CN101763033A - Device for self-correcting control for multi-model RBF neural network of deep submersible rescue vehicle and method thereof - Google Patents

Abstract

The invention provides a device for self-correcting control for a multi-model RBF neural network of a deep submersible rescue vehicle and a method thereof. The device comprises an electric compass, three high-frequency short baseline sonars, a DSP data processing system, a control computer, two vertical channel propellers, two horizontal channel propellers and two main propellers, wherein the three high-frequency short baseline sonars are connected with the DSP data processing system; the controlling computer is embedded with a thrust allocation logic, a filter and an algorithm for the self-correcting control for the multi-model RBF neural network; the electric compass is connected with the control computer by a serial port; the control computer is connected with the two vertical channel propellers, the two horizontal channel propellers and the two main propellers by a digital-analog conversion card of the control computer; and the DSP data processing system is connected with the control computer by a serial port. The invention has high control precision and can successfully complete the butt joint of the deep submersible rescue vehicle and a disabled submarine.

Description

Deep submergence rescue vessel multi-modal RBF neural network device for self-correcting control and method

Technical field

What the present invention relates to is a kind of deep submergence rescue vessel control device.Particularly a kind of multi-modal RBF neural network device for self-correcting control.The present invention also relates to a kind of deep submergence rescue vessel control method.

Background technology

The dynamic positioning of vessels system adopts pid control algorithm more at present, makes boats and ships or deep submergence rescue vessel remain on a certain desired locations dynamically.Though the PID design of Controller is simple, have good control performance in design conditions, PID control has a lot of limitation, is mainly reflected in environment when changing the off-design operating mode, and the control effect is even unstable with variation.And deep submergence rescue vessel with the docking operation of wrecking submarine in the middle of marine environment be more complicated, usually have a lot of disturbing factors and uncertain factor.Under a certain marine environment, there is the pid parameter of optimal control results may just can't reach the control effect of expectation in other cases like this, even can't finishes the work.In addition because marine environment is constantly to change, variation has taken place with regard to the hydrodynamic force coefficient that makes deep submergence rescue vessel in this, and the parameter of traditional PID control is fixed, the controlled variable of PID must be done corresponding adjustment after marine environment changed, and obviously adopted the very difficult dynamically positioning process of finishing deep submergence rescue vessel smoothly of PID control of preset parameter.

Find by literature search, Chinese patent publication number CN1776555A, title: based on the marine power positioning control method of fuzzy adaptive algorithm, it is applicable to waterborne vessel.This method has been described a kind of marine power positioning control method based on fuzzy adaptive algorithm, this system adjusts the rule of fuzzy control to reach best by the information in the control procedure, thereby reach adaptive purpose, there is system's static difference in this method when control.

Summary of the invention

The object of the present invention is to provide and a kind ofly when marine environment changes, also can satisfy the control accuracy requirement, thereby finish the multi-modal RBF neural network of the deep submergence rescue vessel that the docks device for self-correcting control of deep submergence rescue vessel and wrecking submarine smoothly.The present invention also aims to provide a kind of control method based on the multi-modal RBF neural network of deep submergence rescue vessel device for self-correcting control.

The object of the present invention is achieved like this:

The multi-modal RBF neural network of deep submergence rescue vessel of the present invention device for self-correcting control comprises a gyro compass, three high frequency short baseline sonars, a DSP data handling system, control computer, two vertical channels thrusters, two horizontal channel thrusters, two main thrusters, three high frequency short baseline sonars are connected with the DSP data handling system, be embedded with the thrust assignment logic in the control computer, wave filter and multi-modal RBF neural network self-correcting control algolithm, gyro compass is connected by serial ports with control computer, control computer by its digital-to-analog conversion card respectively with two vertical channels thrusters, two horizontal channel thrusters, two main thrusters link to each other, and the DSP data handling system is connected by serial ports with control computer.Wherein multi-modal RBF neural network self-correcting control algolithm is made of Multi-Mode PID Controller and RBF neural network, Multi-Mode PID Controller can be selected a suitable operation mode automatically according to the size of ocean current, and the RBF neural network is finely tuned the parameter of Multi-Mode PID Controller by the resulting Jacobian information of the dynamic model of identification deep submergence rescue vessel.

The multi-modal RBF neural network of deep submergence rescue vessel of the present invention device for self-correcting control can also comprise:

1, the described digital-to-analog conversion card that digital signal is converted to simulating signal be-10VDC～+ 12 D/A cards of 10VDC.

2, described DSP data handling system is the system based on the TMS320VC33PGA chip, and this system comprises several serial ports and an Ethernet interface.

Control method based on the multi-modal RBF neural network of deep submergence rescue vessel of the present invention device for self-correcting control is:

(1) three high frequency short baseline sonar is gathered the relative position value of survival craft and accident ship, give the DSP data handling system by the corresponding interface with data transfer, by resolving, calculate north orientation position, east orientation position and the height value of survival craft, send to control computer by serial ports with respect to the accident ship;

(2) bow of gyro compass collection survival craft sends to control computer with bow to the angle value by serial ports to the angle;

(3) wave filter filtering north orientation position, east orientation position, height value and the high frequency noise of bow in the numerical value of angle are determined the position and the course of deep submergence rescue vessel;

(4) the Multi-Mode PID control algolithm is adjusted the weight coefficient of each PID controller according to ocean current, and then definite its operation mode, finely tune the parameter of Multi-Mode PID Controller by the resulting Jacobian information of the dynamic model of RBF neural network identification deep submergence rescue vessel, determine a multi-modal RBF neural network self tuning controller like this, calculate needed controlled quentity controlled variable by this controller;

(5) the RBF neural network utilizes least square method and gradient descent method to adjust the parameter of self according to the input and output of deep submergence rescue vessel, make it approach the dynamic perfromance of deep submergence rescue vessel and obtain Jacobian information, adjust the parameter of Multi-Mode PID Controller by Jacobian information, the output steering order;

(6) by the thrust assignment logic steering order is converted to the thrust instruction that the actual thrust device needs, and thrust instruction sent to the digital-to-analogue conversion card, the digital-to-analogue conversion card is converted to analog quantity with digital quantity, thereby the control thruster produces thrust, makes deep submergence rescue vessel to the motion of the position of expectation.

Principle of the present invention is: multi-modal RBF neural network self-correcting control algolithm of the present invention is the core of dynamic positioning system, this algorithm is that the RBF neural network is combined with Multi-Mode PID Controller, determine the operation mode of controller according to ocean current, pass through the dynamic model of RBF neural network identification deep submergence rescue vessel then, with resulting Jacobian information, parameter to Multi-Mode PID Controller is finely tuned, and constitutes a multi-modal self tuning controller.

The controller of Multi-Mode PID Controller of the present invention for combining by a plurality of PID controls, wherein each PID control is optimum or the suboptimal control strategy under specific ocean current, these pid control parameters is constituted the controlled variable of Multi-Mode PID Controller in the mode of weighted sum.This controller can be adjusted the weight of each pid control parameter according to the difference of ocean current, thereby makes the bigger effect of controlled variable performance with the high PID controller of current ocean current matching degree, and the low less effect of pid control parameter performance of matching degree.With the proportional component is that example illustrates, Fig. 2 is the weighting function figure of the scale-up factor of designed each PID control of Multi-Mode PID Controller of the present invention, the scale-up factor that this figure has reflected five PID controls from A to E shared weight when the different ocean current, as A is that ocean current is to obtain the weight of scale-up factor of PID control of Optimal Control effect and the relation curve of ocean current at 0 o'clock, B is that ocean current is the weight of scale-up factor of the 0.5 joint PID control that the time can obtain the Optimal Control effect and the relation curve of ocean current, by that analogy.So, can utilize the mode of above-mentioned weighted sum to obtain a more satisfactory scale-up factor according to the size of actual ocean current.

Still be example with the scale-up factor, can express with following formula by the parameter of the Multi-Mode PID Controller of above-mentioned acquisition:

k _p＝ω ₁k _p1+ω ₂k _p2+ω ₃k _p3+...+ω _nk _pn

Here, k _P1Be the scale-up factor of integral-separated PID controller, it is the scale-up factor at the optimum of specific big or small ocean current or suboptimal control device, ω _n=ω _N0/ (ω ₁₀+ ω ₂₀+ ...+ω _N0) be the weight coefficient of the scale-up factor of n PID control, it is the normalized results of a plurality of weights. ${\mathrm{\&omega;}}_{\mathrm{<figure-callout\; id="0"\; label="n"\; filenames="H2009100734343E00032.png,H2009100734343E00041.png"\; state="\{\{state\}\}">n</figure-callout>}0}=\exp (-\frac{{\left|\right|X-{\mathrm{<figure-callout\; id="0"\; label="X\; n"\; filenames="H2009100734343E00032.png,H2009100734343E00041.png"\; state="\{\{state\}\}">X</figure-callout>}}_n\left|\right|}^2}{A})$ Be the weight of n PID control and the funtcional relationship of current ocean current.X is the size of current ocean current, X _N0Be the center of Gaussian function, pairing ocean current size when representing n PID control acquisition optimum or suboptimum effect.

Here the RBF neural network is used to approach the dynamic model of deep submergence rescue vessel, for the adjustment of sound stage width and center vector, and the learning method that the present invention uses gradient to descend, the performance index function that therefore defines identification is:

${\mathrm{<figure-callout\; id="1"\; label="J"\; filenames="H2009100734343E00012.png,H2009100734343E00042.png"\; state="\{\{state\}\}">J</figure-callout>}}_1=\frac{1}{2}{(y\left(k\right)-y_m\left(k\right))}^2$

${\mathrm{\&Delta;b}}_j=(y\left(k\right)-y_m\left(k\right))w_j{h}_j\frac{{\left|\right|X-C_j\left|\right|}^2}{{{\mathrm{<figure-callout\; id="3"\; label="b\; j"\; filenames="H2009100734343E00042.png,H2009100734343E00051.png"\; state="\{\{state\}\}">b</figure-callout>}}_j}^3}$

b _j(k)＝b _j(k-1)+ηΔb _j+α(b _j(k-1)-b _j(k-2))

$\mathrm{\&Delta;}{c}_{\mathrm{ji}}=(y\left(k\right)-y_m\left(k\right))w_j\frac{x_j-c_{\mathrm{ji}}}{{{\mathrm{<figure-callout\; id="2"\; label="b\; j"\; filenames="H2009100734343E00011.png,H2009100734343E00042.png"\; state="\{\{state\}\}">b</figure-callout>}}_j}^2}$

c _ji(k)＝c _ji(k-1)+ηΔc _ji+α(c _ji(k-1)-c _ji(k-2))

In the formula: η is a learning efficiency, and α is a factor of momentum.

For the weights adjustment of RBF neural network, the present invention uses the learning method of least square recursion, and the objective function of definition is:

$J\left(w_i\right)=\underset{p=1}{\overset{L}{\mathrm{\&Sigma;}}}{E}_p\left(w_i\right)=\frac{1}{2}\underset{p=1}{\overset{L}{\mathrm{\&Sigma;}}}{(y-y_m)}^2$

L is the length of sample in the formula.

For making the objective function minimum, i.e. the output y of RBF neural network _mLevel off to the output y of deep submergence rescue vessel, at this moment:

$\frac{\&PartialD;J\left(w_i\right)}{\&PartialD;w_i}=0$

Obtaining weights adjustment Recursive Least Squares is:

$w_i\left(k\right)=w_i(k-1)+\frac{P(k-1)\mathrm{\&Phi;}\left(k\right)}{\mathrm{\&beta;}+{\mathrm{\&Phi;}}^T\left(k\right)P(k-1)\mathrm{\&Phi;}\left(k\right)}[y\left(k\right)-y_m\left(k\right)]$

$P\left(k\right)=[P(k-1)-\frac{P(k-1)\mathrm{\&Phi;}\left(k\right){\mathrm{\&Phi;}}^T\left(k\right)P(k-1)}{\mathrm{\&beta;}+{\mathrm{\&Phi;}}^T\left(k\right)P(k-1)\mathrm{\&Phi;}\left(k\right)}]/\mathrm{\&beta;}$

β in the following formula is a forgetting factor, and y (k) is a RBF neural network desired output, also is the output of deep submergence rescue vessel, y _m(k) be the actual output of neural network, Φ is the input of neural network, exports the vector that y forms by controller output u and controlled device.

When deep submergence rescue vessel is disturbed, can finely tune the parameter of Multi-Mode PID Controller,, need to obtain Jacobian information in order to adjust the parameter of Multi-Mode PID Controller by the RBF neural network.

Jacobian information is:

$\frac{\&PartialD;y\left(k\right)}{\&PartialD;\mathrm{\&Delta;u}\left(k\right)}\&ap;\frac{{\&PartialD;y}_m\left(k\right)}{\&PartialD;\mathrm{\&Delta;u}\left(k\right)}=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{w}_j{h}_j\frac{c_{j1}-x_1}{{\mathrm{<figure-callout\; id="2"\; label="b\; j"\; filenames="H2009100734343E00011.png,H2009100734343E00042.png"\; state="\{\{state\}\}">b</figure-callout>}}_j^{}}$

The adjustment increment that obtains the Multi-Mode PID parameter according to Jacobian information is:

${\mathrm{\&Delta;k}}_p\left(k\right)=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{{\&PartialD;k}_p}=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;\mathrm{\&Delta;u}}\frac{\&PartialD;\mathrm{\&Delta;u}}{\&PartialD;k_p}=\mathrm{\&eta;e}\left(k\right)\frac{\&PartialD;y_m}{\&PartialD;\mathrm{\&Delta;u}}\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="H2009100734343E00012.png,H2009100734343E00042.png"\; state="\{\{state\}\}">e</figure-callout>}1$

${\mathrm{\&Delta;k}}_i\left(k\right)=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{{\&PartialD;k}_i}=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;\mathrm{\&Delta;u}}\frac{\&PartialD;\mathrm{\&Delta;u}}{\&PartialD;k_i}=\mathrm{\&eta;e}\left(k\right)\frac{\&PartialD;y_m}{\&PartialD;\mathrm{\&Delta;u}}\mathrm{<figure-callout\; id="2"\; label="e"\; filenames="H2009100734343E00011.png,H2009100734343E00042.png"\; state="\{\{state\}\}">e</figure-callout>}2$

${\mathrm{\&Delta;k}}_d\left(k\right)=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{{\&PartialD;k}_d}=-\mathrm{\&eta;}\frac{{\&PartialD;E}_p}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;\mathrm{\&Delta;u}}\frac{\&PartialD;\mathrm{\&Delta;u}}{\&PartialD;k_d}=\mathrm{\&eta;e}\left(k\right)\frac{\&PartialD;y_m}{\&PartialD;\mathrm{\&Delta;u}}\mathrm{<figure-callout\; id="3"\; label="e"\; filenames="H2009100734343E00042.png,H2009100734343E00051.png"\; state="\{\{state\}\}">e</figure-callout>}3$

k _p(k)＝k _p(k-1)+Δk _p(k)

k _i(k)＝k _i(k-1)+Δk _i(k)

k _d(k)＝k _d(k-1)+Δk _d(k)

In the formula:

Δu(k)＝k _pe1+k _ie2+k _de3

e1＝e(k)-e(k-1)

e2＝e(k)

e3＝e(k)-2e(k-1)+e(k-2)

e(k)＝r-y

The deep submergence rescue vessel power-positioning control system that adopts such scheme to design can be good at realizing the four-degree-of-freedom dynamically positioning of deep submergence rescue vessel, especially when marine environment or deep submergence rescue vessel self-characteristic change, multi-modal RBF neural network self tuning controller can be selected and the study of RBF neural network is effectively controlled deep submergence rescue vessel and finished the dynamically positioning process by the mode of Multi-Mode PID Controller.

Description of drawings

Fig. 1 is a deep submergence rescue vessel dynamic positioning system block scheme;

Fig. 2 is Multi-Mode PID Controller weighting function figure;

Fig. 3 is RBF neural network structure figure;

Fig. 4 is multi-modal RBF neural network self correcting system structural drawing;

Fig. 5 is a deep submergence rescue vessel thruster distribution plan;

Fig. 6 a to Fig. 6 d is deep submergence rescue vessel dynamically positioning hardware-in-the-loop simulation figure.

Embodiment

For example the present invention is done description in more detail below in conjunction with accompanying drawing:

Fig. 1 is the block scheme of control system, it has shown the annexation of control system and the transitive relation of information, three high frequency short baseline sonars 2,3 link to each other with DSP data handling system 5 by cable with 4, DSP data handling system 5 and control computer 6 are connected by serial ports, wave filter 9, thrust assignment logic 8 and multi-modal RBF neural network self-correcting control algolithm 7 embed in the control computer 6, gyro compass 1 and control computer 6 are connected by serial ports, control computer 6 by its digital-to-analog conversion card respectively with two vertical channels thrusters 10, two horizontal channel thrusters 11, two main thrusters 12 link to each other.

Fig. 2 is a Multi-Mode PID Controller weight coefficient functional arrangement, it has represented the weight coefficient of each PID control and the relation of ocean current, A is illustrated in optimum or the weight coefficient of suboptimum PID control and the funtcional relationship of ocean current under the situation that ocean current is 0Kn, and its explanation weight coefficient when current ocean current is 0Kn is to successively decrease according to rule under 1 other situation.In like manner, B is illustrated in optimum or the weight coefficient of suboptimum PID control and the funtcional relationship of ocean current under the situation that ocean current is 0.5Kn.C is illustrated in optimum or the weight coefficient of suboptimum PID control and the funtcional relationship of ocean current under the situation that ocean current is 1.0Kn.D is illustrated in optimum or the weight coefficient of suboptimum PID control and the funtcional relationship of ocean current under the situation that ocean current is 1.5Kn.E is illustrated in optimum or the weight coefficient of suboptimum PID control and the funtcional relationship of ocean current under the situation that ocean current is 2.0Kn.

Fig. 3 is the structural drawing of RBF neural network, Φ=[φ ₁, φ ₂..., φ _n] ^TBe the input vector of network, H=[h ₁, h ₂..., h _m] ^TBe the radially base vector of network, wherein h _jBe the gaussian basis function $h_j=\exp \left(\frac{{\left|\right|X-C_j\left|\right|}^2}{{2b}_j^2}\right)$ (j＝1，2，...，m)。The center vector of j node of network is C _j=[c _J1, c _J2..., c _Jn] ^T(j=1,2 ..., m).The sound stage width vector of network is B=[b ₁, b ₂..., b _m] ^T, b _jBe the sound stage width degree parameter of node j, and be number greater than zero.The weight vector of network is W=[w ₁, w ₂..., w _j..., w _m] ^TNetwork is output as y _m(k)=w ₁h ₁+ w ₂h ₂+ ...+w _mh _m

Fig. 4 is the structural drawing of multi-modal RBF neural network self-aligning control system, and r is an input vector among the figure, represent respectively bow that position that deep submergence rescue vessel will navigate to and deep submergence rescue vessel will keep to.U is the output of multi-modal RBF neural network self tuning controller, i.e. controlled quentity controlled variable.Y is the output of deep submergence rescue vessel, also be the physical location of deep submergence rescue vessel and bow to.y _mOutput for the RBF neural network.By y and y _mThe difference inner parameter of adjusting the RBF neural network make the dynamic model of its identification deep submergence rescue vessel and obtain Jacobian information, utilize the Jacobian information that obtains to come the Multi-Mode PID Controller parameter is finely tuned.

Fig. 5 is the location map of thruster, and it has described the particular location of thruster on deep submergence rescue vessel.Because controller only calculates steering order according to deviation, and does not consider the virtual rating of thruster, therefore need the thrust assignment logic that steering order is transformed to thrust (square) instruction, when carrying out the thrust assignment of logical, follow following principle:

1), the horizontal channel thruster adopts bow to priority principle, promptly priority allocation is changeed bow moment, remaining force satisfies traversing power again;

2), the angle of rake power of vertical channels just equals vertical control;

3), can't satisfy under the situation of changeing bow moment at the horizontal channel thruster, distributing to main thruster changes bow moment, and remaining power is distributed to longitudinal force.

Control method for position of the present invention is: the relative position value of 1. being gathered survival craft and accident ship by three high frequency short baseline sonars 2,3 and 4, give DSP data handling system 5 by the corresponding interface with data transfer, by resolving, calculate north orientation position, east orientation position and the height value of survival craft, send to control computer 6 by serial ports with respect to the accident ship; 2. the bow that provides survival craft by gyro compass sends to control computer 6 to the angle by serial ports; 3. by wave filter 9 filtering positions, height and the high frequency noise of bow in the numerical value of angle, determine the position and the course of deep submergence rescue vessel; 4. deliver in the multi-modal RBF neural network self-correcting control algolithm 7 to angle numerical value through filtered position, height and bow, calculate required control and moment; 5. the control that will be calculated by controller of thrust assignment logic 8 and torque command are distributed to the principle of preferential assurance according to bow and are obtained thrust (square) instruction, and two vertical channels thrusters 10, two horizontal channel thrusters 11 and two main thrusters 12 are issued in thrust (square) instruction; 6. thruster 10,11 and 12 produces actual thrust by thrust instruction, and deep submergence rescue vessel is moved to the accident ship, finally finishes docking of deep submergence rescue vessel and wrecking submarine.

Fig. 6 is deep submergence rescue vessel dynamic positioning system hardware-in-the-loop simulation result.Simulated conditions is the conventional PID controllers parameter of adjusting under 1 joint ocean current; 5 pid control parameters forming Multi-Mode PID Controller are to adjust under the situation of 0Kn, 0.5Kn, 1Kn, 1.5Kn and 2.0Kn stream respectively, can obtain optimum or suboptimal control effect under the condition of ocean current separately, the coefficient of Multi-Mode PID Controller is determined by the function among Fig. 2.Current direction is made as 160 degree during emulation, and ocean current was increased to 2 joints from 1 joint in 50 seconds, and the deep submergence rescue vessel initial position is: bow is to 0 degree, 0 meter of north orientation position, 0 meter of east orientation position, 25 meters of the degree of depth; The position of expectation is: bow is to-20 degree, 5 meters of north orientation positions, east orientation position-3 meter, 30 meters of the degree of depth.That the simulation curve dotted line is represented is the result who uses traditional PID control, that solid line is represented is the result of multi-modal RBF neural network self-correcting control, by simulation curve as can be seen traditional PID control to control effect under the situation that ocean current changes relatively poor, and multi-modal RBF neural network self-correcting control can reach ideal control effect, has effectively resisted because ocean current changes the disturbance that deep submergence rescue vessel is caused.

Claims (4)

1. the multi-modal RBF neural network of a deep submergence rescue vessel device for self-correcting control comprises a gyro compass (1), three high frequency short baseline sonars (2,3,4), a DSP data handling system (5), control computer (6), two vertical channels thrusters (10), two horizontal channel thrusters (11), two main thrusters (12), it is characterized in that: three high frequency short baseline sonars (2,3,4) be connected with DSP data handling system (5), be embedded with thrust assignment logic (8) in the control computer (6), wave filter (9) and multi-modal RBF neural network self-correcting control algolithm (7), gyro compass (1) is connected by serial ports with control computer (6), control computer (6) by its digital-to-analog conversion card respectively with two vertical channels thrusters (10), two horizontal channel thrusters (11), two main thrusters (12) link to each other, and DSP data handling system (5) is connected by serial ports with control computer (6).

2. the multi-modal RBF neural network of deep submergence rescue vessel according to claim 1 device for self-correcting control is characterized in that: the described digital-to-analog conversion card that digital signal is converted to simulating signal is-10VDC～+ 12 D/A cards of 10VDC.

3. the multi-modal RBF neural network of deep submergence rescue vessel according to claim 1 and 2 device for self-correcting control, it is characterized in that: described DSP data handling system is the system based on the TMS320VC33PGA chip, and this system comprises several serial ports and an Ethernet interface.

4. control method based on the multi-modal RBF neural network of deep submergence rescue vessel of the present invention device for self-correcting control is characterized in that:

(1) three high frequency short baseline sonar is gathered the relative position value of survival craft and accident ship, give the DSP data handling system by the corresponding interface with data transfer, by resolving, calculate north orientation position, east orientation position and the height value of survival craft, send to control computer by serial ports with respect to the accident ship;

(2) bow of gyro compass collection survival craft sends to control computer with bow to the angle value by serial ports to the angle;

(3) wave filter filtering north orientation position, east orientation position, height value and the high frequency noise of bow in the numerical value of angle are determined the position and the course of deep submergence rescue vessel;

(4) the Multi-Mode PID control algolithm is adjusted the weight coefficient of each PID controller according to ocean current, and then definite its operation mode, finely tune the parameter of Multi-Mode PID Controller by the resulting Jacobian information of the dynamic model of RBF neural network identification deep submergence rescue vessel, determine a multi-modal RBF neural network self tuning controller like this, calculate needed controlled quentity controlled variable by this controller;

(5) the RBF neural network utilizes least square method and gradient descent method to adjust the parameter of self according to the input and output of deep submergence rescue vessel, make it approach the dynamic perfromance of deep submergence rescue vessel and obtain Jacobian information, adjust the parameter of Multi-Mode PID Controller by Jacobian information, the output steering order;

(6) by the thrust assignment logic steering order is converted to the thrust instruction that the actual thrust device needs, and thrust instruction sent to the digital-to-analogue conversion card, the digital-to-analogue conversion card is converted to analog quantity with digital quantity, thereby the control thruster produces thrust, makes deep submergence rescue vessel to the motion of the position of expectation.

Images