CN106200655A - The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot - Google Patents
The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot Download PDFInfo
- Publication number
- CN106200655A CN106200655A CN201610479531.2A CN201610479531A CN106200655A CN 106200655 A CN106200655 A CN 106200655A CN 201610479531 A CN201610479531 A CN 201610479531A CN 106200655 A CN106200655 A CN 106200655A
- Authority
- CN
- China
- Prior art keywords
- module
- btt
- guided missile
- controller
- input
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 17
- 230000001537 neural effect Effects 0.000 claims abstract description 28
- 239000011159 matrix material Substances 0.000 claims description 90
- 230000007935 neutral effect Effects 0.000 claims description 18
- 238000009795 derivation Methods 0.000 claims description 8
- 238000013459 approach Methods 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims 1
- 238000004088 simulation Methods 0.000 abstract description 14
- 230000000694 effects Effects 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 73
- 230000008901 benefit Effects 0.000 description 11
- 230000005284 excitation Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 8
- 238000005070 sampling Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 5
- 230000004044 response Effects 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 3
- 230000007306 turnover Effects 0.000 description 3
- 241000196324 Embryophyta Species 0.000 description 2
- XCWPUUGSGHNIDZ-UHFFFAOYSA-N Oxypertine Chemical compound C1=2C=C(OC)C(OC)=CC=2NC(C)=C1CCN(CC1)CCN1C1=CC=CC=C1 XCWPUUGSGHNIDZ-UHFFFAOYSA-N 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241000176964 Mononeuron Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006757 chemical reactions by type Methods 0.000 description 1
- 238000012885 constant function Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 101150056779 rbf-1 gene Proteins 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/08—Control of attitude, i.e. control of roll, pitch, or yaw
- G05D1/0808—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft
- G05D1/0816—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability
- G05D1/0825—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability using mathematical models
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/107—Simultaneous control of position or course in three dimensions specially adapted for missiles
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Automation & Control Theory (AREA)
- Remote Sensing (AREA)
- Aviation & Aerospace Engineering (AREA)
- Evolutionary Computation (AREA)
- Radar, Positioning & Navigation (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses the FPGA implementation method of a kind of BTT guided missile Neural Network Inversion automatic pilot: first, establish the state equation model of BTT guided missile；Secondly RBF neural method is selected to compensate the modeling error of BTT guided missile state equation；Then utilize Backstepping method to derive control input, thus design the controller of BTT guided missile stabilitization；Finally, resolve to transplant by guided missile model and realize in FPGA, when automatic pilot runs, input expectation attitude signal and initial state vector are in controller, control input i.e. angle of rudder reflection is calculated by controller, it is sent in Models computed device resolve the new attitude information drawing BTT guided missile, more all status signals are stored in memorizer FIFO, be consequently formed circulation；Simulation result shows that this automatic pilot not only achieves preferable control effect but also substantially reduces simulation time, it is possible to meet the requirement of real-time.
Description
Technical field
The present invention relates to missile guidance and control technical field, be specifically related to based on RBF neural-Backstepping side
The FPGA design of the BTT missile autopilot of method and realization.
Background technology
For BTT (Bank-to-turn) guided missile, when designing the controller of automatic pilot it is noted that following three points:
(1) method using multivariable Control；(2) robustness of controller is considered；(3) its stability is improved.The control of BTT guided missile at present
In theory processed: method for designing based on classical control theory can only use in the case of angular velocity in roll is less, otherwise can lead
Cause to control to lose efficacy；Controller based on LQR design only just has in the case of mathematical model is accurate and preferably controls effect,
When there is perturbation or there is distracter in controll plant it cannot be guaranteed that the stability of controller；Robust control theory is at the beginning of design
Just the uncertainty of model is taken into account, but when model parameter changes greatly, this controller just shows preferably
Simulated effect and robustness.
Backstepping method is theoretical based on Modern Geometry, is that the one for strict feedback systems proposition is for the most true
Determining the controller design synthesis method of service system, it is often composed promise husband function (Control with Li Ya by modern control theory
Lyapunov Function, CLF) combine, for solving the control problem of the big system of non-linear, multivariate, close coupling, because of
This is very suitable for the design of BTT STT missile device, but when pattern function is unknown, there is distracter or controll plant generation parameter
During perturbation, control effect poor.And Intelligent Control Theory is as a focus of current control field, it is applied to extensive, multiple
The miscellaneous control with uncertain system.In recent years, during intelligent control technology the most more and more occurs in STT missile field.
Digital autopilot is the automatic pilot realized by digital processing chip, conventional digital processing chip
There are PC, single-chip microprocessor MCU (Microcontroller Unit is called for short MCU), digital signal processor DSP (Digital
Signal Processor), arm processor (Acorn RISC Machine, be called for short ARM) and on-site programmable gate array FPGA
(FieldProgrammable Gate Array is called for short FPGA).Complex in order to realize higher processing speed and carrying
System, the design of current digital autopilot typically uses two chip blocks to complete to coordinate collocation, one of chip
Completing control method as main control unit to resolve, another block realizes data acquisition and servo driving as coprocessor.
Based PC/104 module and the framework of MCU, its cpu i/f ability, need to use more peripheral interface device
Coordinating, and volume is big, power consumption is high, thus is gradually replaced by other frameworks；Framework based on DSP and MCU, its DSP core
Sheet control ability is more weak, lacks the support of common software, is unfavorable for realizing the design of complication system, thus applies model in practice
It is with limit；In framework based on ARM embedded microprocessor, ARM is also a micro-place grown up on the basis of MCU
Reason device.FPGA reliability and real-time are the highest, and possess " restructural " characteristic, it is simple to the method upgrading in later stage updates.This
Outward, FPGA is used to replace MCU to complete high-speed data computing required time as association's controller shorter.
Summary of the invention
For above-mentioned the deficiencies in the prior art, it is an object of the invention to provide a kind of BTT guided missile Neural Network Inversion certainly
The FPGA implementation method of dynamic pilot, and make this automatic pilot have preferable transient state and steady-state behaviour and robustness
By force.
In order to achieve the above object, present invention employs techniques below scheme:
1) state equation model of BTT guided missile is set up；
2) RBF neural method is selected to compensate the modeling error of BTT guided missile state equation；Then utilize
Backstepping method derives control input, thus designs BTT missile attitude control device based on FPGA model；
3) guided missile model is resolved transplant and realize in FPGA, FPGA completes whole closed loop feedback, sets up out BTT
Missile autopilot；When automatic pilot runs, input expectation attitude signal and initial state vector in attitude controller,
Calculated control input by attitude controller, utilize control input to carry out Models computed, draw the new attitude information of BTT guided missile,
New attitude information is sent in attitude controller, is consequently formed circulation；Export the attitude angle of BTT guided missile when loop ends, follow
The number of times of ring is determined by the sample size expecting attitude angle.
Described attitude controller uses RBF neural to approach the indeterminate Δ of BTT guided missile state equation model1
(x1) and Δ2(x1,x2), to make up modeling error:
Wherein, x1=[α β φ]T∈R3, x2=[p q r]T∈R3, α is the angle of attack, and β is yaw angle, and φ is roll angle, p
For angular velocity in roll, q is rate of pitch, and r is yaw rate, and H is the hidden layer output matrix of neutral net, and Q is neural
Connection weight value matrix between hidden layer and the output layer of network,WithIt is that indeterminate approaches value.
Described list-directed input list is shown as:
Wherein, f2It is about α, the function of β, p, q, r, g1It is about α, the function of β, φ, θ, g2It is the function about α, e1
It is the error of state vector and the control command signal fed back, e2It is the error of state variable and dummy pilot signal,
It is dummy pilot signal x2dDerivation result.
Described attitude controller includes input state variable module, the first RBF neural module, the 2nd RBF nerve net
Network module, the first matrix operations module, the second matrix operations module, Virtual Controller module and controller input module；Defeated
Enter state variable module by state variable value x1=[α β φ]TAnd x2=[p q r]TBe sent to the first matrix operations module,
One RBF neural module and the second matrix operations module；First matrix operations module calculates f1And g1Value and be sent to void
Intending controller module, the second matrix operations module calculates f2And g2Value and be sent to controller input module, the oneth RBF god
Calculate through mixed-media network modules mixed-mediaAnd it is sent to Virtual Controller module；Virtual Controller module calculates virtual controlling amount x2d
And it is sent to the second RBF neural module；Second RBF neural module calculatesAnd it is defeated to deliver to controller
Enter module；Controller input module calculates angle of rudder reflection and as controlling input.
Described Models computed selects four step Runge-Kutta.
The FPGA architecture of described automatic pilot includes that closed loop main body, described closed loop main body include attitude controller, storage
Device, the first buffer, the second buffer and Models computed module；Attitude controller reads the expectation attitude signal in memorizer
And the laggard row operation of original state signal that second in buffer, operation result sends into the first buffer, Models computed module
Read data in the first buffer and carry out Models computed, resolve the state vector obtained and send into the second buffer, complete closed loop anti-
Feedback.
Beneficial effects of the present invention is embodied in:
When the present invention designs the controller of BTT missile autopilot, on the basis of using Backstepping method,
Model error and Parameter Perturbation are taken into account and utilizes RBF neural to go compensation model error stable to strengthen controller
Property and robustness.The resolving of guided missile model is transplanted and is realized in FPGA by the present invention, constructs one based on FPGA single-chip
The digital autopilot of framework, carries out simulating, verifying to it, and result shows that this automatic pilot has preferably control
Also there is while effect the strongest real-time.
Accompanying drawing explanation
Fig. 1 digital autopilot based on FPGA architecture block diagram；
Fig. 2 digital autopilot based on FPGA architecture RTL view；
Fig. 3 is Back_ode module RTL view；
Fig. 4 is RBF-Backstepping controller hardware structured flowchart；
Fig. 5 is controller RTL view；
Fig. 6 is error module Error_gen structural representation；
Fig. 7 is error norm module Norm_k structural representation；
Fig. 8 is RBF neural hardware configuration；
Fig. 9 is dynamic memory module W_ram structural representation；
Figure 10 is right value update module Updata structural representation；
Figure 11 is Virtual Controller module X2d structural representation；
Figure 12 is trigonometric function module tri structural representation；
Figure 13 is controller module U structural representation；
Figure 14 is state machine module structural representation；
Figure 15 is Models computed module hardware structure chart；
Figure 16 is angle of attack aircraft pursuit course；
Figure 17 is yaw angle aircraft pursuit course；
Figure 18 is roll angle aircraft pursuit course；
Figure 19 is automatic pilot operational process block diagram.
Detailed description of the invention
The present invention is described in detail with embodiment below in conjunction with the accompanying drawings.
It is an object of the invention to design and realize based on Neural Network Inversion (Neural Network
Backstepping) BTT missile autopilot.
1, neutral net uses radial basis function neural network (being called for short RBF), RBF with Backstepping method is tied mutually
Close, and use the RBF-Backstepping method after combination to design BTT missile attitude control device.
1.1, BTT guided missile is write with BTT guided missile state equation when 40,000 feet of high cruises of the speed of Mach number 2.75
For:
Wherein x1=[α β φ]T∈R3, x2=[p q r]T∈R3, x3=[θ φ]T∈R2,It it is the single order of state vector x
Partial derivative, controls input u=[δp δq δr]T∈R3, α is the angle of attack, and β is yaw angle, and φ is roll angle, and p is angular velocity in roll, q
For rate of pitch, r is yaw rate, and θ is the angle of pitch；R3It is a three-dimensional real number field, R2It it is a two-dimentional real number
Territory；f1It is the function about α, β, f2It is about α, the function of β, p, q, r, g1It is about α, the function of β, φ, θ, g2It is about α
Function, h1It it is the function about α, β.
BTT guided missile state equation (formula 1) is removed subsidiary equationAnd simplify, obtain as follows
BTT guided missile state equation:
Introduce RBF neural, by the function of state on-line study in BTT guided missile state equation being solved control mould
Type mismatch problems so that it is can adapt to the Parameters variation of model and the uncertainty of structure, thus obtain equation below:
Wherein, Δ1(x1) and Δ2(x1,x2) it is model indeterminate:
H is the hidden layer output matrix of neutral net, and Q is the connection weight value matrix between hidden layer and output layer, Q*It is
Expectation connection weight value matrix between hidden layer and output layer, uses neutral net to carry out approximate model indeterminate Δ1(x1) and
Δ2(x1,x2) to make up modeling error:
Introduce error vector e1And e2, e1Represent the error of state vector and the control command signal fed back, e2Represent
State variable and the error of dummy pilot signal:
Backstepping method construct is used to draw the target following signal x of BTT guided missile state equation internal ring and outer shroud1d
And x2d, x1dBe given by control command signal, x2dBe given for Virtual Controller.
1.2, Backstepping method specifically comprises the following steps that
The first step, for BTT guided missile state equation internal ring, to e1Derivation:
Design Virtual Controller signal x2d:
k1It is to control parameter；
Formula (9) is substituted into formula (8), and order:
Then right value update rate is:
γ1It it is step parameter；
Structure Lyapnov function, and to its derivation be:
Work as e2When → 0,Therefore this BTT guided missile state equation internal ring is calmed；
Second step, for BTT guided missile state equation outer shroud, to e2Derivation:
Design controller:
k2It is to control parameter；
Formula (13) is substituted into (12), and order
Then right value update rate is:
γ2It it is step parameter
Structure Lyapnov function, and to its derivation be:
Wherein k1,k2> 0 time,Therefore this BTT guided missile state equation outer shroud can be calmed；
2, the FPGA of RBF-Backstepping controller (attitude controller) realizes, and is divided into two parts: RBF neural
FPGA realize and Backstepping method FPGA realize.
2.1, the FPGA of RBF neural realizes；
RBF neural has three layers, is n neuron of input layer, h neuron of hidden layer and output layer m god respectively
Through unit, the weights of input layer to hidden layer are 1, and hidden layer uses Gauss (Gaussian) function as excitation function, Gauss
Function expression isOutput layer is the linear combination of hidden layer node, RBF neural
FPGA realizes being made up of following two part:
1. the realization of mononeuron FPGA
This neuron by input register RAM, weight storage device RAM, excitation function parameter RAM, multiplier I, accumulator,
Excitation function (neuron of input layer and output layer does not comprises), buffer and control unit module composition, during operation:
(1) input signal sends input register RAM to；
(2) control unit produces and enables signal, controls multiplier I, accumulator, excitation function and four modules of buffer and depends on
Secondary operation；
(3) input register RAM and weight storage device RAM reads input vector value and weight vector value respectively；
(4) will element feeding multiplier I corresponding with in weight vector for input vector be multiplied；
(5) by accumulator, multiplied result carried out cumulative summation；
(6) summed result is sent excitation function module to, draw the value of its Gaussian function, buffered by buffer defeated
Go out.
RBF neural excitation function owing to using is Gaussian function, so needing an additional excitation function parameter
RAM is for storing the radial direction sound stage width degree b of Gaussian functioniWith radial direction base center ai, this n+h+m neuron constitute RBF neural
Mixed-media network modules mixed-media；
2. the FPGA of excitation function Gaussian function realizes
Exponential function e is realized by using piecewise function to approach the method combined with look-up tablex, first pass through
MATLAB iunction for curve polyfit is to exNegative semiaxis carries out by stages and approaches, and obtains exBy stages approximating function expression formula with
And the maximum approximate error in each interval, it being divided into 13 intervals, concrete piecewise function is as follows:
This exponential function is by inputting RAM, excitation function mux (mux is a MUX module in FPGA), constant value
Approximating function, quadratic closeness function, time delay and output RAM module composition, during operation:
(1) input RAM storage input；
(2) input value is sent into excitation function mux, it judge the interval ownership of input value；
(3) if input value is in (-∞ ,-5) interval, exwponential function approximation is in constant, approximating function (constant value approximating function)
Use 8 corresponding piecewise constant function y=C；
(4) if input value [-5,0) in interval, exponential function change is violent but smoother, and approximating function (force by secondary
Nearly function) use corresponding 5 segmentation quadratic fit function y=Ax2+ Bx+C carrys out approaching to reality curve；
(5) shorter than the quadratic closeness function calculating time due to constant value approximating function, therefore use time delay module that constant value is approached
The value of calculation of function and quadratic closeness function does output synchronization process (makes the input value of different hidden layers by hidden layer god
After the calculating of unit, can be simultaneously communicating in output neuron)；
(6) result is sent in output RAM.
2.2, from the controller derivation in above-mentioned BTT guided missile state equation, Backstepping method realizes
Process is matrix operations, also relates to part nonlinear function computing, such as sin cos functions in this external BTT guided missile state equation
And square root function, therefore, the FPGA of Backstepping method realizes being made up of following three parts:
1. matrix multiple
If A=is [aij] it is a m × N-shaped matrix, B=[bij] it is n × s type matrix, matrix A is multiplied with B and obtains
Matrix C may be defined as:
Matrix multiple by two input matrixes, three multiplexers, state machine I, line storage, row memorizer, multiplier II,
Adder I and output matrix module I composition.During operation:
(1). stored two according to the matrix multiple module that formula (18) designs respectively by input matrix A and input matrix B
Matrix A and B；
(2). under state machine I controls, matrix A is read by multiplexer I by row, deposits into line storage, and multiplexer II is by square
Battle array B is read by row, deposits into row memorizer；
(3). in two memorizeies, corresponding element carries out multiplying accumulating computing (in multiplier II and adder I)；
(4). result of calculation is stored in relevant position in output matrix module I via multiplexer III, i.e. can get matrix
C。
2. matrix inversion
Using the adjoint matrix tactical deployment of troops to realize matrix inversion, for n rank square formation A, its inverse matrix is:
In formula: A*For the adjoint matrix of matrix A, the determinant that | A | is matrix A.
The matrix inversion module designed according to above thinking, this module is divided into two big submodules, i.e. solves determinant module
M1 and solve adjoint matrix module M2, M1 and M2 parallel running, matrix inversion module by input matrix, two multiplexers, two
State machine, four memorizeies, four multipliers, adder, subtractor, divider, matrix functions are taken advantage of and are formed with output matrix module,
During operation:
(1) for solving determinant module M1, multiplexer IV is sampled out in determinant product term from input matrix module I
Rearrange by row after each element, and increase by a column element and represent sign bit, such as:
WillIt is rearranged forIt is stored in memorizer I；
(2) under state machine II controls, read every row element in memorizer I in order and send into multiplier (multiplier III He
Multiplier IV) and adder II in carry out multiplying accumulating operation and i.e. can get determinant | A | of matrix A；
(3) matrix A determinant is stored in memorizer III；
(4) for solving adjoint matrix module M2, multiplexer V is sampled from input matrix module I, by calculating adjoint matrix
The mode of the required each element of battle array arranges, such as:
WillIt is rearranged forIt is stored in memorizer II；
(5) under state machine III controls, read every row element in memorizer II in order and send into multiplier (multiplier V He
Multiplier VI) and subtractor I in carry out the adjoint matrix A that takes advantage of subtraction i.e. to can get matrix A*, it is stored in memorizer IV；
(6) signal is sent into divider I and is carried out division arithmetic (ask reciprocal) by memorizer III；
(7) matrix function is taken advantage of module that the signal of division arithmetic and the signal of memorizer IV are carried out matrix function to take advantage of；
(8) matrix function is taken advantage of result be stored in output matrix module II；
In view of divider time delay compared with big, to consume resource more, enter with the inverse of each element of adjoint matrix with determinant
Row matrix scale multiplication obtains inverse matrix.
3. the FPGA of sin cos functions realizes
Using CORDIC method simply to add and subtract shifting function by iterating to be converted into by sin cos functions, this is just
The reaction type CORDIC method structure of cosine function is by three multiplexers, three depositors, two shift units and three adder substracters
Module forms, and during operation, uses Parallel Feedback formula structure to realize CORDIC method, altogether iteration n times, and the i-th step operation is as follows:
(1) first by XiSend into multiplexer VI, YiSend into multiplexer VII, ZiSending in multiplexer VIII, theta_i is stored in and posts
In storage III；
(2) result of multiplexer VI being sent into depositor I, the result of multiplexer VII sends into depositor II, multiplexer VIII
Result sends into depositor III；
(3) value of depositor II carries out the shifting function (shift unit II) value afterwards with depositor I carry out adding reducing and (add
Subtract device II), obtain Xi+1；
(4) value that the value of depositor I carries out shifting function (shift unit I) and depositor II carries out adding reducing (plus-minus
Device I), obtain Yi+1；
(5) carry out adding reducing (adder substracter III) by the value of depositor III and the value of multiplexer VIII, obtain Zi+1；
(6) by Xi+1Send into multiplexer VI, Yi+1Send into multiplexer VII, Zi+1Send in multiplexer VIII；
(7) the sine and cosine value of angle theta is obtained after iteration n times.
2.3, the FPGA combining RBF neural realizes and the FPGA realization of Backstepping method, finally sets up out
The RBF-Backstepping controller FPGA model of BTT guided missile, design cycle is as follows, sees Fig. 4, Fig. 5:
2.3.1, input state variable module stores input state variate-value x1=[α β φ]TAnd x2=[p q r]T, and
It is sent in matrix operations I module, neutral net I module and matrix operations II module.
2.3.2, matrix operations I module calculates matrix f1And g1Value, be sent to Virtual Controller module, matrix
Computing II module calculates matrix f2And g2Value, be sent to controller input module, wherein:
KQExpression is a constant relying on flying condition；
Wherein, a1、a2、a3Represent aerodynamic model constant respectively；
Q,S,d,Ixx,Iyy,Izz,Ixz,It is known ginseng
Number, the trigonometric function related in controller has angle of attack, sideslip angle beta, the sine and cosine value of roll angle φ and sideslip angle beta and bows
The tangent value of elevation angle theta, uses in above-mentioned solution procedure and has arrived CORDIC method module and divider module to realize triangle letter
Digital-to-analogue block tri, it should be noted that in CORDIC method iterative process, all parameter values are respectively less than 1, therefore CORDIC module uses
16 fixed-point number forms of 1-1-14, to improve precision, realize the computing of same good word long data, pitching by shifting function
Angle tangent value is divided by by this angle sine and cosine and obtains, and yaw angle is because angle is less close to 0, therefore its tangent value can be approximately it
Sine value.
2.3.3, neutral net I module calculates the output of internal ring neutral netRealize process as follows: first error
Module Error_gen1 calculates inner loop error e1, error module Error_gen2 calculates outer shroud error e2, secondly right value update
Module update1 calculates right value update rateFrom controller derivation, neural network weight turnover rate such as formula
(32) shown in, after one be to accelerate response speed and additional Inertia:
In formula: γ1For Learning Step, e1For inner loop error, H1For neutral net hidden layer output valve；N is inertial factor；K
For inner loop error vector, K=[e1,e2] use error norm module Norm_k to realize, this module is used for calculating error vector K
=[e1,e2] norm, as shown in formula (33), the result calculated uses in right value update module as parameter:
Be can be seen that error norm module is by multiplier, adder and square root module composition by formula (33)；Q1For more
Weights before new, formula (32) is a typical continuous print neural network weight turnover rate, in order to realize on FPGA, enters it
Row sliding-model control, i.e. uses the sampling instant kT of series of discrete to replace t continuous time:
In above formula, T is the sampling period, and k is sampling precision, updates according to the RBF network weight that formula (34) designs
Module update1 hardware configuration by inputting RAM, multiplier, matrix multiple, matrix function are taken advantage of, subtractor and output RAM module group
Becoming, its matrix operations related to have employed model mentioned above, finally calculates neutral net I module defeated in each moment
Go outIt is sent to Virtual Controller module.
2.3.4, Virtual Controller module calculates virtual controlling amount x2d, it is sent to neutral net II module, x2dAs
Shown in formula (35):
Its computing related to has matrix function to take advantage of, matrix is added, matrix multiple and matrix inversion, all uses the above
Matrix operations model realizes, virtual machine controller module by inputting RAM, matrix inversion, matrix function are taken advantage of, two matrixes are added,
Matrix multiple and output RAM module composition.
2.3.5, neutral net II module calculates the output of outer shroud neutral netRealize process as follows: first
Error module Error_gen1 calculates inner loop error e1, error module Error_gen2 calculates outer shroud error e2, secondly weights
More new module update2 calculates right value update rateAs shown in formula (36), after one be in order to accelerate response speed and
Additional Inertia:
In formula: γ2For Learning Step, e2For outer shroud error, H2For neutral net hidden layer output valve；N is inertial factor；K
For outer shroud error vector, K=[e1,e2] use error norm module Norm_k to realize, this module is used for calculating error vector K
=[e1,e2] norm, it is achieved method is ibid；Q2For weights before updating, same formula (36) is that a typical continuous print is neural
Network weight turnover rate, in order to realize on FPGA, carry out sliding-model control to it, i.e. uses the sampling instant of series of discrete
KT replaces t continuous time,
In above formula, T is the sampling period, and k is sampling precision, and the RBF network weight designed according to formula formula (37) is more
New module update2 hardware configuration by inputting RAM, multiplier, matrix multiple, matrix function are taken advantage of, subtractor and output RAM module group
Becoming, its matrix operations related to have employed model mentioned above, finally calculates the neutral net II module in each moment
OutputIt is sent to controller input module.
2.3.6, controller module U (i.e. controller input module) is used for calculating angle of rudder reflection output, as shown in formula (38):
Wherein g2And f2For controller nonlinear parameter, k2For the controller parameter of design, e1,e2It is respectively inner and outer ring by mistake
Difference,Export for outer shroud neutral net, g1For Virtual Controller nonlinear parameter, its computing is also that matrix function is taken advantage of, matrix is added
And matrix multiple, controller module U by inputting RAM, matrix inversion, matrix function are taken advantage of, two matrixes are added, two matrix multiples and
Output RAM module composition, is finally sent to output module by u.
2.3.7, output module for the angle of rudder reflection output valve that receives of storage, thus have devised the RBF-of BTT guided missile
The FPGA model of Backstepping controller.
3, the guided missile digital autopilot of FPGA single-chip framework is set up, first with expectation attitude angle information and feedback
The current state value of circuit feedback is input quantity, after input quantity enters into attitude controller, attitude controller and the control calculated
Guided missile control surface deflection is ordered about in instruction processed, thus changes body attitude, and sends into angle of rudder reflection in the Models computed module of FPGA
Row Models computed, resolves the state vector obtained and attitude information feeds back to input and enters in attitude controller, thus structure
Closed loop is become to be circulated, when, after loop ends, outfan exports attitude angle, designs FPGA single-chip according to above thinking
The guided missile digital autopilot of framework.
Seeing Fig. 1, Figure 19, the flow process that this automatic pilot runs is:
(1) first FPGA receives the desired signal of external world's input by serial ports and is passed to memory RAM；
(2) under state machine orders about, attitude controller reads expects attitude angle and from buffer storage II in RAM
Original state signal carries out computing；
(3) during operation result sends into buffer storage I；
(4) Models computed is carried out by data in Models computed module read buffer memory I；
(5) resolve the state vector obtained and send into buffer storage II, complete closed loop feedback；
(6) data every circulation primary in closed loop, each state signal value all can be stored in FIFO by automatic pilot
Memorizer FIFO (First In First Out is called for short FIFO)；
(7), after end of run, use serial ports sending module data in FIFO to be sequential read out in order, i.e. can get each shape
The dynamic response process of state variable.
Emulation example
When BTT guided missile with the speed of Mach number 2.75 40,000 feet of high cruises time, by aerodynamic parameter and the physics of guided missile
Parameter substitutes into guided missile model and obtains following differential equation group:
Simulation input signal is the BTT guided missile angle of attack, yaw angle and the desired signal of roll angle.Input signal tool in simulations
Body is configured to:
Design digital autopilot based on FPGA architecture as it is shown in figure 1, its structural rtl as shown in Figure 2.Exp_
Data module receives the desired signal of external world's input, and Sys_fsm module is state machine, controls the operation of program, Back_ode mould
Block is closed loop main body, and it can be further divided into some submodules, as it is shown on figure 3, Back_ctrl is missile attitude control device,
ODE be guided missile model resolve module, W_ram4 and W_ram3 is that dynamic memory module uses as buffer, Addr_gen4 with
Addr_gen3 is respectively the address module of W_ram4 and W_ram3, and UART_FIFO is used for storing each shape during automatic pilot runs
State vector dynamic value, to obtain each attitude angle response curve, Bo_fsm is state machine, controls the operation of program, automatic pilot
Carrying out practically process as follows:
First, serial received module receives the desired signal of external world's input and is passed to RAM (RAM referring in Fig. 1).
2. secondly, under state machine orders about, controller (referring to Back_ctrl) read RAM expects attitude angle and from
The original state signal of buffer storage II carries out computing.
As shown in Figure 4, its structural rtl is as shown in Figure 5 for controller (referring to Back_ctrl) hardware configuration.In Figure 5,
Error_gen1 module is used for calculating quantity of state x1And the deviation between expected value, i.e. inner loop error；Error_gen2 module is used
Calculate quantity of state x2And the deviation between intermediate virtual controller, i.e. outer shroud error, Error_gen1 and Error_gen1
Module hardware structure is as shown in Figure 6；Norm_k module is used for calculating inner and outer ring error norm, updates for neural network weight,
Its hardware configuration is as shown in Figure 7；RBF1 is the forward direction operation link of first RBF neural；RBF2 is second RBF nerve
The forward direction operation link of network, RBF neural hardware configuration is as shown in Figure 8；W_ram1 and W_ram2 is memory module, point
Not Cun Chu the value information of two neutral nets, W_ram modular structure is as shown in Figure 9；Addr_gen1 Yu Addr_gen2 is W_
The address module of ram1 and W_ram2 module；Updata1 and Updata2 is right value update module, is respectively intended to update two god
Through the weights of network, Updata module hardware structure is as shown in Figure 10；X2d module is method intermediate virtual controller, its module
Hardware configuration is as shown in figure 11；Some non-linear matrix modules that F1, G1 and B use needed for being to solve for Virtual Controller；Tri is
Trigonometric function module, is also used for solving virtual controller, and its module hardware structure is as shown in figure 12；U is controller input module,
Exporting three angle of rudder reflection values, its module hardware structure is as shown in figure 13；F2, G2, C and GE module is and solves controller U time institute
Need some matrix modules；Control unit in Fig. 4 is automatic pilot control module, finite state machine realize, and its module is tied
Structure is as shown in figure 14.
3. its operation result is sent into buffer storage 1 (referring to W_ram4 in Fig. 3) by controller (referring to Back_ctrl).
4. in Models computed module (referring to ODE) read buffer memory 1, data carry out Models computed, at mathematical model FPGA
In solver (referring to ODE), selection four step Runge-Kutta computing formula:
In formula, h is material calculation, k1,k2,k3,k4It is respectively different time points slope value in step-length, ynFor present moment
Result of calculation, yn+1For the result of calculation in next step-length moment, based on formula (39) and formula (40) design as shown in figure 15 hard
Part structure resolves for implementation model；This mathematical model FPGA calculates 5 big modules K1 of device, K2, K3, K4 and Y, respectively representative formula
(40) 5 formula；K1, K2, K3, K4 can be divided into again several submodules, and wherein, g0, g1, g2, g3, g4, g5 module is respectively
Corresponding formula (39) shown six differential equations；TRI is trigonometric function module, is used in the formula that solves (39) relevant trigonometric function；
SA1, SA2, SA3 are shifter-adder module, for formula argument value each in calculating formula (40)；When program is run, K1, K2, K3,
K4 module is run successively under the control of state machine ode_fsm module and operation result is delivered to Y module, and Y module is according to above-mentioned
The state vector of result and current time calculates the state vector value in next step-length moment.
5. the state vector that Models computed obtains sends into buffer storage 2 (referring to W_ram3 in Fig. 3), completes closed loop feedback.
6. last, serial ports sending module sequential reads out the data in UART_FIFO in order, thus obtains exporting attitude
Angle.
Set up the emulation initial condition to be: BTT guided missile with the speed of Mach number 2.75 in 40,000 feet of high cruises, emulation step
The long fixed step size being set to 0.0001s, initial state value is as shown in table 1.
Table 1 guided missile initial state value
In simulation result, angle of attack aircraft pursuit course as shown in figure 16, as shown in figure 17, follow the tracks of yaw angle aircraft pursuit course by roll angle
Curve is as shown in figure 18.
FPGA simulation result is contrasted with Simulink simulation result, as shown in table 2.
Table 2 mean square error contrasts
FPGA simulation result was not so good as in second and third stage of the angle of attack and the phase III of roll angle as can be seen from Table 2
Simulink emulates, and remaining stage is all better than Simulink, particularly yaw angle.
Simulink emulation, FPGA-PC HWIL simulation, the simulation time of FPGA HWIL simulation are contrasted, knot
Fruit is as shown in table 3:
Table 3 simulation time contrasts
From table 3 it can be seen that HWIL simulation required time based on FPGA is less than 1s, it is far superior to Simulink emulation
As a result, there is the highest real-time；Occupation condition is as shown in table 4.
Table 4 consumed resource
By above each simulation result it can be seen that the automatic pilot of RBF-Backstepping method based on FPGA
Good control effect is achieved for BTT guided missile, and has the highest real-time.
Claims (6)
- The FPGA implementation method of 1.BTT guided missile Neural Network Inversion automatic pilot, it is characterised in that: comprise the following steps:1) state equation model of BTT guided missile is set up；2) RBF neural method is selected to compensate the modeling error of BTT guided missile state equation；Then Backstepping side is utilized Method derives control input, thus designs BTT missile attitude control device based on FPGA model；3) guided missile model is resolved transplant and realize in FPGA, FPGA completes whole closed loop feedback, sets up out BTT guided missile Automatic pilot；When automatic pilot runs, input expectation attitude signal and initial state vector are in attitude controller, by appearance State controller calculates control input, utilizes control input to carry out Models computed, draws the new attitude information of BTT guided missile, will be new Attitude information is sent in attitude controller, is consequently formed circulation；The attitude angle of BTT guided missile is exported when loop ends, circulation Number of times is determined by the sample size expecting attitude angle.
- The most according to claim 1, the FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot, its feature exists In: described attitude controller uses RBF neural to approach the indeterminate Δ of BTT guided missile state equation model1(x1) and Δ2(x1,x2), to make up modeling error:Wherein, x1=[α β φ]T∈R3, x2=[p q r]T∈R3, α is the angle of attack, and β is yaw angle, and φ is roll angle, and p is rolling Tarnsition velocity, q is rate of pitch, and r is yaw rate, and H is the hidden layer output matrix of neutral net, and Q is neutral net Hidden layer and output layer between connection weight value matrix,WithIt is that indeterminate approaches value.
- The most according to claim 2, the FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot, its feature exists In: described list-directed input list is shown as:Wherein, f2It is about α, the function of β, p, q, r, g1It is about α, the function of β, φ, θ, g2It is the function about α, e1It is anti- The state vector being fed back to and the error of control command signal, e2It is the error of state variable and dummy pilot signal,It is empty Intend control signal x2dDerivation result.
- The most according to claim 3, the FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot, its feature exists In: described attitude controller includes input state variable module, the first RBF neural module, the second RBF neural mould Block, the first matrix operations module, the second matrix operations module, Virtual Controller module and controller input module；Input shape State variable module is by state variable value x1=[α β φ]TAnd x2=[p q r]TBe sent to the first matrix operations module, first RBF neural module and the second matrix operations module；First matrix operations module calculates f1And g1Value and be sent to virtual Controller module, the second matrix operations module calculates f2And g2Value and be sent to controller input module, a RBF is neural Mixed-media network modules mixed-media calculatesAnd it is sent to Virtual Controller module；Virtual Controller module calculates virtual controlling amount x2dAnd It is sent to the second RBF neural module；Second RBF neural module calculatesAnd deliver to controller input mould Block；Controller input module calculates angle of rudder reflection and as controlling input.
- The most according to claim 1, the FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot, its feature exists In: described Models computed selects four step Runge-Kutta.
- The most according to claim 1, the FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilot, its feature exists In: the FPGA architecture of described automatic pilot include closed loop main body, described closed loop main body include attitude controller, memorizer, One buffer, the second buffer and Models computed module；Attitude controller read the expectation attitude signal in memorizer and The laggard row operation of original state signal in second buffer, operation result sends into the first buffer, the reading of Models computed module In first buffer, data carry out Models computed, resolve the state vector obtained and send into the second buffer, complete closed loop feedback.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610479531.2A CN106200655B (en) | 2016-06-27 | 2016-06-27 | The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilots |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610479531.2A CN106200655B (en) | 2016-06-27 | 2016-06-27 | The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilots |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106200655A true CN106200655A (en) | 2016-12-07 |
| CN106200655B CN106200655B (en) | 2017-06-27 |
Family
ID=57462063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610479531.2A Active CN106200655B (en) | 2016-06-27 | 2016-06-27 | The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilots |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106200655B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2563011A (en) * | 2017-05-25 | 2018-12-05 | Mbda Uk Ltd | Mission planning for weapons systems |
| CN109376853A (en) * | 2018-10-26 | 2019-02-22 | 电子科技大学 | Echo State Networks export aixs cylinder circuit |
| CN109992198A (en) * | 2017-12-29 | 2019-07-09 | 深圳云天励飞技术有限公司 | The data transmission method and Related product of neural network |
| US11029130B2 (en) | 2017-05-25 | 2021-06-08 | Mbda Uk Limited | Mission planning for weapons systems |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3395874A (en) * | 1965-07-31 | 1968-08-06 | Contraves Italiana Spa | Accelerometric automatic pilot for missiles |
| CN103353759A (en) * | 2013-06-25 | 2013-10-16 | 西安交通大学 | CDM (Coefficient Diagram Method)-based missile autopilot design method |
| CN104197793A (en) * | 2014-08-25 | 2014-12-10 | 中国人民解放军海军航空工程学院 | Missile PID controller parameter self-adaptive adjustment method |
-
2016
- 2016-06-27 CN CN201610479531.2A patent/CN106200655B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3395874A (en) * | 1965-07-31 | 1968-08-06 | Contraves Italiana Spa | Accelerometric automatic pilot for missiles |
| CN103353759A (en) * | 2013-06-25 | 2013-10-16 | 西安交通大学 | CDM (Coefficient Diagram Method)-based missile autopilot design method |
| CN104197793A (en) * | 2014-08-25 | 2014-12-10 | 中国人民解放军海军航空工程学院 | Missile PID controller parameter self-adaptive adjustment method |
Non-Patent Citations (4)
| Title |
|---|
| 施敏良 等: "BTT导弹自适应反演控制律设计与仿真", 《电光与控制》 * |
| 李前国 等: "基于干扰观测器的导弹自动驾驶仪的反演设计", 《电光与控制》 * |
| 董朝阳 等: "基于积分滑模的BTT导弹鲁棒反演控制律设计", 《航空兵器》 * |
| 郭强 等: "自适应反演技术在高超声速巡航导弹自动驾驶仪设计中的应用研究", 《飞行力学与飞行试验（2004）学术交流年会论文》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2563011A (en) * | 2017-05-25 | 2018-12-05 | Mbda Uk Ltd | Mission planning for weapons systems |
| US11029130B2 (en) | 2017-05-25 | 2021-06-08 | Mbda Uk Limited | Mission planning for weapons systems |
| GB2563011B (en) * | 2017-05-25 | 2022-04-27 | Mbda Uk Ltd | Mission planning for weapons systems |
| CN109992198A (en) * | 2017-12-29 | 2019-07-09 | 深圳云天励飞技术有限公司 | The data transmission method and Related product of neural network |
| CN109992198B (en) * | 2017-12-29 | 2020-07-24 | 深圳云天励飞技术有限公司 | Data transmission method of neural network and related product |
| CN109376853A (en) * | 2018-10-26 | 2019-02-22 | 电子科技大学 | Echo State Networks export aixs cylinder circuit |
| CN109376853B (en) * | 2018-10-26 | 2021-09-24 | 电子科技大学 | Echo state neural network output axon circuit |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106200655B (en) | 2017-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107491616B (en) | Structure finite element parametric modeling method suitable for grid configuration control surface | |
| CN106200655B (en) | The FPGA implementation method of BTT guided missile Neural Network Inversion automatic pilots | |
| Bauer et al. | Supercritical Wing Sections II: A Handbook | |
| CN103080941A (en) | Apparatus for generating computational data, method for generating computational data, and program for generating computational data | |
| CN110018637B (en) | Spacecraft attitude tracking performance-guaranteeing control method considering completion time constraint | |
| CN105955031A (en) | Non-linear-model-predictive-control FPGA hardware acceleration controller and acceleration realization method | |
| CN111813146A (en) | Reentry prediction-correction guidance method based on BP neural network prediction voyage | |
| CN114444214A (en) | Aircraft control method based on control surface efficiency | |
| CN109446581A (en) | The measurement method and system of the Hydrodynamic of floating body under a kind of wave action | |
| CN110134018A (en) | A kind of underwater multi-foot robot system polypody cooperative control method | |
| Gao et al. | Broad learning system-based adaptive optimal control design for dynamic positioning of marine vessels | |
| Zang et al. | Standoff tracking control of underwater glider to moving target | |
| CN115618498A (en) | Prediction method, device, equipment and medium for cross-basin flow field of aircraft | |
| Simon | Fighter Aircraft Maneuver Limiting Using MPC: Theory and Application | |
| Dębski | An adaptive multi-spline refinement algorithm in simulation based sailboat trajectory optimization using onboard multi-core computer systems | |
| CN114861304A (en) | Nonlinear aerodynamic force data rapid modeling method and system and storage medium | |
| CN115407294B (en) | Aircraft dynamic radar scattering cross section simulation method for double-station radar system | |
| Oruc et al. | Towards real-time fully coupled flight dynamics and cfd simulations of the helicopter/ship dynamic interface | |
| CN114897146B (en) | Model generation method and device and electronic equipment | |
| Ripepi | Model order reduction for computational aeroelasticity | |
| CN108197368A (en) | It is a kind of to be suitable for the geometrical constraint of aircraft complexity aerodynamic configuration and weight function Two Simple Methods | |
| Wang et al. | Event-triggered hierarchical learning control of air-breathing hypersonic vehicles with predefined-time convergence | |
| Baayen | Vortexje-an open-source panel method for co-simulation | |
| CN114115256A (en) | Ship course control method based on cloud model | |
| Shu et al. | Parametric Aeroelastic Reduced-Order Modeling with Hyperparameter Optimization for Flutter Analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201231 Address after: 236200 south side of Guanying Road, Shencheng Town Economic Development Zone, Yingshang County, Fuyang City, Anhui Province Patentee after: Anhui bairunyuan Beverage Co.,Ltd. Address before: Beilin District Xianning West Road 710049, Shaanxi city of Xi'an province No. 28 Patentee before: XI'AN JIAOTONG University |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230117 Address after: 236200 No. 266, Zhengzheng South Road, Shencheng Town, Yingshang County, Fuyang City, Anhui Province Patentee after: Yingshang Xiangsheng Construction Management Co.,Ltd. Address before: 236200 south side of Guanying Road, Shencheng Town Economic Development Zone, Yingshang County, Fuyang City, Anhui Province Patentee before: Anhui bairunyuan Beverage Co.,Ltd. |
|
| TR01 | Transfer of patent right |
Effective date of registration: 20240906 Address after: No. 0001 Shuya Road, Yingshang County, Fuyang City, Anhui Province, China 236200 Patentee after: Yingshang County Shenhe Garden Construction Co.,Ltd. Country or region after: China Address before: 236200 No. 266, Zhengzheng South Road, Shencheng Town, Yingshang County, Fuyang City, Anhui Province Patentee before: Yingshang Xiangsheng Construction Management Co.,Ltd. Country or region before: China |