CN104238367A - Method for controlling consistency of vibration of surfaces of shell structures on basis of neural networks - Google Patents

Abstract

The invention discloses a method for controlling the consistency of vibration of the surfaces of shell structures on the basis of neural networks. The method includes firstly, constructing vibration consistency control architectures on the basis of a neural network optimization algorithm; secondly, deducing weight and threshold iterative formulas of identifiers and controllers on the basis of a gradient descent process; thirdly, further computing global errors and frequency-point errors, combining the global errors and the frequency-point errors with one another and creating a novel error judgment criterion for judging difference between control effects and targets; fourthly, synchronously controlling the consistency of vibration response of multiple points of the surfaces of the shell structures so as to cooperatively controlling the vibration response of the multiple points of the surfaces of the shell structures. Each control architecture mainly comprises two core modules which are the corresponding identifier and the corresponding controller respectively, the identifiers are used for identifying controlled shell models and predicting the vibration response, and the controllers are used for implementing excitation parameter optimization and control targets. The weight and threshold iterative formulas are used for optimizing and updating loop parameters.

Description

A kind of shell structure surface vibration consistance control method based on neural network

Technical field

The invention belongs to mechanical oscillatory structure response optimization control field, be specifically related to a kind of shell structure surface vibration consistance control method based on neural network.

Background technology

At present, Vibration Optimization and Control mainly concentrates on structure of modification, Passive Control and ACTIVE CONTROL at present.Structure of modification reproducibility is strong, but restricted large, very flexible, update cost are high; Passive Control is to increase equipment quality for cost, and low frequency inhibition of vibration is limited; ACTIVE CONTROL is the vibration signal according to detecting, adopts Real-Time Control Strategy, and drive ram applies power or moment to control object, thus reaches vibration optimization aim, has the advantages that adaptivity is strong, cost is low, efficiency is high, is therefore subject to extensive concern.

The research of Active Vibration Control mainly focuses on the suppression aspect of vibration at present, namely external forces is utilized to reduce the vibratory output of controlled device to greatest extent, but, along with the high speed of mechanized equipment, precise treatment and mute demand for development, Vibration Active Control is faced with new challenges: how to ensure that the maximization of mechanized equipment power source and minimizing of vibration noise become conflict.For Shell-Class structure, on the basis solving this kind of vibration problem, how to realize the consistance of body structure surface vibration noise radiation signal, ensure that the homogenising of body structure surface vibration noise becomes a difficult problem.

Summary of the invention

For the deficiencies in the prior art, the invention discloses a kind of shell structure surface vibration consistance control method based on neural network, the consistance that described method is used for shell structure vibration suppression and surface vibration controls, and it is characterized in that, said method comprising the steps of:

S100, structure control framework: described control framework comprises identifier, controller and controlled housing, and described identifier is used for the identification of controlled housing and the prediction of vibratory response, and described controller is for optimizing the realization of excitation parameters and control objectives;

S200, input stimulus: the intended vibratory response signal setting controlled hull vibration frequency domain; By initial excitation parameter, excitation is carried out to controlled housing and obtain vibration response signal;

S300, differentiation difference: global error combined with frequency error, instrument error passes judgment on criterion, for differentiating the difference of vibration response signal and intended vibratory response signal;

S400, termination: first target setting precision; Then the difference of vibration response signal and intended vibratory response signal and aimed at precision are compared; If both differences can not meet aimed at precision, then identifier and controller are optimized until both differences meet aimed at precision.

Technique effect of the present invention is:

1) achieve Shell-Class body structure surface vibration consistance to control, and easily extensible is in the vibration suppression of other mechanized equipments and Vibration Optimization and Control.

2) the Vibration Optimization and Control existence of solution of multi-source vibration coupling system is given.

3) construct the judgment of error criterion that overall error of frequency domain combines with frequency error, effectively improve stability and the anti-interference of control system.

Accompanying drawing explanation

Fig. 1 is vibration consistance control system frame diagram; In figure, NNC is controller, and NNI is identifier, the collection that " sampling " is data and collection process; " T → F " is for time domain is to the transfer process of frequency domain data; [D] is the response of surface of shell intended vibratory; [E] is intended vibratory response and the error signal controlling vibratory response; [F] is excitation parameters; The time domain vibratory response that y (t) is controlled housing; The frequency domain vibratory response that [Y] is controlled housing; for the controlled hull vibration response that identifier model exports.

Fig. 2 is the control system identifier network architecture; [F] is input layer; for output layer; [U] is the weights coefficient between input layer and output layer; [F]=[F ₁, F ₂, F ₃..., F _n] be excitation parameters; Y=[Y ₁, Y ₂..., Y _i] be actual vibration response.

Fig. 3 is control system controller network framework; [E] is input layer; [H] is hidden layer; [O] is output layer; [W] is the weights coefficient between input layer and hidden layer; [A] is the threshold coefficient between input layer and hidden layer; [V] is the weights coefficient between hidden layer and output layer; [B] is the threshold coefficient between hidden layer and output layer.

Fig. 4 is that time-domain signal contrasts figure with frequency-region signal consistance;

Fig. 4 (a) is time-domain signal;

Fig. 4 (b) is frequency-region signal;

Fig. 5 is overall error of frequency domain convergence key diagram;

Fig. 5 (a) is echo signal;

Fig. 5 (b) is real-time vibration signal;

Fig. 6 is frequency error convergence key diagram;

Fig. 6 (a) is echo signal;

Fig. 6 (b) is real-time vibration signal;

Fig. 7 is that cantilever shell experimental system arranges schematic diagram;

Fig. 8 is experiment effect comparison diagram;

Fig. 8 (a) is sampled point 1 initial vibration response spectrum;

Fig. 8 (b) is the controlled after vibration response spectrum of sampled point 1;

Fig. 8 (c) is sampled point 2 initial vibration response spectrum;

Fig. 8 (d) is the controlled after vibration response spectrum of sampled point 2;

Fig. 8 (e) is sampled point 3 initial vibration response spectrum;

Fig. 8 (f) is the controlled after vibration response spectrum of sampled point 3;

Fig. 8 (g) is sampled point 4 initial vibration response spectrum;

Fig. 8 (h) is the controlled after vibration response spectrum of sampled point 4;

Embodiment

Below in conjunction with drawings and the specific embodiments, content of the present invention is described in further detail:

A kind of shell structure surface vibration consistance control method of the present invention comprises the following steps:

(1) build vibration consistance and control overall architecture, instruct the overall operation of control system.

Figure 1 shows that the controlled body diagram of vibration consistance, core link comprises controller NNC and identifier NNI two parts.Wherein, identifier is responsible for identification and the response prediction of controlled housing, and the realization optimizing excitation parameters and control objectives is responsible for by controller.In addition, " sampling " refer to collection and the collection process of data; " T → F " refers to that time domain arrives the transfer process of frequency domain data; [D] is the response of surface of shell intended vibratory; [E] refers to intended vibratory response and the error signal controlling vibratory response; [F] refers to excitation parameters; Y (t) refers to the time domain vibratory response of controlled housing; [Y] refers to the frequency domain vibratory response of controlled housing; refer to the controlled hull vibration response that identifier model exports.

Whole control flow is: system cloud gray model starts, and first sets the frequency domain intended vibratory response signal [D] of hull vibration; Then by initial excitation parameter (amplitude, phase place, frequency), controlled housing is encouraged; Collected time domain vibration response signal y (t) of controlled housing by sensor and data acquisition unit, and by Fourier transform, the time domain vibration response signal collected is transformed into frequency domain [Y]; Trained as identifier input identifier by excitation parameters [F], identifier exports as system frequency domain vibratory response prediction the frequency domain vibratory response predict identifier and intended vibratory respond [D] and compare, if both meet aimed at precision at error, then termination of iterations, if both errors can not meet aimed at precision, error signal [E] input control device is optimized iteration, and output drive parameter encourages controlled housing.Circulation like this, until error signal meets aimed at precision.As can be seen from this circulation process, a Fourier transform is only used in each cyclic process, does not use inverse Fourier transform, therefore, effectively simplifies control flow, saves the control time.

(2) based on optimum algorithm of multi-layer neural network, orecontrolling factor System Discrimination device, the iteration optimization formula of derivation identifier parameter, identification system model, prediction vibratory response;

Figure 2 shows that identifier framework.Identifier is made up of double-layer structure: input layer [F] and output layer [U] is the weights coefficient between input layer and output layer.The workflow of identifier is: by excitation parameters [F]=[F ₁, F ₂, F ₃..., F _n] as input, by weights coefficient, obtain predicated response by predicated response y=[Y is responded with actual vibration ₁, Y ₂..., Y _i] compare, as both overall error of frequency domain and frequency error are all less than the accuracy requirement of target, then identification completes, if do not meet, continues iteration.The pass of input layer and output layer is:

$\left[\hat{Y}\right]=\left[U\right]\left[F\right]$

Weights coefficient between input layer and output layer is the key parameter that whole training iterative process is optimized, and the Optimized Iterative formula of weights coefficient [U] is:

$u_i(k+1)=u_i\left(k\right)+\frac{{\mathrm{\α}}_i\·e_i(k+1)\·F_i\left(k\right)}{\mathrm{\ϵ}+{F_i}^T\left(k\right)\·F_i\left(k\right)}$

In formula, e _ifor the error amount of identifier identification Output rusults and controlled housing actual vibration frequency domain response; a _i∈ (0,2) is decay factor; ε is very little non-zero constant, usually gets ε=0.001; K is iterative steps counter; F _ik () represents the kth step iteration input of identifier i-th input quantity; u _ik () represents the kth step iteration map function of identifier i-th input quantity;

(3) based on optimum algorithm of multi-layer neural network, orecontrolling factor device, the iteration optimization formula of derivation controller parameter, optimize output drive parameter, exciting is carried out to controlled housing, obtain new controlled after vibration response, so according to the circulation of control overall architecture, until reach control objectives.

Figure 3 shows that controller architecture.Controller is made up of three-layer network framework, comprises input layer [E]=[E ₁, E ₂..., E _i], hidden layer [H]=[H ₁, H ₂, H ₃..., H _j] and output layer [O]=[O ₁, O ₂..., O _k], the pass between three layers is:

[H]＝[E][W]+[A]

[O]＝[H][V]+[B]

Wherein, $\left[W\right]=\left[\begin{array}{cccc}{w}_{11}& w_{12}& \·\·\·& w_{1i}\\ w_{21}& w_{22}& \·\·\·& w_{2i}\\ \·\·\·& \·\·\·& \·\·\·& \·\·\·\\ w_{j1}& w_{j2}& \·\·\·& w_{\mathrm{ji}}\end{array}\right]$ For the weights coefficient between input layer and hidden layer; [A]=[a ₁, a ₂, a ₃..., a _j] be the threshold coefficient between input layer and hidden layer; $\left[V\right]=\left[\begin{array}{cccc}{v}_{11}& v_{12}& \·\·\·& v_{1i}\\ v_{21}& v_{22}& \·\·\·& v_{2i}\\ \·\·\·& \·\·\·& \·\·\·& \·\·\·\\ v_{j1}& v_{j2}& \·\·\·& v_{\mathrm{ji}}\end{array}\right]$ For the weights coefficient between hidden layer and output layer; [B]=[b ₁, b ₂..., b _k] be the threshold coefficient between hidden layer and output layer.

Controller hidden layer selects tangent S type transport function f (x)=1/ (1+e ^-x), output layer selects linear transfer function g (x)=x.Based on gradient descent method, the optimization formula that can obtain related coefficient between described controller input layer and hidden layer is:

w _ji(k+1)＝w _ji(k)+Δw _ji(k)

a _ji(k+1)＝a _ji(k)+Δa _ji(k)

Wherein: w _ji(k+1), w _jik () represents that k+1 walks jth row i-th column element of kth step iteration middle controller input layer and hidden layer weights coefficient matrix of coefficients [W] respectively; Δ w _jik () represents the adjustment amount of kth step iteration middle controller input layer and hidden layer weights matrix of coefficients [W] jth row i-th column element; a _ji(k+1), a _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller input layer and hidden layer threshold coefficient matrix [A] respectively; Δ a _jik () represents the adjustment amount of kth step iteration middle controller input layer and hidden layer threshold matrix [A] jth row i-th column element;

Δw _ji(k)＝ηδ _jE _i

Δa _ji(k)＝ηδ _j

${\mathrm{\δ}}_j=\underset{k}{\mathrm{\Σ}}(D_k-O_k)f^{\′}\left({\mathrm{net}}_k\right)V_{\mathrm{kj}}\·*f^{\′}\left({\mathrm{net}}_j\right)$

Wherein, E _irepresent i-th element of controller input vector E; η is the learning rate of controller, for controlling the knots modification of controller weights coefficient; D _krepresent a kth element of echo signal vector D; O _krepresent the kth element controlling output signal; V _kjrepresent the row k jth column element f ' (net of controller hidden layer and output layer weights matrix of coefficients [V] _k), f ' (net _j) represent the derivative of controller input layer and hidden layer, network function between hidden layer and output layer respectively;

Related coefficient between described controller hidden layer and output layer optimizes formula:

v _ji(k+1)＝v _ji(k)+Δv _i(k)

b _ji(k+1)＝b _ji(k)+Δb _ji(k)

Wherein: v _ji(k+1), v _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller hidden layer and output layer weights matrix of coefficients [V] respectively; Δ v _jik () represents the adjustment amount of kth step iteration middle controller hidden layer and output layer weights matrix of coefficients [V] jth row i-th column element; b _ji(k+1), b _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller input layer and hidden layer threshold coefficient matrix [B] respectively;

Δv _kj(k)＝ηδ _kH _j

Δb _ji(k)＝ηδ _k

${\mathrm{\δ}}_j=\underset{k}{\mathrm{\Σ}}(D_k-O_k)f^{\′}\left({\mathrm{net}}_k\right)$

Wherein, H _jrepresent a jth element of hidden layer output vector H;

(4) the judgment of error criterion that optimal control process iterates stops constructs based on overall error of frequency domain and frequency error.In addition, before control, need to explore the existence condition that multi-point cooperative consistance under coupled vibrations response condition controls to separate, instruct the selection of Sensor and actuator number.

The overall error of frequency domain J of described control system is the quadratic sum of housing intended vibratory response signal and real-time vibration response signal difference:

$J=\frac{1}{2}\underset{i}{\mathrm{\Σ}}{(D_i-Y_i)}^2$

In above formula, D _ifor i-th sequential value of controlled housing intended vibratory response signal D; Y _ifor i-th sequential value of controlled housing, vibration response signal;

The frequency error J of described control system ₁for intended vibratory response signal and real-time vibration response signal n target frequency difference in magnitude and:

$J_1=\frac{1}{n}\underset{n}{\mathrm{\Σ}}(D_i-Y_i)$

In above formula, D _ifor i-th sequential value of controlled housing intended vibratory response signal D; Y _ifor i-th sequential value of controlled housing, vibration response signal;

Described control system iteration optimization end condition is:

J≤err_goal1 and J ₁≤ err_goal2;

In above formula, err_goal1 characterizes the approximation ratio of the real-time vibration response signal of full frequency band and intended vibratory response signal; Err_goal2 characterizes the approximation ratio of characteristic frequency point amplitude in real-time vibration response signal and intended vibratory response signal; Stopping criterion for iteration characterizes the iteration ends when the difference of shell structure real-time vibration response signal Y and intended vibratory response signal D is less than err_goal1 and the difference of n target frequency is less than err_goal2.

The advantage constructing this iteration optimization end condition mainly contains following 2 points:

The first, the structure of overall error of frequency domain effectively can be saved the control time and make whole iterative process more steadily smooth.First, in the inventive method, intended vibratory response signal is vibrated frequency domain response signal with control and is all concentrated on frequency domain, and in each iterative process, only use a Fourier transform, therefore, the control and optimize time is effectively saved.Secondly, as shown in Figure 4, time-domain signal and frequency-region signal are not one-one relationship, may get well the corresponding same frequency-region signal of several time-domain signal.The structure of overall situation error of frequency domain can accurately reflect the relation between real-time frequency domain vibration information and target information can ensure the stationarity of control and optimize process.

The second, the structure of the iteration optimization end condition in the present invention effectively can improve adaptivity and the antijamming capability of algorithm.As shown in Figure 5, suppose that (a) is for echo signal, (b), for vibrate frequency-region signal in real time, under normal circumstances, our focus point is the amplitude of characteristic frequency point.If the amplitude of characteristic frequency point and target are closely and in the tolerance interval of the amplitude error of other frequencies, so this controls result is acceptable, i.e. Fig. 5.But if adopt separately overall error of frequency domain as stopping criterion for iteration, for this result, iterative optimization procedure cannot stop, and needs the time more grown go circulation or cause whole optimizing process to restrain.On the contrary, if adopt separately frequency error as stopping criterion for iteration, then unacceptable control result may be produced.Frequency error only focuses on the amplitude of characteristic frequency point, if this amplitude meets the demands, then stops optimizing.But as shown in Figure 6, if the amplitude of characteristic frequency point and target are enough close, and other frequency amplitudes are flooded by noise, adopt frequency error separately, then may produce this unacceptable control result.In addition, by the precision err_goal1 in suitable stopping criterion for iteration and err_goal2, reliability and the antijamming capability of algorithm can effectively be improved.

The present invention is directed to multi-source vibration coupling System Implementation, therefore, the existence condition that under coupled vibrations response condition, multi-point cooperative consistance controls to separate also must be probed into.

Be the system of n for one degree of freedom, can be expressed as:

$M\stackrel{\·\·}{x}+C\stackrel{\·}{x}+\mathrm{Kx}=F_{F}F+F_{v}v$

$F=(K_{1}y+K_2\stackrel{\·}{y})$

y＝F _yx

minJ(K ₁，K ₂，F _F，F _v)

Wherein, M, C, K are respectively the Mass matrix of system, damping battle array and Stiffness Matrix.X is the initial vibration response of system; F is Systematical control input; V is the original exciting force of system; F _fand F _vbe respectively the location matrix of control inputs and original exciting force; Y is the system vibration response of test; K ₁and K ₂be respectively the gain matrix of system displacement and speed; J is objective function.

Error feedback signal can be expressed as:

E＝[E ₁?E ₂?...?E _i]

Suppose T _vfor N _vthe vector that individual original exciting force amplitude is formed, T _ffor N _fthe vector that individual control inputs power amplitude is formed, therefore, error feedback signal can be transformed to further:

E＝P ₁T _v+P ₂T _f·D

Wherein, P ₁for original exciting force and intended vibratory respond between transition matrix, P ₂for control inputs power and intended vibratory respond between transition matrix.

Objective function can be transformed to:

$J=\underset{i=1}{\overset{i}{\mathrm{\Σ}}}{\left|E_i\right|}^2={T_v}^T{P_1}^T{P}_1{T}_v+T_f^T{P}_2^T{P}_1{T}_v+{T_v}^T{P_1}^T{P}_2{T}_f+T_f^T{P}_2^T{P}_2{T}_f-D$

Therefore, optimum control input is by order obtain:

${T_v}^{\mathrm{opt}}=-{\left(P_2^T{P}_2\right)}^{-1}{P}_2^T{P}_1{T}_v$

The minimum value of objective function is:

$J_{\min }={T_v}^T{P_1}^T[I-P_2\left(P_2^T{P}_2\right)]P_1{T}_v$

Therefore, there are three kinds of different situations for this optimization problem: if 1 error signal number is greater than control signal, then being a row non-singular matrix, there is non-zero only optimal solution in objective function; If 2 error signal numbers are relative with control signal, for non-singular matrix, objective function existence anduniquess null solution, if namely with the vibration of n actuator control n test response point, then can obtain the optimum vibratory response of n response point respectively; If 3 error signal numbers are less than control signal, for non-non-singular matrix, the optimum solution of objective function is not unique.Therefore, in the enforcement of the inventive method, ensure that actuator is consistent with the number of sensor.

Figure 7 shows that experimental system arrangenent diagram, experimental system is made up of cantilever shell structure, clamped frame, actuator and sensor.Experiment flow is: cantilever shell produces vibration after being subject to actuator initial excitation, vibratory response is gathered by sensor, excitation parameters and vibratory response are inputted controlled housing, loop optimization, export new excitation parameters, the difference that vibratory response under the new excitation of contrast and intended vibratory respond, if meet the demands, iteration ends, if do not meet, continue loop optimization, until reach target.

Figure 8 shows that the Vibration Optimization and Control result of four points in cantilever shell surface.Control objectives makes the vibratory response of four test points reach consistent, even if 70Hz place amplitude reaches 0.003,80Hz reach 0.004,90Hz place is 0.001,100Hz be 0.002,110Hz place is 0.003,120Hz place is 0.004,130Hz place be 0.001,150Hz place is 0.002.Control result shows, and the vibratory response of four test points in cantilever shell surface all well reaches target, reaches the requirement that shell structure surface vibration consistance controls.

The foregoing is only some embodiments of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1., based on a shell structure surface vibration consistance control method for neural network, the consistance that described method is used for shell structure vibration suppression and surface vibration controls, and it is characterized in that, said method comprising the steps of:

S100, structure control framework: described control framework comprises identifier, controller and controlled housing, and described identifier is used for the identification of controlled housing and the prediction of vibratory response, and described controller is for optimizing the realization of excitation parameters and control objectives;

S200, input stimulus: the intended vibratory response signal setting controlled hull vibration frequency domain; By initial excitation parameter, excitation is carried out to controlled housing and obtain vibration response signal;

S300, differentiation difference: global error combined with frequency error, instrument error passes judgment on criterion, for differentiating the difference of vibration response signal and intended vibratory response signal;

S400, termination: first target setting precision; Then the difference of vibration response signal and intended vibratory response signal and aimed at precision are compared; If both differences can not meet aimed at precision, then identifier and controller are optimized until both differences meet aimed at precision.

2. method according to claim 1, preferably, is characterized in that: the formation of described control framework also comprises acceleration transducer, data acquisition unit and Fourier transform unit.

3. method according to claim 2, is characterized in that:

Described step S200 comprises:

S201: by the time domain vibration response signal of acceleration transducer and the controlled housing of data acquisition unit acquires;

S202: the time domain vibration response signal collected is transformed into frequency domain by Fourier transform unit.

4. method according to claim 3, is characterized in that: in described step S100,

Identifier utilizes neural network to build, and its input/output relation is:

$\left[\hat{Y}\right]=\left[U\right]\left[F\right]$

In formula, [F] represents that identifier inputs, and is initial excitation parameter; for identifier exports, represent the vibration response signal of the controlled housing of identifier prediction; [U] represents identifier mapping function.

5. method according to claim 4, it is characterized in that, described S400 is specially: described identifier exported and the difference of described intended vibratory response signal and described aimed at precision compare, if described difference can not meet aimed at precision, then using the input of its difference as controller.

6. method according to claim 5, is characterized in that,

Described controller utilizes neural network to build, and its input/output relation is:

[H]＝[E][W]+[A]

[O]＝[H][V]+[B]

In formula, [W] is the weights coefficient between input layer and hidden layer; [E] is the input of controller input layer; [A] is the threshold coefficient between input layer and hidden layer; [V] is the weights coefficient between hidden layer and output layer; [H] is hidden layer output; [B] is the threshold coefficient between hidden layer and output layer; [O] is the output of controller output layer.

7. method according to claim 6, it is characterized in that, described S400 is specially: exported by described controller as in excitation input identifier, be optimized, until stop circulating when the difference of described vibration response signal and described intended vibratory response signal meets aimed at precision.

8. method according to claim 7, is characterized in that:

Optimization in described step S400 comprises optimization to identifier mapping function, to the optimization of related coefficient between controller input layer and hidden layer and the optimization to hidden layer and output layer related coefficient;

The optimization formula of the mapping function of identifier is:

$u_i(k+1)=u_i\left(k\right)+\frac{{\mathrm{\α}}_i\·e_i(k+1)\·F_i\left(k\right)}{\mathrm{\ϵ}+F_i^T\left(k\right)\·F_i\left(k\right)}$

In formula, e _ifor the error amount of identifier identification Output rusults and controlled housing actual vibration frequency domain response; α _i∈ (0,2) is decay factor; ε is very little non-zero constant, usually gets ε=0.001; K is iterative steps counter; F _ik () represents the kth step iteration input of identifier i-th input quantity; u _ik () represents the kth step iteration map function of identifier i-th input quantity;

Between described controller input layer and hidden layer, the optimization formula of related coefficient is:

w _ji(k+1)＝w _ji(k)+Δw _ji(k)

a _ji(k+1)＝a _ji(k)+Δa _ji(k)

Wherein: w _ji(k+1), w _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller input layer and hidden layer weights coefficient matrix of coefficients [W] respectively; Δ w _jik () represents the adjustment amount of kth step iteration middle controller input layer and hidden layer weights matrix of coefficients [W] jth row i-th column element; a _ji(k+1), a _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller input layer and hidden layer threshold coefficient matrix [A] respectively; Δ a _jik () represents the adjustment amount of kth step iteration middle controller input layer and hidden layer threshold matrix [A] jth row i-th column element;

Δw _ji(k)＝ηδ _jE _i

Δa _ji(k)＝ηδ _j

${\mathrm{\δ}}_j=\underset{k}{\mathrm{\Σ}}(D_k-O_k)f^{\′}\left({\mathrm{net}}_k\right)V_{\mathrm{kj}}\·*f^{\′}\left({\mathrm{net}}_j\right)$

Wherein, E _irepresent i-th element of controller input vector E; η is the learning rate of controller, for controlling the knots modification of controller weights coefficient; D _krepresent a kth element of echo signal vector D; O _krepresent the kth element controlling output signal; V _kjrepresent the row k jth column element f ' (net of controller hidden layer and output layer weights matrix of coefficients [V] _k), f ' (net _j) represent the derivative of controller input layer and hidden layer, network function between hidden layer and output layer respectively;

Related coefficient between described controller hidden layer and output layer optimizes formula:

v _ji(k+1)＝v _ji(k)+Δv _j(k)

b _ji(k+1)＝b _ji(k)+Δb _ji(k)

Wherein: v _ji(k+1), v _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller hidden layer and output layer weights matrix of coefficients [V] respectively; Δ v _jik () represents the adjustment amount of kth step iteration middle controller hidden layer and output layer weights matrix of coefficients [V] jth row i-th column element; b _ji(k+1), b _jik () represents jth row i-th column element of k+1 step and kth step iteration middle controller input layer and hidden layer threshold coefficient matrix [B] respectively;

Δv _kj(k)＝ηδ _kH _j

Δb _ji(k)＝ηδ _k

${\mathrm{\δ}}_j=\underset{k}{\mathrm{\Σ}}(D_k-O_k)f^{\′}\left({\mathrm{net}}_k\right)$

Wherein, H _jrepresent a jth element of hidden layer output vector H.

9. method according to claim 8, is characterized in that: the formula of global error described in described step S300 is:

$J=\frac{1}{2}\underset{i}{\mathrm{\Σ}}{(D_i-Y_i)}^2$

In above formula, D _ifor i-th sequential value of intended vibratory response signal D; Y _ifor i-th sequential value of controlled hull vibration response signal;

The formula of described frequency error is:

$J_1=\frac{1}{n}\underset{n}{\mathrm{\Σ}}(D_i-Y_i)$

10. method according to claim 9, is characterized in that:

The condition that described step S400 stops is:

J≤err_goal1 and J ₁≤ err_goal2;

In above formula, err_goal1 and err_goal2 is the aimed at precision of setting, and err_goal1 characterizes the approximation ratio of vibration response signal and intended vibratory response signal; Characteristic frequency point amplitude in err_goal2 sign vibration response signal and the approximation ratio of intended vibratory response signal.