regression model of the incoming flow velocity by adopting a mean square error and a decision coefficient to obtain an optimal prediction regression model of the incoming flow velocity; and evaluating the prediction regression model of the wall-to-wall distance by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the wall-to-wall distance. The method can identify the near-wall flow field, realizes the prediction of the incoming flow speed and the near-wall distance of the near-wall swimming of the bionic robot fish, and provides a new idea for the perception of the underwater complex non-structural environment.

Description

Bionic robot fish near-wall flow field identification method and system based on artificial lateral line
Technical Field
The invention relates to the technical field of bionic robot fish, in particular to a method and a system for recognizing a near-wall flow field of the bionic robot fish based on an artificial lateral line.
Background
With the deep development of the ocean by human beings, the facing underwater operation environment is also more dangerous. Autonomous Underwater Vehicles (AUV) are important tools for ocean exploration, and have put higher demands on their performance. The traditional underwater robot has the defects of low efficiency, high noise, poor maneuverability and the like due to the adoption of propeller propulsion, and cannot adapt to the increasingly developed underwater operation requirements. In recent years, inspired by the superior swimming performance of fishes, researchers carry out deep research on the swimming mechanism of the fishes and simulate and research various high-performance bionic underwater robots. As autonomous underwater operation equipment, the bionic robot fish needs to effectively sense and identify the surrounding flow field environment. However, due to the influence of water turbidity and the complicated unstructured underwater terrain environment, the application of the traditional optical imaging and sonar detection technology is limited, and the operation capability of the bionic underwater robot is severely restricted.
Disclosure of Invention
Therefore, the invention aims to solve the technical problems that the underwater robot in the prior art is difficult to operate and the application of the traditional optical imaging and sonar detection technology is limited.
In order to solve the technical problem, the invention provides a bionic robot fish near-wall surface flow field identification method based on an artificial lateral line, which comprises the following steps:
s1, configuring a plurality of pressure sensors on the bionic robot fish, and acquiring pressure data through the plurality of pressure sensors;
s2, constructing a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance by adopting a multi-layer feedforward neural network according to the pressure data;
s3, evaluating the prediction regression model of the incoming flow velocity by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the incoming flow velocity;
and evaluating the prediction regression model of the wall-to-wall distance by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the wall-to-wall distance.
Preferably, the step three is further followed by:
acquiring and processing pressure coefficients obtained by a plurality of pressure sensors under different wall-approaching distances and different incoming flow speeds, and solving the variance to obtain a pressure coefficient variance curve;
and reducing the characteristics of the input data by adopting a characteristic variable stepwise elimination method according to the pressure coefficient variance curve, and optimizing a characteristic set.
Preferably, the S1 includes:
the head of the bionic robot fish is provided with a pressure sensor, and pressure sensor groups are uniformly arranged along the body length direction of the bionic robot fish;
acquiring pressure data through a plurality of pressure sensors to obtain an integral pressure data set, wherein the integral pressure data set comprises pressure data acquired by the pressure sensors at the head of the bionic robot fish and pressure data acquired by a pressure sensor group in the length direction of the bionic robot fish;
and averaging the pressure data acquired by the pressure sensors at the head of the bionic robot fish, and summing and averaging the pressure data acquired by the pressure sensor group in the length direction of the bionic robot fish.
Preferably, the S1 further includes:
and carrying out non-dimensionalization processing on the pressure data acquired by the pressure sensor to obtain a normalized pressure coefficient.
Preferably, in S2, the structural parameters of the multi-layer feedforward neural network include the number of input data features, the number of hidden layers, the number of hidden layer neurons, and the selection of activation functions of hidden layers and output layers.
Preferably, the ReLU function is used as a hidden layer activation function of a multi-layer feedforward neural network, and the output layer of the multi-layer feedforward neural network uses a linear activation function.
Preferably, S2 further includes:
obtaining an optimized neural network structure, specifically:
gradually increasing the number of hidden layer layers from 1 to 5, gradually increasing the number of neurons of the first hidden layer from the number of input features to 3 times, configuring the number of neurons of each hidden layer in a descending rule, wherein the number of neurons of the next layer is 2/3 of the previous layer.
Preferably, in the S3,
mean square error
Figure RE-GDA0003138301760000031
Determining coefficients
Figure RE-GDA0003138301760000032
Wherein the content of the first and second substances,
Figure RE-GDA0003138301760000033
Yi,
Figure RE-GDA0003138301760000034
the predicted value, the observed value, and the mean value are respectively represented.
Preferably, the reducing the input data features by adopting a feature variable stepwise elimination method according to the pressure coefficient variance curve, and optimizing the feature set specifically includes:
and gradually eliminating corresponding input data characteristics from small to large according to the pressure coefficient variance value acquired by the pressure sensor.
The invention discloses a bionic robot fish near-wall surface flow field identification system based on an artificial lateral line, which comprises:
the data acquisition module is used for configuring a plurality of pressure sensors on the bionic robot fish and acquiring pressure data through the plurality of pressure sensors;
the model building module is used for building a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance by adopting a multi-layer feedforward neural network according to the pressure data;
the incoming flow velocity model optimization module evaluates the prediction regression model of the incoming flow velocity by adopting a mean square error and a decision coefficient to obtain an optimal prediction regression model of the incoming flow velocity;
and the wall-to-wall distance model optimization module evaluates the prediction regression model of the wall-to-wall distance by adopting a mean square error and a decision coefficient to obtain an optimal prediction regression model of the wall-to-wall distance.
Compared with the prior art, the technical scheme of the invention has the following advantages:
1. according to the invention, the near-wall surface wave propulsion of the bionic robot fish can cause the asymmetric distribution of the surrounding flow field structure, and a basis is provided for flow field parameter identification based on an artificial lateral line.
2. The invention obtains the pressure coefficient variance values of the side line pressure sensor array under different incoming flow speeds and wall-approaching distances, and discloses the identification degrees of the pressure sensors at different positions on the flow field parameter change.
3. The invention provides a near-wall flow field identification method based on ALL and a multilayer feedforward neural network, which realizes the prediction of the incoming flow speed and the near-wall distance of the near-wall swimming of the bionic robot fish and provides a new idea for the perception of an underwater complex non-structural environment.
Drawings
FIG. 1 is a flow chart of a bionic robot fish near-wall surface flow field identification method based on an artificial lateral line in the invention;
FIG. 2 is a simulation model of a biomimetic fish and its environment;
FIG. 3 is a schematic diagram of an ALL sensor layout, wherein (a) is a pressure field cloud and (b) is a velocity field cloud;
FIG. 4 is a cloud of flow field distributions, where (a) is the wall approach distance and (b) is the incoming flow distance;
FIG. 5 is a pressure coefficient variance plot;
FIG. 6 is an evaluation index chart of an incoming flow velocity prediction regression model, in which (a) is R2And (b) is MSE;
FIG. 7 is an evaluation index graph of a wall distance prediction regression model, in which (a) is a determination coefficient and (b) is an average error;
FIG. 8 is an evaluation index graph of data feature elimination of an incoming flow velocity prediction regression model, in which (a) is R2And (b) is MSE;
FIG. 9 is a graph of evaluation indexes for feature elimination of regression model predicted by wall distance, wherein (a) is R2And (b) is MSE;
FIG. 10 is a comparison of the predicted effect of data feature elimination in an incoming flow velocity prediction regression model, wherein (a) is no-data feature elimination (R)20.998), (b) to eliminate 6 data features (R)2＝0.994)；
FIG. 11 is a comparison graph of the predicted effect of feature elimination for a regression model predicted by wall distance, wherein (a) is the elimination of no-data feature (R)20.912), (b) to eliminate 5 data features (R)2＝0.883)。
Detailed Description
The present invention is further described below in conjunction with the following figures and specific examples so that those skilled in the art may better understand the present invention and practice it, but the examples are not intended to limit the present invention.
Referring to fig. 1, the invention discloses a bionic robot fish near-wall flow field identification method based on an artificial lateral line, which comprises the following steps:
step one, dispose a plurality of pressure sensors on bionical machine fish, gather pressure data through a plurality of pressure sensors, specifically include:
the head of the bionic robot fish is provided with a pressure sensor, and pressure sensor groups are uniformly arranged along the body length direction of the bionic robot fish;
acquiring pressure data through a plurality of pressure sensors to obtain an integral pressure data set, wherein the integral pressure data set comprises pressure data acquired by the pressure sensors at the head of the bionic robot fish and pressure data acquired by a pressure sensor group in the length direction of the bionic robot fish;
averaging pressure data acquired by a pressure sensor at the head of the bionic robot fish, summing the pressure data acquired by a pressure sensor group in the length direction of the bionic robot fish body and averaging;
and carrying out non-dimensionalization processing on the pressure data acquired by the pressure sensor to obtain a normalized pressure coefficient.
And step two, constructing a prediction regression model of the incoming flow speed and a prediction regression model of the wall-leaning distance by adopting a multilayer feedforward neural network according to the pressure data.
The structural parameters of the multilayer feedforward neural network comprise the number of input data features, the number of hidden layers, the number of hidden layer neurons and the selection of activation functions of the hidden layers and the output layers.
And adopting a ReLU function as a hidden layer activation function of the multilayer feedforward neural network, wherein the output layer of the multilayer feedforward neural network adopts a linear activation function.
The second step further comprises: obtaining an optimized neural network structure, specifically: gradually increasing the number of hidden layer layers from 1 to 5, gradually increasing the number of neurons of the first hidden layer from the number of input features to 3 times, configuring the number of neurons of each hidden layer in a descending rule, wherein the number of neurons of the next layer is 2/3 of the previous layer.
And step three, evaluating the prediction regression model of the incoming flow speed by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the incoming flow speed.
And evaluating the prediction regression model of the wall-to-wall distance by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the wall-to-wall distance.
In the third step, mean square error
Figure RE-GDA0003138301760000071
Determining coefficients
Figure RE-GDA0003138301760000072
Wherein the content of the first and second substances,
Figure RE-GDA0003138301760000073
Yi,
Figure RE-GDA0003138301760000074
the predicted value, the observed value, and the mean value are respectively represented.
The third step further comprises the following steps:
acquiring and processing pressure coefficients obtained by a plurality of pressure sensors under different wall-approaching distances and different incoming flow speeds, and solving the variance to obtain a pressure coefficient variance curve;
and reducing the input data characteristics by adopting a characteristic variable stepwise elimination method according to the pressure coefficient variance curve, and optimizing a characteristic set, namely gradually eliminating the corresponding input data characteristics from small to large according to the pressure coefficient variance value acquired by the pressure sensor.
The invention discloses a bionic robot fish near-wall surface flow field identification system based on an artificial lateral line.
The data acquisition module is used for configuring a plurality of pressure sensors on the bionic robot fish and acquiring pressure data through the plurality of pressure sensors;
the model construction module adopts a multilayer feedforward neural network to construct a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance according to the pressure data;
the incoming flow velocity model optimization module adopts a mean square error and a decision coefficient to evaluate a prediction regression model of the incoming flow velocity to obtain an optimal prediction regression model of the incoming flow velocity;
and the wall-to-wall distance model optimization module evaluates the prediction regression model of the wall-to-wall distance by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the wall-to-wall distance.
The technical solution of the present invention is further described below with reference to specific examples.
1. Theoretical analysis
The fish lateral line system comprises a surface neural hill and a lateral conduit neural hill which are respectively used for sensing speed and acceleration (related to pressure) signals of fluid, transmitting the micro information of the space-time dynamic variation to a central nerve hub, providing instant orientation and environmental water dynamic information for a fish body, assisting the fish body to adjust the behavior mode of the body and achieving the purpose of adapting to the environment. Considering an incompressible, isothermal newtonian fluid (density ρ, viscosity μ), the navier-stokes equation can be expressed as:
Figure RE-GDA0003138301760000081
according to the above formula, pressure
Figure RE-GDA0003138301760000082
And momentum
Figure RE-GDA0003138301760000083
The pressure value is increased due to the reduction of the speed, so that the incoming flow speed can be estimated by collecting and analyzing the change of the pressure value of the surface of the fish body. In addition, the fish body actively and symmetrically fluctuates and the passive motion of the fish body under the action of fluid causes the periodic change of the surrounding flow field, thereby influencing the body surface pressure distribution of the fish body. The lateral force R acting on a fish body per unit length can be expressed as:
Figure RE-GDA0003138301760000084
wherein m (x) is the virtual mass of the fish body per unit length, and w (x, t) is the lateral movement velocity of the fish body relative to the fluid. When the wave is close to the wall surface, the fish body pushes the fluid to move towards the side wall surface, the fluid is blocked by the side wall surface, the speed is reduced, the value of w (x, t) is increased, and the lateral force R is increased. Therefore, the wall-against distance can be predicted and estimated by detecting the body surface pressure difference value of the symmetrical positions of the two sides of the fish body.
2. Data acquisition and processing
2.1 simulation modeling
And (3) performing simulation calculation on the near-wall surface wave propulsion process of the bionic robot fish by adopting a Computational Fluid Dynamics (CFD) method. As shown in fig. 2, the simulation calculation model has a flow field inlet on the left, an incoming flow velocity v, and a flow field outlet on the right. Simulating the machine fish to do wave motion at a position d away from the side wall surface by using a two-dimensional Joukowski wing profile with the length of L, wherein the adopted carangidae wave equation is expressed as follows:
Figure RE-GDA0003138301760000085
wherein x is a coordinate in the body length direction; a (x) is the amplitude envelope of the transverse motion, and the coefficient is a0＝0.02，a1＝-0.008，a20.16; y (x, t) is the lateral displacement at time x at t; k is 2 pi/lambda is the wave number of the bulk wave, and lambda is the wavelength of the bulk wave; f is the tail fin oscillation frequency.
In order to collect the flow field pressure change information, as shown in fig. 3, a series of virtual pressure sensors are configured on the body surface of the biomimetic robotic fish, and a flow field identification ALL system is constructed for extracting real-time body surface pressure data in the wave propulsion process of the biomimetic robotic fish. Wherein the head pressure sensor is marked as S1The pressure sensors uniformly and symmetrically distributed along the length direction are recorded as
Figure RE-GDA0003138301760000091
SiLIs a body left side pressure sensor, SiRIs a body right side pressure sensor.
Considering the influence of the inflow velocity v, the wall-leaning distance d and the fluctuation frequency f on the body surface pressure of the bionic robot fish, the invention carries out a series of parameterized simulation experiments. Table 1 shows the flow field simulation parameters. As shown in table 1, the incoming flow velocity ranges from 0 to 1.0m/s, the wall approach distance ranges from 0.1 to 0.8L, and the state without the wall effect is represented by d ═ 2L. The fluctuation frequency is 0.5-2.5Hz, and the corresponding ALL sampling frequency is 0.5-2.5 KHz. t is tiTime of day, head pressure sensor S1The pressure data collected is noted as P(s)1,ti) Pressure sensor group SiThe collected pressure data is expressed as
Figure RE-GDA0003138301760000092
The overall pressure data collected during the test period T may be representativeComprises the following steps:
Figure RE-GDA0003138301760000093
in order to eliminate the influence of the fluctuation motion of the fish body on the lateral pressure component, the head pressure sensor S is provided1Averaging the collected pressure data while simultaneously averaging the sensor set SiThe collected pressure data are summed and averaged to obtain:
Figure RE-GDA0003138301760000094
further, the pressure data collected by each pressure sensor is subjected to non-dimensionalization processing, and the obtained normalized pressure coefficient is represented as:
Figure RE-GDA0003138301760000101
wherein U ═ λ f.
Figure RE-GDA0003138301760000102
2.2 simulation results analysis
Fig. 4 shows the distribution clouds of the flow field structure of the biomimetic robotic fish at different times when the biomimetic robotic fish moves near the wall (d is 0.2L). From the pressure field cloud chart (see fig. 4(a)), it can be known that a low pressure area is always present between the fish body and the wall surface during the whole movement period. Under the influence of the wall effect, the pressure field formed by the symmetrical fluctuation of the fish body presents asymmetrical distribution. Similarly, as shown in the speed field cloud chart of fig. 4(b), a high-speed region always exists between the fish body and the wall surface due to the wall surface effect, and the speed field also presents an asymmetric distribution. The asymmetric distribution of the pressure field and the velocity field provides a basis for the identification of the near-wall environment.
Head sensor S1And a sensor group Si(i-2, …,11) at different wall-contact distances d and incoming flow ratesThe pressure coefficient acquired and processed at v is squared to obtain a variance, and a pressure coefficient variance curve as shown in fig. 5 can be obtained. As can be seen from fig. 5(a), the head sensor S has a given incoming flow velocity range (v ═ 0.2 to 1.0m/S)1The pressure coefficient variance values collected at different wall-leaning distances are increased along with the increase of the flow speed, and the sensor group (S) in the middle of the body length direction2-S6) The variance value of the acquired pressure coefficient is kept at a low level, and a sensor group (S) at the tail part7-S11) The acquired pressure coefficient variance value is changed in an ascending trend. As can be seen from fig. 5(b), within a given wall-to-wall distance range (d ═ 0.2-0.8L), the variation trend of the pressure coefficient variance values collected by the sensor array at different incoming flow velocities along the body length direction is consistent with fig. 5 (a). The variance of the data samples reflects the discrete degree of the data, the identification degree of the sensor on the flow field parameters can be measured by utilizing the variance value of the pressure coefficient acquired by the sensor, the larger the variance value is, the more sensitive the sensor group is to the flow field parameter change is, and the larger the information weight of the sensor group in the whole array is. Therefore, the layout and the number of the sensors can be optimized according to the variance value of the pressure data collected by the sensor group.
3. Neural network modeling
3.1 Multi-layer feedforward neural network architecture analysis
According to pressure data collected by a simulation experiment, a prediction regression model of the incoming flow speed and the wall-approaching distance is established by adopting a multilayer feedforward neural network, and the incoming flow speed and the wall-approaching distance of the bionic robot fish swimming close to the wall surface are predicted. The structural parameters of the multi-layer feedforward neural network comprise the number of input data features, the number of hidden layers, the number of hidden layer neurons and the selection of activation functions of the hidden layers and the output layers. Table 2 shows the structural parameters of the multi-layer feedforward neural network. As shown in table 2, two prediction regression models of the incoming flow velocity and the wall approach distance are established, the ReLU function is used as the hidden layer activation function, and the output layer uses the linear activation function. To find an optimized neural network structure, the number of hidden layer layers is gradually increased from 1 to 5, and the number of neurons in the first hidden layer is gradually increased from the number of input features to 3 times. The neuron number of each hidden layer is configured in a descending rule, and the neuron number of the next layer is 2/3 of the previous layer.
TABLE 2
Figure RE-GDA0003138301760000111
Using Mean-Square Error (MSE) and coefficient of determination (R)2) Evaluating the network structure of different configurations:
Figure RE-GDA0003138301760000112
wherein the content of the first and second substances,
Figure RE-GDA0003138301760000113
Yi,
Figure RE-GDA0003138301760000114
the predicted value, the observed value, and the mean value are respectively represented.
As shown in FIG. 6, the incoming flow velocity prediction regression model adopts evaluation indexes of different structural parameters, and it can be known from the figure that the hidden layer number and hidden layer neuron number pair R in a given structural parameter range2And the influence of MSE is small, and the structure of the optimal incoming flow velocity neural network prediction regression model is determined by considering the complexity of the structural parameters of the model as follows: 12-36-1.
As shown in FIG. 7, R is an evaluation index obtained by using different structural parameters in the wall distance prediction regression model, and is shown in FIG. 7(a)2The number of hidden layer neurons is changed to R along with the increase of the number of hidden layers2Has little effect. As can be seen from fig. 7(b), MSE shows a significant downward trend with the number of hidden layers, and the MSE is affected little by the number of hidden layer neurons. Comprehensively considering the model evaluation index and the model parameter complexity, and determining the optimal structure of the wall-dependent distance prediction regression model as follows: 13-13-8-5-3-2-1.
3.2 data feature reduction
And reducing the characteristics of the input data by adopting a characteristic variable stepwise elimination method according to the analysis result of the pressure coefficient variance value acquired by the manual lateral line at the 2.2 node. Table 3 is a data feature elimination order list. As shown in table 3, the pressure coefficient variance values collected by the pressure sensors are gradually eliminated from small to large according to the corresponding input data characteristics.
TABLE 3
Figure RE-GDA0003138301760000121
Fig. 8 is a graph showing the influence of the number of removed data features on the evaluation index in the incoming flow velocity prediction regression model. As can be seen from FIG. 8(a), when the number of eliminated data features is 6 or less, R is found on the training set and the test set2Remains substantially stationary, R with further increase in the number of eliminated data features2And the change is in a rapid descending trend. It can also be seen that for MSE (see fig. 8(b)), the change is small when the number of eliminated data features is 6 or less, and changes in a rapidly increasing trend when it is greater than 6. From this, the data feature sequence S5,S6,S4,S7,S3,S8The influence on the prediction effect of the incoming flow velocity prediction regression model is small, and the optimized feature set after the features are eliminated is { S }2,S9,S10,S11,S1,f}。
FIG. 9 is a graph showing the influence of the number of feature removals of the regression model data on the evaluation index by the wall distance prediction. As can be seen from FIG. 9(a), when the number of eliminated data features is 5 or less, R is found on the training set and the test set2The variation is small, and when the number of eliminated data features is greater than 5, R on the training set and the test set2And the change is in a rapid descending trend. Similarly, as can be seen from fig. 9(b), when the number of eliminated data features is greater than 5, the MSE on the test set changes in a rapidly increasing trend. From this, the data feature sequence S3,S4,S2,S5,S6The influence on the prediction effect of the prediction regression model by the wall distance is little, and the influence should be eliminated, so that the optimized model input data can be obtained finallyThe feature set is { S7,S8,S9,S1,S10,S11, f,v}。
Fig. 10 and 11 show the comparison of the prediction effects of the incoming flow velocity and the wall approach distance prediction regression model before and after the elimination of the data feature, respectively. As can be seen from FIG. 10, the incoming flow velocity prediction regression model is very effective in predicting the incoming flow velocity in a given range (R)20.998), the predicted effect remains substantially unchanged after 6 data features are eliminated (R)20.994). As can be seen from fig. 11, the near-wall distance prediction regression model has a good effect of predicting a state near the wall surface, but the prediction effect of a state far from the wall surface (d 2L) is poor (R is 2L)20.912). After 5 weakly correlated data features are eliminated, the prediction effect is not changed greatly (R)2＝0.883)。
The invention provides a near-wall surface fluctuation propelling bionic robot fish flow field identification method based on an artificial lateral line, which adopts a computational fluid dynamics method to develop a parametric simulation experiment, collects body surface pressure data of the bionic robot fish under different flow field conditions, trains and establishes a flow field parameter prediction regression model based on a multilayer feedforward neural network, and optimizes a model structure and reduces data characteristics. The invention has the following advantages:
(1) the near-wall surface wave propulsion of the bionic robot fish can cause the asymmetrical distribution of the surrounding flow field structure, and provides a basis for flow field parameter identification based on artificial lateral lines.
(2) The pressure coefficient variance values of the side line pressure sensor array under different incoming flow speeds and wall-approaching distances are obtained, and the identification degrees of the pressure sensors at different positions on the flow field parameter change are disclosed.
(3) The influence of the number of hidden layers and the number of hidden layer neurons on the evaluation index of the incoming flow velocity prediction regression model is small, and the increase of the number of hidden layers of the wall distance prediction regression model causes R2Increasing the MSE decreases.
(4) The influence of the pressure sensor group on the prediction effect of the incoming flow speed and the wall-approaching distance along the middle part of the body length direction is small, and the influence is eliminated, and the result shows that the method provided by the invention has a better prediction effect on the incoming flow speed and the wall-approaching distance.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are within the scope of the invention.

Claims (10)

1. A bionic robot fish near-wall surface flow field identification method based on an artificial lateral line is characterized by comprising the following steps:
s1, configuring a plurality of pressure sensors on the bionic robot fish, and acquiring pressure data through the plurality of pressure sensors;
s2, constructing a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance by adopting a multi-layer feedforward neural network according to the pressure data;
s3, evaluating the prediction regression model of the incoming flow velocity by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the incoming flow velocity;
and evaluating the prediction regression model of the wall-to-wall distance by adopting the mean square error and the decision coefficient to obtain the optimal prediction regression model of the wall-to-wall distance.
2. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines as claimed in claim 1, further comprising, after S3:
acquiring and processing pressure coefficients obtained by a plurality of pressure sensors under different wall-approaching distances and different incoming flow speeds, and solving the variance to obtain a pressure coefficient variance curve;
and reducing the characteristics of the input data by adopting a characteristic variable stepwise elimination method according to the pressure coefficient variance curve, and optimizing a characteristic set.
3. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines as claimed in claim 1, wherein the S1 includes:
the head of the bionic robot fish is provided with a pressure sensor, and pressure sensor groups are uniformly arranged along the body length direction of the bionic robot fish;
acquiring pressure data through a plurality of pressure sensors to obtain an integral pressure data set, wherein the integral pressure data set comprises pressure data acquired by the pressure sensors at the head of the bionic robot fish and pressure data acquired by a pressure sensor group in the length direction of the bionic robot fish;
and averaging the pressure data acquired by the pressure sensors at the head of the bionic robot fish, and summing and averaging the pressure data acquired by the pressure sensor group in the length direction of the bionic robot fish.
4. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines as claimed in claim 1, wherein the S1 further comprises:
and carrying out non-dimensionalization processing on the pressure data acquired by the pressure sensor to obtain a normalized pressure coefficient.
5. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines in claim 1, wherein in S2, the structural parameters of the multilayer feedforward neural network include the number of input data features, the number of hidden layers, the number of hidden layer neurons, and the selection of activation functions of hidden layers and output layers.
6. The method for recognizing the near-wall flow field of the bionic robotic fish based on the artificial lateral line as claimed in claim 5, wherein a ReLU function is adopted as a hidden layer activation function of a multilayer feedforward neural network, and a linear activation function is adopted as an output layer of the multilayer feedforward neural network.
7. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines as claimed in claim 5, wherein the step S2 further comprises:
obtaining an optimized neural network structure, specifically:
gradually increasing the number of hidden layer layers from 1 to 5, gradually increasing the number of neurons of the first hidden layer from the number of input features to 3 times, configuring the number of neurons of each hidden layer in a descending rule, wherein the number of neurons of the next layer is 2/3 of the previous layer.
8. The method for recognizing the near-wall flow field of the biomimetic robotic fish based on artificial lateral lines as claimed in claim 1, wherein in S3,
mean square error
Figure FDA0003060613950000031
Determining coefficients
Figure FDA0003060613950000032
Wherein the content of the first and second substances,
Figure FDA0003060613950000033
Yi,
Figure FDA0003060613950000034
the predicted value, the observed value, and the mean value are respectively represented.
9. The method for recognizing the near-wall flow field of the bionic robot fish based on the artificial lateral line as claimed in claim 2, wherein the step-by-step elimination method of the characteristic variables is adopted to reduce the characteristics of the input data according to the pressure coefficient variance curve, and the optimization of the characteristic set specifically comprises the following steps:
and gradually eliminating corresponding input data characteristics from small to large according to the pressure coefficient variance value acquired by the pressure sensor.
10. The utility model provides a bionical machine fish near-wall flow field identification system based on artifical lateral line which characterized in that includes:
the data acquisition module is used for configuring a plurality of pressure sensors on the bionic robot fish and acquiring pressure data through the plurality of pressure sensors;
the model building module is used for building a prediction regression model of the incoming flow speed and a prediction regression model of the wall-to-wall distance by adopting a multi-layer feedforward neural network according to the pressure data;
the incoming flow velocity model optimization module evaluates the prediction regression model of the incoming flow velocity by adopting a mean square error and a decision coefficient to obtain an optimal prediction regression model of the incoming flow velocity;
and the wall-to-wall distance model optimization module evaluates the prediction regression model of the wall-to-wall distance by adopting a mean square error and a decision coefficient to obtain an optimal prediction regression model of the wall-to-wall distance.
CN202110511911.0A 2021-05-11 2021-05-11 Bionic robot fish near-wall flow field identification method and system based on artificial lateral line Active CN113255243B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110511911.0A CN113255243B (en) 2021-05-11 2021-05-11 Bionic robot fish near-wall flow field identification method and system based on artificial lateral line

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110511911.0A CN113255243B (en) 2021-05-11 2021-05-11 Bionic robot fish near-wall flow field identification method and system based on artificial lateral line

Publications (2)

Publication Number Publication Date
CN113255243A true CN113255243A (en) 2021-08-13
CN113255243B CN113255243B (en) 2022-07-01

Family

ID=77223997

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110511911.0A Active CN113255243B (en) 2021-05-11 2021-05-11 Bionic robot fish near-wall flow field identification method and system based on artificial lateral line

Country Status (1)

Country Link
CN (1) CN113255243B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117077579A (en) * 2023-10-17 2023-11-17 深圳十沣科技有限公司 Airfoil flow field prediction method, device, equipment and storage medium
CN117723771A (en) * 2023-12-15 2024-03-19 清华大学深圳国际研究生院 Speed measuring method for soft robotic fish and soft robotic fish

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105333988A (en) * 2015-11-25 2016-02-17 中国海洋大学 Artificial lateral line pressure detection method
CN110286592A (en) * 2019-06-28 2019-09-27 山东建筑大学 A kind of multi-modal movement technique of machine fish based on BP neural network and system
CN110488611A (en) * 2019-09-02 2019-11-22 山东建筑大学 A kind of biomimetic robot fish movement control method, controller and bionic machine fish

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105333988A (en) * 2015-11-25 2016-02-17 中国海洋大学 Artificial lateral line pressure detection method
CN110286592A (en) * 2019-06-28 2019-09-27 山东建筑大学 A kind of multi-modal movement technique of machine fish based on BP neural network and system
CN110488611A (en) * 2019-09-02 2019-11-22 山东建筑大学 A kind of biomimetic robot fish movement control method, controller and bionic machine fish

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
刘贵杰等: "基于人工侧线系统的动载体流场感知", 《兵工自动化》 *
谢鸥等: "仿生机器鱼靠壁游动的壁面效应特性研究", 《华中科技大学学报(自然科学版)》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117077579A (en) * 2023-10-17 2023-11-17 深圳十沣科技有限公司 Airfoil flow field prediction method, device, equipment and storage medium
CN117077579B (en) * 2023-10-17 2024-02-06 深圳十沣科技有限公司 Airfoil flow field prediction method, device, equipment and storage medium
CN117723771A (en) * 2023-12-15 2024-03-19 清华大学深圳国际研究生院 Speed measuring method for soft robotic fish and soft robotic fish

Also Published As

Publication number Publication date
CN113255243B (en) 2022-07-01

Similar Documents

Publication Publication Date Title
CN113255243B (en) Bionic robot fish near-wall flow field identification method and system based on artificial lateral line
Cheng et al. Path planning and obstacle avoidance for AUV: A review
Kruusmaa et al. Filose for svenning: A flow sensing bioinspired robot
CA3078530A1 (en) Gradient normalization systems and methods for adaptive loss balancing in deep multitask networks
CN113033119B (en) Underwater vehicle target area floating control method based on double-critic reinforcement learning technology
Liu et al. A new bionic lateral line system applied to pitch motion parameters perception for autonomous underwater vehicles
CN117198330B (en) Sound source identification method and system and electronic equipment
Zhou et al. Bio-inspired flow sensing and prediction for fish-like undulating locomotion: A CFD-aided approach
Djavareshkian et al. Neuro-fuzzy based approach for estimation of Hydrofoil performance
Wei et al. Trans-media resistance investigation of hybrid aerial underwater vehicle base on hydrodynamic experiments and machine learning
CN115329459A (en) Underwater vehicle modeling method and system based on digital twinning
Li et al. A new artificial lateral line attitude perception method based on mixed activation function-multilayer perceptron (MAF-MLP)
Zhang et al. Intelligent vector field histogram based collision avoidance method for auv
Qiu et al. Development of hybrid neural network and current forecasting model based dead reckoning method for accurate prediction of underwater glider position
CN111026145A (en) Ups and downs and gesture control system suitable for upper ocean robot
Yari et al. Applying the artificial neural network to estimate the drag force for an autonomous underwater vehicle
Sun et al. Fault diagnosis method of autonomous underwater vehicle based on deep learning
López-Barajas et al. Automatic Visual Inspection of a Net for Fish Farms by Means of Robotic Intelligence
Liu et al. Hydrodynamic modeling with grey-box method of a foil-like underwater vehicle
Maertens Fish swimming optimization and exploiting multi-body hydrodynamic interactions for underwater navigation
CN116562094B (en) AUV formation flow field prediction method based on PINN model
Awano et al. Human-robot cooperation in arrangement of objects using confidence measure of neuro-dynamical system
Li et al. The Performance Evaluation of Static Obstacle Recognition by Underwater Pressure Characteristics for Underwater Spherical Robot
Ardilla et al. Multi-Scale Batch-Learning Growing Neural Gas Efficiently for Dynamic Data Distributions
Xu et al. A novel artificial lateral line sensing system of robotic fish based on BP neural network

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