CN112114521B - Intelligent prediction control entry guidance method for spacecraft - Google Patents

Intelligent prediction control entry guidance method for spacecraft Download PDF

Info

Publication number
CN112114521B
CN112114521B CN202010751626.1A CN202010751626A CN112114521B CN 112114521 B CN112114521 B CN 112114521B CN 202010751626 A CN202010751626 A CN 202010751626A CN 112114521 B CN112114521 B CN 112114521B
Authority
CN
China
Prior art keywords
time
guidance
spacecraft
prediction
output
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.)
Active
Application number
CN202010751626.1A
Other languages
Chinese (zh)
Other versions
CN112114521A (en
Inventor
胥彪
李翔
冯建鑫
李爽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202010751626.1A priority Critical patent/CN112114521B/en
Publication of CN112114521A publication Critical patent/CN112114521A/en
Application granted granted Critical
Publication of CN112114521B publication Critical patent/CN112114521B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Economics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Evolutionary Biology (AREA)
  • Automation & Control Theory (AREA)
  • Game Theory and Decision Science (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses an intelligent prediction control entry guidance method for a spacecraft. Firstly, constraint conditions of spacecraft atmosphere entrance are considered, a guidance system is designed by adopting a nonlinear model prediction control method, an attenuation memory filter is introduced, a prediction model correction method based on error information estimation is provided, and robustness of the control system to model errors is enhanced by combining a variable prediction time domain strategy. And then, taking the guidance system as a guidance template, generating a sample data set, performing off-line training of a multilayer neural network, and approaching to the mapping relation between the real-time flight state of the spacecraft and the guidance instruction. And finally, in the guidance process, the trained neural network is used for replacing the process of solving the complex optimization problem and integral prediction, so that the time for solving the guidance instruction on line is reduced. Simulation results show that the guidance precision and the real-time performance of the guidance method provided by the invention can meet task requirements under the condition of larger errors such as model uncertainty.

Description

Intelligent prediction control entry guidance method for spacecraft
Technical Field
The invention relates to the technical field of spacecraft guidance, in particular to a spacecraft intelligent prediction control entry guidance method.
Background
The guidance of atmospheric admission is the precondition for the spacecraft to land the planet successfully and carry out various scientific researches and experiments. As the planet in the solar system, which is most similar to the earth, mars have become important target celestial bodies for deep space exploration in humans. The mars atmosphere is very thin, has great uncertainty, often appears weather such as wild wind, sand and dust. In the process of entering, descending and landing of the spacecraft, the atmospheric entering section lasts for the longest time, the working condition is the worst, the state of the spacecraft changes quickly, and the requirement on the deceleration performance is also high. As the Mars 'curious number' Mars vehicle adopts the closed-loop guidance method to successfully realize the Mars surface soft landing in the Mars atmosphere entrance section, the guidance method is adopted to realize the high-precision safe landing of the Mars landing task, and the Mars landing task in the future is inevitably selected.
At present, spacecraft atmosphere entry guidance methods are generally divided into reference trajectory guidance and prediction correction guidance. The reference track guidance method is to design a reference profile (such as a resistance acceleration profile) in an off-line manner according to requirements in advance, store the reference profile in a spacecraft computer, and design a guidance law according to real-time tracking errors to track the track on line. For example, the PID control law has been successfully applied to reentry guidance of Apollo spacecraft and space shuttle, but the gain is relatively complicated to calculate and is obtained under certain assumed conditions. The guidance law is designed by using the idea of feedback linearization, and the nonlinearity of the resistance dynamics can be counteracted by using a state feedback term. However, when there is a large model error and control saturation, the performance of trajectory tracking is not ideal. On the basis, some scholars propose that the control precision is improved by estimating the model error by using the sliding mode state observer, but due to the characteristics of the sliding mode, the obtained controlled variable has jitter, and great challenges are brought to a roll angle executing mechanism. Higher tracking accuracy can be obtained by using the active disturbance rejection control method to perform resistance tracking. However, none of the above methods takes into account the control quantity constraint. The prediction correction guidance method does not depend on a standard track, continuously predicts the terminal state in the flight process, then corrects the control quantity according to the deviation from the expected terminal state, has higher accuracy of the falling point and is insensitive to the entering initial condition. However, the guidance precision of the method depends on the precision of the online model to a great extent, the method has high calculation complexity and needs to rely on strong computer performance to improve the calculation speed, and the real-time performance of online guidance is difficult to ensure.
The model prediction control method capable of processing constraint on line is applied to the problem of guidance of atmosphere entry in recent years, and can well solve the problem of control saturation existing in the process of tracking a reference track. For a complex nonlinear model of an atmospheric inlet section, most scholars design a guidance system by utilizing a linear prediction control method after linearizing the model, and the landing precision of the guidance system is influenced to a certain extent by the strategy. For the prediction control of the nonlinear model, a nonlinear prediction guidance law of a Mars atmosphere entry section based on resistance tracking is provided in the 2 nd, 137 nd and 143 nd page & ltnonlinear prediction guidance law design of Mars atmosphere entry section based on resistance tracking & gt volume 2 in 2015 of journal, the nonlinear prediction control algorithm with the analytic solution can realize safe landing under the condition of meeting the control constraint, but the Taylor series expansion is adopted to approximately predict future output, the prediction precision is not high under the condition of large model uncertainty, and the guidance precision can be reduced. The Nonlinear Model Predictive Control (NMPC) method for solving the corresponding Nonlinear programming problem on line by using a numerical optimization algorithm can avoid the influence of Model mismatch caused by adopting a linear Model, has higher precision, but greatly increases the complexity of calculation and can cause instruction delay of a Control system.
The artificial neural network is an algorithmic mathematical model simulating animal neural network behavior characteristics and performing distributed parallel information processing. The network achieves the purpose of processing information by adjusting the mutual connection relationship among a large number of internal nodes depending on the complexity of the system, and has self-learning and self-adapting capabilities. And the artificial neural network designed aiming at specific complex problems can exert the high-speed computing capability of the computer and has the capability of searching the optimal solution at high speed. In recent years, the design idea of applying the advantages of the neural network to the guidance method has also become a research hotspot.
Disclosure of Invention
The invention aims to provide a spacecraft intelligent prediction control atmosphere entering guidance method, which solves the problem that high-precision guidance and rapid command calculation in the current guidance technology are difficult to meet simultaneously. A spacecraft atmosphere entry guidance system is designed based on a nonlinear model prediction control method capable of processing various constraints on line, a prediction model correction method based on error information estimation is provided in consideration of the influence of larger model uncertainty on the performance of an NMPC system, and the robustness of the NMPC system on model errors is enhanced by combining a variable prediction time domain strategy. Aiming at the defects of high complexity and long time consumption of an NMPC method for online optimization and solving of a guidance instruction, a Back Propagation (BP) neural network is trained offline by using sample data generated by an NMPC system. The neural network is used for replacing the process of solving the complex optimization problem and integral prediction, the control quantity meeting various constraints is quickly solved on line, the speed of instruction resolving is increased, the real-time performance of the guidance system is improved, and the influence of guidance instruction delay on guidance precision is reduced.
The invention adopts the following technical scheme for solving the technical problems:
an intelligent prediction control entry guidance method for a spacecraft specifically comprises the following steps:
the method comprises the following steps: according to the state quantity of the initial moment of the atmospheric entering section of the spacecraft and the target constraint condition of the terminal of the entering section, a reference track is generated in an optimized mode, a corresponding rolling angle change curve is obtained, the change relation of the resistance acceleration and the energy is extracted, and the relation of the change of the derivative of the resistance acceleration along with the time is obtained through deduction;
step two: establishing a spacecraft entry guidance model, and designing an NMPC guidance system by adopting a nonlinear model predictive control NMPC method;
step three: considering errors existing in the process of entering the atmosphere, setting the distribution range and the form of random errors of each state quantity, performing offline tracking on a reference track for multiple times based on the NMPC guidance system designed in the step two, and storing the optimal control quantity at each moment and each corresponding state quantity obtained by performing rolling solution by using a hybrid optimization algorithm when performing tracking each time to generate a sample data set taking the NMPC guidance system as a guidance template;
step four: taking the state quantities in the three generated sample data sets as input and the corresponding optimal control quantity as output, carrying out back propagation BP neural network training, and finishing the training of the BP neural network when the value of the loss function converges to a set error range or reaches the maximum iteration number of the training to obtain the neural network based on the NMPC guidance system;
step five: and (3) considering errors existing in the actual guidance process, adopting a neural network based on an NMPC guidance system in the fourth step to track the reference track generated in the first step on line: and inputting the state quantity obtained by real-time measurement of the spacecraft navigation system into a neural network based on an NMPC guidance system to obtain the corresponding optimal control quantity, thereby realizing intelligent guidance of the spacecraft.
Further, the step one is specifically as follows: according to the speed, longitude and latitude, height, course angle, track angle of an initial moment of an atmospheric entering section of the spacecraft and target constraint conditions of a terminal of the entering section, under the condition that path constraint including overload, dynamic pressure and heat flow rate and control quantity constraint are met, a reference track is generated in an optimized mode, a corresponding rolling angle change curve is obtained, the change relation of the resistance acceleration and the energy is extracted, and the relation of the change of the derivative of the resistance acceleration along with the time is obtained through deduction.
Further, in the second step, the kinetic equation of the spacecraft entering the guidance model is as follows:
Figure BDA0002610224820000031
Figure BDA0002610224820000032
Figure BDA0002610224820000033
wherein, the spacecraft state vector at the time t
Figure BDA0002610224820000034
r is the distance from the center of mass of the spacecraft to the target, V is the speed of the spacecraft,
Figure BDA0002610224820000035
is latitude, theta is longitude, gamma is track angle, psi is course angle, u (t) is control input vector at time t; y isc(t) is the output value at time t, D (t) and
Figure BDA0002610224820000036
the resistance acceleration and the first derivative thereof at the time t; y isb(t) is the value of the path constraint at time t,
Figure BDA0002610224820000037
is the heat flow rate at time t, q (t) is the incoming flow pressure at time t, nLAnd (t) is the overload of the spacecraft at the time t.
Further, in the second step, an NMPC guidance system is designed by adopting a nonlinear model predictive control NMPC method, which specifically comprises the following steps:
1) at time t, a constrained nonlinear programming problem of the form:
Figure BDA0002610224820000038
the following constraints are satisfied:
Figure BDA0002610224820000039
wherein x (t) is the state of the spacecraft at time t; t iscAnd TpRespectively a control time domain and a prediction time domain, and satisfies Tc≤Tp；σminFor minimum value of allowable roll angle, σmaxIn order to allow for a maximum value of roll angle,
Figure BDA00026102248200000311
to maximize the allowable roll angle rate, δ is the sampling time, Tc＝Ncδ,Tp＝Npδ，NcFor controlling the number of steps, NpPredicting the step number;
Figure BDA00026102248200000310
for the upper limit of the allowable heat flow rate, qmaxFor an allowable upper limit of dynamic pressure, nLmaxIs the allowable overload upper limit; the objective function J at time t is specifically defined as
Figure BDA0002610224820000041
Figure BDA0002610224820000042
And
Figure BDA0002610224820000043
the predicted value of the resistance acceleration and the first derivative thereof at the t moment to the tau moment is represented;
Figure BDA0002610224820000044
Dr(τ) and
Figure BDA0002610224820000045
representing the desired output at time τ, i.e. the resistive acceleration of the reference track and its first derivative, ur(τ) with yr(τ) a corresponding control input, namely roll angle;
Figure BDA0002610224820000046
control input value representing the time T to be solved at time T for any time T e [ T, T + T ∈c]，
Figure BDA0002610224820000047
i-int (τ -t/δ), int (·) being a downward rounding function;
Figure BDA0002610224820000048
to-be-solved optimization sequence U for forming nonlinear programming problem with constraint at time ttI.e. by
Figure BDA0002610224820000049
Figure BDA00026102248200000410
A predicted value representing the path constraint output from the time t to the time tau; q is an output error weighting matrix, and R is a control quantity weighting matrix
Figure BDA00026102248200000411
And
Figure BDA00026102248200000412
calculated by the following prediction model:
Figure BDA00026102248200000413
wherein
Figure BDA00026102248200000414
For the predicted value of the state quantity at time t to time t,
Figure BDA00026102248200000415
the values of r, V, θ,
Figure BDA00026102248200000416
predicted values of γ, ψ;
2) correcting the prediction model in 1) by adopting a prediction model correction method based on error information estimation:
first, a first-order attenuation memory filter is designed to obtain error estimation information, wherein the first-order attenuation memory filter is Z (t) -Z (t-delta) + (1-epsilon) (Z)*(t) -Z (t- δ)), the state quantity at time t of the first-order attenuation memory filter is
Figure BDA00026102248200000417
Z (t-delta) represents the state quantity of the filter at the t-delta moment, Z (0) is 1, 0 & ltepsilon & lt 1 is a gain coefficient, L (t) represents the actual lift acceleration at the t moment,
Figure BDA00026102248200000418
and
Figure BDA00026102248200000419
respectively representing the predicted values of the prediction model at the time t to the resistance acceleration and the lift acceleration;
secondly, the output Z (t) of the first-order attenuation memory filter is used as a correction factor of the prediction model, and
Figure BDA00026102248200000420
and
Figure BDA00026102248200000421
make real-time corrections, i.e.
Figure BDA00026102248200000422
And
Figure BDA00026102248200000423
respectively predicting values of the corrected resistance acceleration and the corrected lift acceleration;
finally, the prediction model is modified as follows:
Figure BDA00026102248200000424
wherein
Figure BDA00026102248200000425
Correcting the predicted value of the prediction model to the state quantity at the time tau for the time t,
Figure BDA00026102248200000426
represents the predicted value output by the tau time correction prediction model to the tau time,
Figure BDA0002610224820000051
the predicted value of the path constraint output at the time t of the correction prediction model is represented;
3) and (3) adding a correction link for prediction errors at each optimization moment of the constrained nonlinear programming problem:
first, the error between the predicted output and the actual output at time t is defined:
Figure BDA0002610224820000052
wherein
Figure BDA0002610224820000053
The predicted output value of the corrected prediction model at the time t, and y (t) is the actual output value at the time t;
secondly, correcting the predicted value output by the modified prediction model at the time t to the time tau in a manner of weighting e (t):
Figure BDA0002610224820000054
wherein the content of the first and second substances,
Figure BDA0002610224820000055
for the purpose of the corrected predicted output value,
Figure BDA0002610224820000056
and
Figure BDA0002610224820000057
predicted values of the corrected resistance acceleration and its first derivative, eta ═ eta1,η2]Is a correction vector, and η1＞0,η2＞0；
4) After the correction of the prediction model and the feedback correction of the prediction error, the objective function J at the time t is redefined as:
Figure BDA0002610224820000058
further, in the third step, a mixed optimization algorithm combining the sequence quadratic programming SQP and the particle swarm optimization PSO algorithm is adopted to solve the nonlinear programming problem with the constraint: firstly, solving a nonlinear programming problem with constraint by using a PSO algorithm, and then searching the solving result of the PSO algorithm as an optimized initial value near the position by using an SQP algorithm so as to converge to a global optimal solution.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
the invention designs a spacecraft atmosphere entering guidance system based on a nonlinear model prediction control method capable of processing constraint on line, takes the influence of larger model uncertainty on the system performance into consideration, improves an NMPC (non-uniform matrix computer) method, and provides a prediction model correction method and a variable prediction time domain strategy based on error information estimation. The method has the advantages that the traditional model prediction control method can process the control quantity and state quantity constraints on line, the robustness of a control system to model errors is enhanced, and the precision of track tracking is improved. By utilizing the strong self-learning capability of the neural network, the mapping relation between the real-time flight state and the guidance instruction in the NMPC guidance system is approximated, and the optimal control quantity can be rapidly calculated on line according to the current state quantity. Compared with an NMPC method, the neural network is used for replacing the process of solving a complex NLP problem and integral solving prediction output by a numerical optimization method in each guidance period, the defects of high solving complexity and long time consumption of an NMPC guidance system are overcome to a great extent, the solving time of guidance instructions is shortened, and the influence of instruction delay on guidance precision is reduced. The real-time requirement of a control system can be well met while high-precision guidance is ensured, intelligent guidance of an atmospheric entry section of the spacecraft is realized, and the method has important engineering application value.
Drawings
FIG. 1 is a roll angle profile of a reference trajectory according to the present invention.
Fig. 2 is a resistance acceleration curve of a reference trajectory according to the present invention.
FIG. 3 is a flow chart of a nonlinear model predictive control method in accordance with the present invention.
Fig. 4 is a graph of the NMPC method height variation according to the present invention.
Fig. 5 is a speed profile of the NMPC method according to the present invention.
Fig. 6 is a roll angle profile for the NMPC method of the present invention.
Fig. 7 is a resistance acceleration curve of the NMPC method according to the present invention.
Fig. 8 is a change curve of longitude and latitude of the NMPC method according to the present invention.
FIG. 9 is a flow chart of the spacecraft intelligent predictive entry guidance of the present invention.
FIG. 10 is a simulation result of the intelligent predictive control entry guidance method of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the attached drawings:
the invention provides an intelligent prediction control guidance entering method for a spacecraft, which comprises the following steps:
the method comprises the following steps: optimizing and generating a reference track according to the initial flight state and the constraint conditions of the atmospheric entering section of the spacecraft; according to the state quantity (speed, longitude and latitude, height, course angle and track angle) of the spacecraft at the initial atmospheric entering time and the target entering terminal constraint conditions, under the condition that path constraint (overload, dynamic pressure and heat flow rate) and control quantity constraint are met, a reference track is generated in an optimized mode, a corresponding rolling angle change curve is obtained, and the change relation between the resistance acceleration and the energy is extracted. The derivative of the resistive acceleration is then derived as a function of time.
In the embodiment, Mars detection is used as a background, and a spacecraft atmosphere entry section three-degree-of-freedom kinetic equation can be expressed in a nonlinear form as follows;
Figure BDA0002610224820000061
wherein r is the distance from the center of mass of the spacecraft to the fire center, namely the sum of the height h of the spacecraft and the average radius Re of the Mars, V is the speed of the spacecraft,
Figure BDA0002610224820000062
is latitude and theta is longitude. Gamma is the track angle, i.e. the angle between the speed and the local horizontal plane. ψ is the heading angle, i.e. the angle between the projection of the velocity on the horizontal plane and the east direction. σ is the roll angle, i.e. the rotation angle of the spacecraft with respect to the velocity vector. g is the acceleration of gravity.
Figure BDA0002610224820000063
Wherein m isvThe mass of the spacecraft is shown, S is the aerodynamic reference area of the spacecraft, rho is the atmospheric density, D is the resistance acceleration, L is the lift acceleration, C is the lift accelerationDIs the coefficient of aerodynamic drag, CLIs the aerodynamic lift coefficient.
The reference track guidance method firstly reasonably plans a reference track in a given initial state, and meets path constraint, terminal target constraint and control quantity constraint.
Entering an initial state:
Figure BDA0002610224820000071
wherein, r (t)0),θ(t0),
Figure BDA0002610224820000072
V(t0),ψ(t0),γ(t0) The state quantity of the spacecraft at the initial moment of the entering section is obtained.
And (3) path constraint:
(1) the heat flow rate constraint is:
Figure BDA0002610224820000073
wherein the content of the first and second substances,
Figure BDA0002610224820000074
as heat flow rate, coefficient of heat flow k of spacecraftq＝1.898e-4，
Figure BDA0002610224820000075
The upper limit of the allowable heat flow rate.
(2) The dynamic pressure constraint is:
Figure BDA0002610224820000076
wherein q is the incoming flow pressure, q ismaxIs the upper limit of allowable dynamic pressure.
(3) The overload constraint is:
Figure BDA0002610224820000077
wherein n isLFor spacecraft overload, nLmaxIs the upper limit of allowable overload.
And (3) entering a segment terminal target constraint:
(1) entry segment terminal conditions, including altitude and speed limits.
h(tf)≥hf,V(tf)≤Vf (7)；
Wherein, h (t)f),V(tf) Respectively the altitude and the speed, h, of the spacecraft at the end of the entry sectionf,VfAre respectively sailingAltitude and speed limits at the end of the antenna entry segment.
(2) Location (latitude and longitude) constraints.
Figure BDA0002610224820000078
Wherein the content of the first and second substances,
Figure BDA0002610224820000079
respectively the longitude and latitude of the end time of the spacecraft entering segment,
Figure BDA00026102248200000710
respectively, the longitude and latitude of the terminal target point.
And (3) controlling quantity constraint:
Figure BDA00026102248200000711
wherein σminFor minimum value of allowable roll angle, σmaxIs the maximum allowable roll angle. Setting taking into account saturation of actuator response speed
Figure BDA00026102248200000712
Is the maximum allowable roll angle rate.
The specific values entering the initial state and various constraints are as follows:
entering an initial state:
Figure BDA00026102248200000713
and (3) path constraint:
Figure BDA0002610224820000081
qmax＝10kpa/m2,nLmax＝30；
and (3) entering a segment terminal target constraint: h isf＝8km,Vf＝400m/s，θf＝-74.395°,
Figure BDA0002610224820000082
And (3) controlling quantity constraint: sigmamin＝10°,σmax＝80°,
Figure BDA0002610224820000083
Model parameters included in the spacecraft atmosphere entry phase: mars attraction coefficient mu is 4.283 multiplied by 1013m3/s2The average Mars radius Re is 3396200m, and the reference area S of the spacecraft is 15.9m2Coefficient of aerodynamic lift CL0.426, coefficient of aerodynamic drag CD1.723 and spacecraft mass mv2804 kg. The Mars atmospheric density model adopts an approximate exponential model, i.e.
Figure BDA0002610224820000084
Wherein the atmospheric density ρ of the Mars surface0＝0.0158kg/m3Height h of equivalent density of atmospheres＝9354m。
Because the spacecraft is not acted by the external force exerted by people in the process of entering the atmosphere, the energy conservation law is satisfied, the energy of the spacecraft can be expressed by the following formula,
Figure BDA0002610224820000085
therefore, when the terminal energy position is reached, if the spacecraft can reach the terminal target height, the terminal speed of the spacecraft can also reach the requirement. Energy is therefore used as an argument in the reference trajectory planning.
The optimal reference trajectory satisfying various constraints is obtained through Gaussian pseudo-spectrum optimization, and the change curve of the roll angle is shown in figure 1. The variation of the resistance acceleration and the energy is extracted, and the curve is shown in figure 2.
The resistance acceleration is obtained by derivation of time,
Figure BDA0002610224820000086
the first derivative of the resistance acceleration with time is obtained from equation (11).
Step two: (1) and establishing a spacecraft entry guidance model.
The final range error of the spacecraft largely determines the guidance precision, and the range is only related to the magnitude of the resistance acceleration through the kinematic analysis. Therefore, as long as the tracking of the resistance acceleration is realized, the upper reference track can be tracked. The resistance acceleration is used as a tracking target.
Note the book
Figure BDA0002610224820000087
If u (t) is a control input vector and u (t) is a state vector at time t, a kinetic equation entering the guidance model is obtained according to equation (1),
Figure BDA0002610224820000088
wherein y isc(t) is the output value at time t, including the resistive acceleration and its first derivative. y isb(t) is a value of the path constraint at time t, and is expressed by equations (4) to (6).
(2) And designing a guidance system by using a nonlinear model predictive control method.
The model predictive control is essentially a model-based closed-loop optimization control strategy and comprises three parts, namely a predictive model, rolling optimization and feedback control. The method has the main idea that the future state quantity and the output value of a prediction model prediction system are utilized, and based on the thought of finite time domain online repeated optimization, a quadratic programming or nonlinear programming problem with constraints is solved online at each sampling moment, namely, the method is equivalent to solving an open-loop optimal control problem. Each solution results in a control sequence, but only the first control quantity is applied to the system. The above process is repeated until the next sampling moment, and the process is sequentially performed forward in a rolling manner, so that the closed-loop optimal control is realized, and the model predictive control is also called as rolling time domain control. In addition, the model predictive control can process the control quantity and state quantity constraint on line, and can avoid the phenomenon of roll angle control saturation when applied to the guidance entering process, and meanwhile, the constraint on the flight path is realized. A method flow diagram is shown in fig. 3.
At the current time t, there is a measured state value x (t), a constrained nonlinear programming problem of the following form is constructed, as shown in equations (13) and (14),
Figure BDA0002610224820000091
the following constraint conditions are satisfied,
Figure BDA0002610224820000092
in the above problem, x (t) is the state of the spacecraft at the current time, UtAn optimized sequence for the solution required; t iscAnd TpRespectively a control time domain and a prediction time domain, and satisfies Tc≤Tp。
The objective function J at time t in equation (13) is specifically defined as,
Figure BDA0002610224820000093
wherein
Figure BDA0002610224820000094
And
Figure BDA0002610224820000095
the predicted value of the resistance acceleration and the first derivative thereof at the t moment to the tau moment is represented;
Figure BDA0002610224820000096
control input value representing the time T to be solved at time T for any time T e [ T, T + T ∈c]The definition is that,
Figure BDA0002610224820000097
where int (·) is a floor function and δ is the sampling time, satisfies Tc＝Ncδ,Tp＝Npδ，NcFor controlling the number of steps, NpTo predict the number of steps. Therefore, the temperature of the molten metal is controlled,
Figure BDA0002610224820000098
the optimization sequence to be solved constituting the constrained nonlinear programming problem at time t, i.e. U in equation (13)t，
Figure BDA0002610224820000099
Figure BDA00026102248200000910
Dr(τ) and
Figure BDA00026102248200000911
representing the desired output at time τ, i.e. the resistive acceleration of the reference track and its first derivative, ur(τ) with yr(τ) a corresponding control input, namely roll angle; q is an output error weighting matrix, and R is a control quantity weighting matrix. The first term of the objective function has the effect of minimizing the tracking error of the resistance acceleration, and the first derivative of the resistance acceleration is also used as a tracking variable, so that better tracking effect can be achieved. The second term of the objective function can make the optimization algorithm easier to solve the optimal control quantity by introducing the control quantity error.
In the formula (14), σmin10 ° is the minimum allowable roll angle, σ max80 is the maximum allowable roll angle,
Figure BDA0002610224820000101
is the maximum value of the allowable roll angle rate of change;
Figure BDA0002610224820000102
a predicted value representing a constraint of the path,
Figure BDA0002610224820000103
indicating the allowed path constraint maximum.
Figure BDA0002610224820000104
And
Figure BDA0002610224820000105
can be calculated by the following prediction model,
Figure BDA0002610224820000106
wherein x (t) is the initial condition of the prediction model;
Figure BDA0002610224820000107
the predicted value of the state quantity at the time t to the time tau is obtained.
Because large pneumatic coefficient and atmospheric density model errors exist in the actual atmospheric entering process, and the prediction model (18) does not consider the errors, the calculated prediction output value has large errors, and the robustness of the NMPC method can be reduced. Therefore, the invention provides a prediction model correction method based on error information estimation. Error estimation information is first obtained by designing a first order attenuated memory filter. As can be seen from equations (1), (2) and (11), the aerodynamic coefficient and the atmospheric density error affect the output of the system and the path constraint only by changing the magnitudes of the drag acceleration and the lift acceleration, and therefore the ratio of the actual drag acceleration and lift acceleration to the corresponding quantity calculated by the prediction model (18) is used as the state of the filter, that is, the state of the filter is the ratio
Figure BDA0002610224820000108
Where l (t) denotes the actual lift acceleration at time t,
Figure BDA0002610224820000109
and
Figure BDA00026102248200001010
respectively represents the predicted values of the prediction model (18) at the time t on the resistance acceleration and the lift acceleration, Z*(t) is the state quantity at time t of the filter. The first order decay memory filter is shown as equation (20) below,
Z(t)＝Z(t-δ)+(1-ε)(Z*(t)-Z(t-δ)) (20)；
wherein Z (t-delta) represents the state quantity of the filter at the time of t-delta, and the gain coefficient is 0 < epsilon < 1. In order to reduce the influence of model uncertainty, the gain factor may be appropriately set to a larger value to enhance the correction effect of the past state quantity of the filter on the current output value z (t), where ∈ is 0.9. The initial filter value Z (0) is taken to be 1. The output Z (t) of the filter is used as a correction factor of the prediction model (18) to correct the values of the resistance acceleration and the lift acceleration in real time, namely
Figure BDA00026102248200001011
Wherein
Figure BDA00026102248200001012
And
Figure BDA00026102248200001013
the corrected values of the resistance acceleration and the lift acceleration are respectively. The modified prediction model (18) may be represented as a modified prediction model (22),
Figure BDA0002610224820000111
wherein
Figure BDA0002610224820000112
Correcting the predicted value of the prediction model to the state quantity at the time tau for the time t,
Figure BDA0002610224820000113
represents the predicted value output by the modified prediction model at the time t to the time tau,
Figure BDA0002610224820000114
and the predicted value of the path constraint output at the time t by the correction prediction model at the time t is shown.
In order to further improve the capability of the control system to overcome uncertainty, a correction link for prediction errors is added at each optimization moment. Defining the error between the predicted output and the actual output at time t,
Figure BDA0002610224820000115
wherein
Figure BDA0002610224820000116
The predicted output value of the modified prediction model at time t, and y (t) the actual output value at time t. The predicted value output by the modified prediction model at the time t to the time tau is corrected by weighting e (t),
Figure BDA0002610224820000117
wherein the content of the first and second substances,
Figure BDA0002610224820000118
for the purpose of the corrected predicted output value,
Figure BDA0002610224820000119
and
Figure BDA00026102248200001110
predicted values of the corrected resistance acceleration and its first derivative, eta ═ eta1,η2]Is a correction vector, and η1＞0,η2＞0。
After the correction of the prediction model and the feedback correction of the prediction error, the objective function (15) at time t needs to be redefined as,
Figure BDA00026102248200001111
predicting time domain T when designing model predictive control systemspOnce determined, is generally fixed and constant throughout the control process, while the parameter TpThe impact on the control system performance is large. If TpToo large, it will increase the computational burden; if TpToo small, it will make the system less robust.
The analysis of the guidance problem of the atmospheric air entering shows that the density of the atmospheric air entering the initial section is very small, and the speed of the spacecraft at the final section is relatively low. As can be seen from the formula (2), the resistance acceleration in the two stages is small, and uncertain factors such as model errors have little influence on the tracking effect of the resistance acceleration, so that the T entering the initial stage and the end stagepSelecting a smaller value, and reducing the calculation amount for predicting the output; the flight state of the spacecraft in the middle section changes rapidly, the resistance acceleration is large, so that the change of the tracking error is severe, and a large T is selectedpTo improve the robustness of the system. The invention provides a variable prediction time domain strategy, which takes the magnitude of the error change rate between the prediction output and the expected output as the judgment basis, reasonably selects the magnitudes of the prediction time domains at different flight phases and improves the performance of the NMPC guidance system.
The error between the predicted output and the desired output at time t is first defined,
Figure BDA00026102248200001112
the rate of change of the calculated error is as follows,
Figure BDA0002610224820000121
wherein k is1(t) and k2(t) the rate of change of the error in the resistive acceleration and its first derivative, respectively. The predicted time domain T at time TpI.e. the predicted number of steps Np(t) is determined by the following formula,
Figure BDA0002610224820000122
wherein λ1And λ2Predict the number of steps N for an appropriate rate of change threshold1<N2。
Finally, the invention adopts a hybrid algorithm combining the sequential quadratic programming and the particle swarm optimization algorithm to process the nonlinear programming problem with constraints formed by the formulas (13) and (14). The PSO algorithm has strong global optimization capability in the initial iteration stage, can quickly converge to a position close to a global optimal solution in the constraint range of the roll angle control quantity in the formula (14), and then the SQP algorithm takes the position as an initial value of optimization and searches nearby the position so as to converge to the global optimal solution. The hybrid optimization algorithm avoids the problem that the SQP algorithm is easy to generate a local optimal solution to a great extent, and the optimal solution can be found more quickly and accurately.
Recording optimized solution vectors obtained by solving NLP problem through hybrid optimization algorithm as
Figure BDA0002610224820000123
Only the first optimal solution of the obtained optimized sequence is acted on the actual system, namely the roll angle control quantity at the time t is,
Figure BDA0002610224820000124
and at the next moment, predicting future output according to the actual flight state of the spacecraft, repeatedly solving the NLP problem, optimizing forward in a rolling mode, and realizing closed-loop optimal control.
The distribution range and form of various random errors shown in table 1 are set, including aerodynamic coefficient, atmospheric density error, and initial state error. Under the condition that the random error exists, a designed guidance system based on an NMPC method is used for tracking the reference track generated in the step one, and values of all parameters of a control system are shown in a table 2. The results of the single simulation are shown in fig. 4-8. Fig. 4 is a height variation curve, fig. 5 is a speed variation curve, fig. 6 is a roll angle variation curve, and fig. 7 is a resistance acceleration variation curve. Fig. 8 is a longitude and latitude variation curve. According to the result, the NMPC guidance system can well track the reference track, and the guidance precision can meet the requirement.
TABLE 1 simulation error term parameters
Error of the measurement Error value Type (B)
Coefficient of aerodynamics CL,CD ±30％ Gaussian distribution
Atmospheric density ρ ±30％ Is uniformly distributed
Height (Km) ±0.2 Gaussian distribution
Longitude (deg) ±0.02 Gaussian distribution
Latitude (deg) ±0.02 Gaussian distribution
Speed (m/s) ±2 Gaussian distribution
Course angle (deg) ±0.05 Gaussian distribution
Track angle (deg) ±0.03 Gaussian distribution
TABLE 2 control of system parameters
Parameter(s) Numerical value
Nc 3
Q [100,0；0,500]
R 10
δ 0.01
η [1,1]
[N1,N2] [5,10]
[λ1,λ2] [0.05,0.01]
Although the designed NMPC guidance system can realize high-precision guidance, the design method still has some defects: solving a complex NLP problem in each guidance period, and consuming a long time to obtain an optimal solution in an iteration mode; because numerical integration is performed on the kinetic equation when the prediction output is calculated by using the prediction model, a large calculation load is brought when the prediction time domain is large. In order to overcome the defects, an intelligent predictive control entry guidance method is provided by combining the respective advantages of an NMPC method and a neural network.
The flow chart for realizing the intelligent prediction of the spacecraft to enter the guidance is shown in fig. 9, and can be summarized into sample data generation, network off-line training and on-line intelligent guidance, namely steps three to five.
Step three: and generating sample data by taking the NMPC guidance system designed in the step two as a guidance template. The process of data generation includes the steps of,
(1) setting a nominal initial condition I ═ x0,CL,CD,ρ]TIncluding an initial state x0(altitude, speed, latitude and longitude, track angle and course angle), aerodynamic coefficient CLAnd CDAnd the atmospheric density ρ.
(2) Introducing an error term pl＝[Δx0,ΔCL,ΔCD,Δρ]TThe distribution range and form of each error are shown in the table1, where l is 1, …, and m is the number of tracking tracks. The actual initial condition of the system is dl＝I+pl。
(3) With dlAs an actual initial condition, the reference trajectory is tracked using an NMPC guidance system. According to the state quantity of the current time
Figure BDA0002610224820000131
And (4) optimizing and solving the corresponding NLP problem to obtain an optimal control quantity u, forming a pair of arrays (x, u) and storing. Then rolling forward to the next optimization moment, and repeating the process until the actual condition d is completedlAnd entering a guidance process under the condition, wherein all the obtained array pairs are the sample data generated by the tracking.
(4) Generating a new error term pl+1Obtaining corresponding actual initial conditions dl+1And repeating the generation process of the sample data in the step (3) until l is m. Finally, obtaining a sample data set for off-line training of the neural network
Figure BDA0002610224820000132
s-1, …, mi, where i is the number of times the optimization problem is rolled out each time the reference trajectory is tracked. To ensure training accuracy, m is 1000, so the data set Q contains approximately 2 × 107And (4) array pairs of state quantity-optimal control quantity.
Step four: and training a neural network algorithm by using the data set to obtain the neural network based on the NMPC guidance system.
The BP neural network is a multilayer forward feedback neural network trained according to an error back propagation algorithm, has strong self-learning capability, can fully mine the relation between data, approaches a nonlinear complex system to the maximum extent, and is one of the most widely applied neural network models at present. Thus, embodiments of the present invention use the flight state quantities in data set Q
Figure BDA00026102248200001420
As the input of BP neural network, the roll angle corresponding to the current state quantity
Figure BDA00026102248200001419
And as the output of the BP neural network, training and testing the neural network to approximately obtain the relationship between the state quantity of the spacecraft and the corresponding roll angle. Randomly partitioning a data set Q into training sets Q1And test set Q2The ratios are set at 95% and 5%.
The BP neural network is composed of an input layer, an output layer and one or more hidden layers. As shown in fig. 9, the present embodiment adopts a four-layer neural network structure, that is, includes two hidden layers, which can improve the training precision of the BP neural network.
An input layer:
Figure BDA0002610224820000141
representing an input matrix in which each input vector
Figure BDA0002610224820000142
s is 1, …, mi, which contains six flight state variables.
Hidden layer: and determining the appropriate number of the hidden layer nodes through experience and repeated experiments, wherein the number of the first hidden layer node is 6, and the number of the second hidden layer node is 4. The transfer function is selected from a hyperbolic tangent sigmoid function tansig with a faster convergence rate and a wider output range.
An output layer:
Figure BDA0002610224820000143
the output matrix is represented by a matrix of values,
Figure BDA0002610224820000144
representing a desired output matrix, wherein each output variable is
Figure BDA0002610224820000145
s is 1, …, mi. The output layer transfer function is a linear transfer function purelin.
Before training, the training set Q is firstly compared1Input matrix and output matrix of (1)Normalization processing is performed to normalize the value to [ -1, 1]And the training efficiency of the network is improved.
Supervised learning of the neural network is a process of adjusting weights w and thresholds B between nodes of each layer to reduce network output errors.
Figure BDA0002610224820000146
And
Figure BDA0002610224820000147
representing the weight and threshold from the input layer to the first hidden layer;
Figure BDA0002610224820000148
and
Figure BDA0002610224820000149
representing weights and thresholds between two hidden layers;
Figure BDA00026102248200001410
and
Figure BDA00026102248200001411
representing the weight and threshold from the hidden layer to the output layer. Input vector
Figure BDA00026102248200001412
The state quantities are summed according to the corresponding weight and threshold, the output of each neuron is obtained through the calculation of a transfer function, and the information is transferred to an output layer through two hidden layers in sequence
Figure BDA00026102248200001413
The output values of the hidden layer and the output layer are shown as follows,
Figure BDA00026102248200001414
wherein f is1aAnd f2bRepresenting the output values of the first and second hidden layers respectively,
Figure BDA00026102248200001415
representing the output value of the output layer.
To prevent overfitting, improve the accuracy of the network approximation, estimate the loss function of the network performance using the mean square error,
Figure BDA00026102248200001416
wherein C represents the size of the error,
Figure BDA00026102248200001417
which represents the input value of the sample,
Figure BDA00026102248200001418
represents the sample output value and mi represents the total number of samples. And (4) reversely transmitting the obtained network output error C, and continuously adjusting the weight and the threshold between nodes of each layer so as to reduce the output error.
The network training algorithm adopts Levenberg-Marquardt algorithm. Compared with the traditional BP neural network training algorithm, the L-M algorithm can achieve the advantages of combining the Gauss-Newton algorithm and the gradient descent method by adaptively adjusting the damping factor, and the training speed of the neural network is accelerated.
Updating the weight value and the threshold value of each neuron by adopting a momentum gradient descent algorithm, wherein the formula is as follows,
Figure BDA0002610224820000151
wherein
Figure BDA0002610224820000154
Is the current loss function pair weight wnPartial derivatives (gradients) of (a), using exponentially weighted averages of the preceding
Figure BDA0002610224820000155
In connection with this, a momentum gradient is obtained
Figure BDA0002610224820000152
The updating formula of the threshold is the same as the weight. Beta is a momentum factor, which affects the exponentially weighted average, usually 0.9; α is the learning rate, i.e., the magnitude of each parameter update, affecting the convergence rate, here taken to be 0.05.
Based on training set Q1Training the determined BP network structure, wherein the target error of the network training is set to be 1 x 10-7The maximum number of iterations is set to 5000. And finishing the training of the BP neural network when the value of the loss function is converged to a set error range or reaches the maximum iteration number of the training. Using test set Q2The data in the step (1) tests the network performance, and the neural network with better training effect and the corresponding weight and threshold are finally obtained through multiple experiments.
Step five: and (4) taking the error of the actual guidance process into consideration, replacing the NMPC guidance system with the BP neural network trained in the fourth step, and carrying out online tracking on the reference track generated in the first step.
When the spacecraft enters the guidance closed loop optimal control, the problem of solving the nonlinear programming with the constraint and the integral prediction process are not required to be solved repeatedly in each guidance period, and only the state quantity of the spacecraft at the current t moment obtained by real-time measurement of the navigation system is required
Figure BDA0002610224820000153
Inputting the data into an offline trained BP neural network, performing weighting operation according to a predetermined weight and a predetermined threshold value, and calculating and outputting an optimal roll angle control quantity U required by guidance entering at the current time tt＝σtThe value of (c). And repeating the process of generating the control quantity at the previous moment to the next moment, so that the guidance instruction can be quickly solved on line, and the intelligent guidance of the spacecraft is realized.
Assuming that random errors in the actual guidance process are shown in table 1, 100 monte carlo targeting simulations were performed using the trained neural network, and the results are shown in fig. 10. Statistical analysis is performed on the actual terminal position and the distance from the target point position, and the calculation result is shown in table 3. Therefore, the guidance precision of the method designed by the invention can well meet the requirement on the drop point precision. In addition, compared with the average consumed time of about 60ms for solving the optimization problem each time by the NMPC guidance system, the trained neural network can generate the guidance instruction only by carrying out simple weighted operation, so that the instruction calculation average time at each moment is only 9ms (the simulation operation environment is Matlab R2019a programming language under a Windows 10 operating system, and the processor is Intel Core i7-9750H 2.60Ghz), the problem that the solution time of the NMPC method is long is solved, and guidance real-time performance is guaranteed.
TABLE 3 simulation results
Mean distance error Probability of 5 km 3 km probability Probability of 2 km
1.59km 99％ 89％ 73％
The spacecraft intelligent predictive control entry guidance method firstly designs a non-linear model predictive control guidance system for spacecraft atmospheric entry according to a proposed prediction model correction method based on error information estimation and a variable prediction time domain strategy, and solves a corresponding NLP problem in a rolling mode by using a hybrid optimization algorithm to obtain an optimal control quantity meeting control constraints and path constraints. And then, by combining the strong self-learning capability of the neural network, based on sample data generated by the NMPC guidance template, the mapping relation between the real-time flight state and the guidance instruction is approximated by adopting the multilayer BP neural network, so that the speed of solving the guidance instruction on line is increased, and the intelligent and accurate guidance of the spacecraft atmosphere entering section can be realized.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can understand that the modifications or substitutions within the technical scope of the present invention are included in the scope of the present invention, and therefore, the scope of the present invention should be subject to the protection scope of the claims.

Claims (3)

1. The spacecraft intelligent prediction control entry guidance method is characterized by comprising the following steps:
the method comprises the following steps: according to the state quantity of the initial moment of the atmospheric entering section of the spacecraft and the target constraint condition of the terminal of the entering section, a reference track is generated in an optimized mode, a corresponding rolling angle change curve is obtained, the change relation of the resistance acceleration and the energy is extracted, and the relation of the change of the derivative of the resistance acceleration along with the time is obtained through deduction;
step two: establishing a spacecraft entry guidance model, and designing an NMPC guidance system by adopting a nonlinear model predictive control NMPC method;
the kinetic equation of the spacecraft entering the guidance model is as follows:
Figure FDA0003292561570000011
Figure FDA0003292561570000012
Figure FDA0003292561570000013
wherein, the spacecraft state vector at the time t
Figure FDA0003292561570000014
r is the distance from the center of mass of the spacecraft to the target, V is the speed of the spacecraft,
Figure FDA0003292561570000015
the latitude is, theta is longitude, gamma is track angle, psi is heading angle, u (t) is control input at the time t, namely roll angle sigma (t); y isc(t) is the output value at time t, D (t) and
Figure FDA00032925615700000119
the resistance acceleration and the first derivative thereof at the time t; y isb(t) is the value of the path constraint at time t,
Figure FDA00032925615700000118
is the heat flow rate at time t, q (t) is the incoming flow pressure at time t, nL(t) spacecraft overload at time t;
the NMPC guidance system is designed by adopting a nonlinear model predictive control NMPC method, and the method specifically comprises the following steps:
1) at time t, a constrained nonlinear programming problem of the form:
Figure FDA0003292561570000016
the following constraints are satisfied:
Figure FDA0003292561570000017
Figure FDA0003292561570000018
Figure FDA0003292561570000019
Figure FDA00032925615700000110
wherein x (t) is the state of the spacecraft at time t; t iscAnd TpRespectively a control time domain and a prediction time domain, and satisfies Tc≤Tp；σminFor minimum value of allowable roll angle, σmaxIn order to allow for a maximum value of roll angle,
Figure FDA00032925615700000111
to maximize the allowable roll angle rate, δ is the sampling time, Tc＝Ncδ，Tp＝Npδ，NcFor controlling the number of steps, NpPredicting the step number;
Figure FDA00032925615700000112
Figure FDA00032925615700000113
for the upper limit of the allowable heat flow rate, qmaxFor an allowable upper limit of dynamic pressure, nLmaxIs the allowable overload upper limit; the objective function J at time t is specifically defined as
Figure FDA00032925615700000114
Figure FDA00032925615700000115
Figure FDA00032925615700000116
And
Figure FDA00032925615700000117
the predicted value of the resistance acceleration and the first derivative thereof at the t moment to the tau moment is represented;
Figure FDA0003292561570000021
Dr(τ) and
Figure FDA0003292561570000022
representing the desired output at time τ, i.e. the resistive acceleration of the reference track and its first derivative, ur(τ) with yr(τ) a corresponding control input, namely roll angle;
Figure FDA0003292561570000023
control input value representing the time T to be solved at time T for any time T e [ T, T + T ∈c]，
Figure FDA0003292561570000024
i-int (τ -t/δ), int (·) being a downward rounding function;
Figure FDA0003292561570000025
to-be-solved optimization sequence U for forming nonlinear programming problem with constraint at time ttI.e. by
Figure FDA0003292561570000026
Figure FDA0003292561570000027
A predicted value representing the path constraint output from the time t to the time tau; q is an output error weighting matrix, and R is a control quantity weighting matrix;
Figure FDA0003292561570000028
and
Figure FDA0003292561570000029
calculated by the following prediction model:
Figure FDA00032925615700000210
wherein
Figure FDA00032925615700000211
For the predicted value of the state quantity at time t to time t,
Figure FDA00032925615700000212
the values of r, V, θ,
Figure FDA00032925615700000213
predicted values of γ, ψ;
2) correcting the prediction model in 1) by adopting a prediction model correction method based on error information estimation:
first, a first-order attenuation memory filter is designed to obtain error estimation information, wherein the first-order attenuation memory filter is Z (t) -Z (t-delta) + (1-epsilon) (Z)*(t) -Z (t- δ)), the state quantity at time t of the first-order attenuation memory filter is
Figure FDA00032925615700000214
Z (t-delta) represents the state quantity of the filter at the t-delta moment, Z (0) is 1, 0 & ltepsilon & lt 1 is a gain coefficient, L (t) represents the actual lift acceleration at the t moment,
Figure FDA00032925615700000215
and
Figure FDA00032925615700000216
respectively representing the predicted values of the prediction model at the time t to the resistance acceleration and the lift acceleration;
secondly, the output Z (t) of the first-order attenuation memory filter is used as a correction factor of the prediction model, and
Figure FDA00032925615700000217
and
Figure FDA00032925615700000218
make real-time corrections, i.e.
Figure FDA00032925615700000219
And
Figure FDA00032925615700000220
respectively predicting values of the corrected resistance acceleration and the corrected lift acceleration;
finally, the prediction model is modified as follows:
Figure FDA00032925615700000221
wherein
Figure FDA00032925615700000222
Correcting the predicted value of the prediction model to the state quantity at the time tau for the time t,
Figure FDA00032925615700000223
represents the predicted value output by the tau time correction prediction model to the tau time,
Figure FDA00032925615700000224
the predicted value of the path constraint output at the time t of the correction prediction model is represented;
3) and (3) adding a correction link for prediction errors at each optimization moment of the constrained nonlinear programming problem:
first, the error between the predicted output and the actual output at time t is defined:
Figure FDA0003292561570000031
wherein
Figure FDA0003292561570000032
At time tCorrecting the predicted output value of the prediction model, wherein y (t) is the actual output value at the time t;
secondly, correcting the predicted value output by the modified prediction model at the time t to the time tau in a manner of weighting e (t):
Figure FDA0003292561570000033
wherein the content of the first and second substances,
Figure FDA0003292561570000034
for the purpose of the corrected predicted output value,
Figure FDA0003292561570000035
and
Figure FDA0003292561570000036
predicted values of the corrected resistance acceleration and its first derivative, eta ═ eta1，η2]Is a correction vector, and η1＞0，η2＞0；
4) After the correction of the prediction model and the feedback correction of the prediction error, the objective function J at the time t is redefined as:
Figure FDA0003292561570000037
step three: considering errors existing in the process of entering the atmosphere, setting the distribution range and the form of random errors of each state quantity, performing offline tracking on a reference track for multiple times based on the NMPC guidance system designed in the step two, and storing the optimal control quantity at each moment and each corresponding state quantity obtained by performing rolling solution by using a hybrid optimization algorithm when performing tracking each time to generate a sample data set taking the NMPC guidance system as a guidance template;
step four: taking the state quantity of the sample data set generated in the third step as input, taking the corresponding optimal control quantity as output, carrying out back propagation BP neural network training, finishing the training of the BP neural network when the value of the loss function converges to a set error range or reaches the maximum iteration number of the training, and obtaining the neural network based on the NMPC guidance system;
step five: and (3) considering errors existing in the actual guidance process, adopting a neural network based on an NMPC guidance system in the fourth step to track the reference track generated in the first step on line: and inputting the state quantity obtained by real-time measurement of the spacecraft navigation system into a neural network based on an NMPC guidance system to obtain the corresponding optimal control quantity, thereby realizing intelligent guidance of the spacecraft.
2. The spacecraft intelligent predictive control entry guidance method of claim 1, wherein step one is specifically: according to the speed, longitude and latitude, height, course angle, track angle of an initial moment of an atmospheric entering section of the spacecraft and target constraint conditions of a terminal of the entering section, under the condition that path constraint including overload, dynamic pressure and heat flow rate and control quantity constraint are met, a reference track is generated in an optimized mode, a corresponding rolling angle change curve is obtained, the change relation of the resistance acceleration and the energy is extracted, and the relation of the change of the derivative of the resistance acceleration along with the time is obtained through deduction.
3. The spacecraft intelligent predictive control entry guidance method of claim 1, wherein in step three, a hybrid optimization algorithm combining a Sequence Quadratic Programming (SQP) algorithm and a Particle Swarm Optimization (PSO) algorithm is adopted to solve the nonlinear programming problem with constraints: firstly, solving a nonlinear programming problem with constraint by using a PSO algorithm, and then searching the solving result of the PSO algorithm as an optimized initial value near the position by using an SQP algorithm so as to converge to a global optimal solution.
CN202010751626.1A 2020-07-30 2020-07-30 Intelligent prediction control entry guidance method for spacecraft Active CN112114521B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010751626.1A CN112114521B (en) 2020-07-30 2020-07-30 Intelligent prediction control entry guidance method for spacecraft

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010751626.1A CN112114521B (en) 2020-07-30 2020-07-30 Intelligent prediction control entry guidance method for spacecraft

Publications (2)

Publication Number Publication Date
CN112114521A CN112114521A (en) 2020-12-22
CN112114521B true CN112114521B (en) 2022-03-01

Family

ID=73799282

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010751626.1A Active CN112114521B (en) 2020-07-30 2020-07-30 Intelligent prediction control entry guidance method for spacecraft

Country Status (1)

Country Link
CN (1) CN112114521B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113031598B (en) * 2021-03-02 2024-02-06 中国人民解放***箭军工程大学 Path tracking obstacle avoidance guidance method and system
CN114115302B (en) * 2021-06-08 2022-08-16 中国人民解放军国防科技大学 Spacecraft attitude control method based on deep neural network approximation MPC
CN114227687B (en) * 2021-12-28 2023-08-15 深圳市优必选科技股份有限公司 Robot control method and device, terminal equipment and storage medium
CN113985740B (en) * 2021-12-30 2022-05-06 中国科学院空天信息创新研究院 Stability control method and device based on particle active disturbance rejection
CN114967729B (en) * 2022-03-28 2022-12-13 广东工业大学 Multi-rotor unmanned aerial vehicle height control method and system
CN115268276B (en) * 2022-09-26 2023-02-03 北京航天自动控制研究所 Gradient descent-based guidance parameter online correction method and system
CN115586721A (en) * 2022-10-08 2023-01-10 厦门大学 Short-range accurate guidance method based on model predictive control
CN115793675B (en) * 2023-02-09 2023-05-16 陕西欧卡电子智能科技有限公司 Coupling calibration method and device for unmanned ship power system and computer equipment
CN116911004A (en) * 2023-07-06 2023-10-20 山东建筑大学 Trajectory drop point correction method based on neural network
CN117311390B (en) * 2023-11-06 2024-03-19 天津大学 Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6185470B1 (en) * 1997-11-07 2001-02-06 Mcdonnell Douglas Corporation Neural network predictive control method and system
CN105222648A (en) * 2015-09-11 2016-01-06 北京航空航天大学 A kind of linear pseudo-spectrum broad sense mark control miss distance Guidance and control method
CN106054604A (en) * 2016-06-02 2016-10-26 西北工业大学 Reentry aircraft robust optimal guidance method based on model predictive control theory
CN106707751A (en) * 2016-12-20 2017-05-24 哈尔滨工业大学 Collision avoidance control method based on limited time saturation for spacecraft terminal approaching
CN107121929A (en) * 2017-05-23 2017-09-01 西北工业大学 Robust reentry guidance method based on linear covariance model PREDICTIVE CONTROL
CN107202584A (en) * 2017-07-06 2017-09-26 北京理工大学 A kind of planet precision landing anti-interference method of guidance
CN109976349A (en) * 2019-04-12 2019-07-05 大连海事大学 A kind of design method containing the path trace guidance and control structure that constrain unmanned boat
CN110989669A (en) * 2019-12-11 2020-04-10 西安智翔防务技术有限公司 Online self-adaptive guidance algorithm for active section of multistage boosting gliding aircraft
CN111351488A (en) * 2020-03-03 2020-06-30 南京航空航天大学 Intelligent trajectory reconstruction reentry guidance method for aircraft

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6185470B1 (en) * 1997-11-07 2001-02-06 Mcdonnell Douglas Corporation Neural network predictive control method and system
CN105222648A (en) * 2015-09-11 2016-01-06 北京航空航天大学 A kind of linear pseudo-spectrum broad sense mark control miss distance Guidance and control method
CN106054604A (en) * 2016-06-02 2016-10-26 西北工业大学 Reentry aircraft robust optimal guidance method based on model predictive control theory
CN106707751A (en) * 2016-12-20 2017-05-24 哈尔滨工业大学 Collision avoidance control method based on limited time saturation for spacecraft terminal approaching
CN107121929A (en) * 2017-05-23 2017-09-01 西北工业大学 Robust reentry guidance method based on linear covariance model PREDICTIVE CONTROL
CN107202584A (en) * 2017-07-06 2017-09-26 北京理工大学 A kind of planet precision landing anti-interference method of guidance
CN109976349A (en) * 2019-04-12 2019-07-05 大连海事大学 A kind of design method containing the path trace guidance and control structure that constrain unmanned boat
CN110989669A (en) * 2019-12-11 2020-04-10 西安智翔防务技术有限公司 Online self-adaptive guidance algorithm for active section of multistage boosting gliding aircraft
CN111351488A (en) * 2020-03-03 2020-06-30 南京航空航天大学 Intelligent trajectory reconstruction reentry guidance method for aircraft

Non-Patent Citations (10)

* Cited by examiner, † Cited by third party
Title
Finite time sliding sector control for spacecraft atmospheric entry guidance;Biao Xu等;《Acta Astronautica》;20181205;108-113 *
Neural network based online predictive guidance for high lifting vehicles;Zhenhua Li等;《Aerospace Science and Technology》;20180907;149-160 *
基于多模型预测的再入飞行器制导方法;张军等;《航空学报》;20080530;25-30 *
基于深度学习的高超声速飞行器再入预测校正容错制导;余跃等;《兵工学报》;20200415(第04期);34-47 *
多对空导弹拦截多目标的协同制导方法研究;葛鲁亲等;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20200215(第02期);全文 *
多约束条件下的制导律研究综述;孙未蒙等;《飞行力学》;20100415(第02期);3-7 *
有推力约束航天器交会的模型预测制导策略;叶东等;《哈尔滨工程大学学报》;20171231(第05期);103-109 *
滑翔式飞行器再入制导与控制方法研究;吴旭忠;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20150715(第07期);全文 *
火星大气进入滑模自抗扰制导方法;李翔等;《飞控与探测》;20190731;第2卷(第4期);37-45 *
线性伪谱模型预测能量最优姿态机动控制方法;冯逸骏等;《北京航空航天大学学报》;20180418(第10期);148-158 *

Also Published As

Publication number Publication date
CN112114521A (en) 2020-12-22

Similar Documents

Publication Publication Date Title
CN112114521B (en) Intelligent prediction control entry guidance method for spacecraft
CN111351488B (en) Intelligent trajectory reconstruction reentry guidance method for aircraft
Cheng et al. Multiconstrained real-time entry guidance using deep neural networks
CN110806759B (en) Aircraft route tracking method based on deep reinforcement learning
Han et al. Online policy iteration ADP-based attitude-tracking control for hypersonic vehicles
CN111813146B (en) Reentry prediction-correction guidance method based on BP neural network prediction voyage
CN113110560B (en) Satellite formation reconstruction model prediction control method based on Chebyshev inequality
CN113377121B (en) Aircraft intelligent disturbance rejection control method based on deep reinforcement learning
CN109062040B (en) PID (proportion integration differentiation) predicting method based on system nesting optimization
CN111007724A (en) Hypersonic aircraft designated performance quantitative tracking control method based on interval II type fuzzy neural network
CN112257847A (en) Method for predicting geomagnetic Kp index based on CNN and LSTM
Dai et al. Entry trajectory optimization for hypersonic vehicles based on convex programming and neural network
Wang et al. Intelligent control of air-breathing hypersonic vehicles subject to path and angle-of-attack constraints
Zhou et al. Optimal guidance for hypersonic vehicle using analytical solutions and an intelligent reversal strategy
Cheng et al. Adaptive control of hypersonic vehicles with unknown dynamics based on dual network architecture
CN110851911B (en) Terminal state calculation model training method, control sequence searching method and device
Du et al. Adaptive recurrent-functional-link-network control for hypersonic vehicles with atmospheric disturbances
CN114237268A (en) Unmanned aerial vehicle strong robust attitude control method based on deep reinforcement learning
Lew et al. Chance-constrained optimal altitude control of a rocket
CN116432539A (en) Time consistency collaborative guidance method, system, equipment and medium
Chen et al. Trajectory tracking guidance of interceptor via prescribed performance integral sliding mode with neural network disturbance observer
Tang et al. Reconfigurable Fault Tolerant Control for nonlinear aircraft based on concurrent SMC-NN adaptor
Inanc et al. Long short-term memory for improved transients in neural network adaptive control
Jia et al. Robust Guidance Control of Vertical Landing Aircraft with Deep Reinforcement Learning Algorithm
Liao et al. Rocket sensor data prediction based on LSTM with attention mechanism

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant