Intelligent predictive guidance method for aerobraking of mars

Abstract

This invention relates to an intelligent predictive guidance method for Mars aerodynamic-assisted orbit descent, belonging to the field of deep space exploration. The invention constructs a high-precision dynamic model for Mars aerodynamic-assisted orbit descent, uses this model to generate a network training dataset, and builds intelligent prediction models for the near-Mars and far-Mars states. Simultaneously, it establishes near-Mars heat flow constraints and designs a predictive guidance method using orbital altitude deviation and control input as optimization performance indicators. The method optimizes and solves the control input to control the probe until its orbital altitude is lowered to near the predetermined orbital altitude, achieving the descent of the probe to the predetermined orbit while satisfying heat flow and control amplitude constraints and reducing fuel consumption. This invention reduces the training cost of the neural network by solving the near-Mars and far-Mars states under the two-body dynamic model and constructing a network to predict the deviation between the near-Mars and far-Mars states under the high-precision dynamic model and the two-body model prediction results, thereby solving for the near-Mars and far-Mars states under the two-body dynamic model.

Description

Technical Field

[0001] This invention relates to an intelligent predictive guidance method, and more particularly to an intelligent predictive guidance method for Mars aerodynamic-assisted orbit descent, belonging to the field of deep space exploration technology. Background Technology

[0002] Mars exploration is of great significance in the search for the origins of the universe and life, as well as the search for extraterrestrial life and habitable planets, and has therefore always been a focus of deep space exploration for major spacefaring nations. To reduce fuel consumption during the capture and descent phases, aerodynamic-assisted orbit control technology, utilizing aerodynamic flight for orbit adjustment on atmospheric bodies like Mars, has received widespread attention. In Mars aerodynamic-assisted descent, the use of predictive guidance methods not only enables precise descent of the probe to the predetermined orbit while meeting heat flux and control amplitude constraints, but also reduces fuel consumption, allowing the probe to carry more scientific payloads and significantly improving the scientific benefits and resource utilization efficiency of Mars exploration missions.

[0003] Because the aerodynamic assisted descent process on Mars is complex and lengthy, the dynamic model used in the prediction guidance method must consider the influence of the irregular spherical gravitational perturbation of Mars in order to make the predicted probe state as close as possible to the actual state. However, using a low-order irregular spherical gravitational model of Mars will result in a large deviation between the predicted trajectory and the actual trajectory, failing to meet the accuracy requirements; while using a high-order irregular spherical gravitational model of Mars can improve accuracy, it will significantly reduce computational efficiency, making it difficult to meet the real-time requirements of orbital applications. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent predictive guidance method for Mars aerodynamic-assisted orbit descent. This method utilizes a high-precision dynamic model of Mars aerodynamic-assisted orbit descent to generate training data, uses a neural network to learn the mapping relationship between input and output, constructs a fast and high-precision intelligent state prediction model, and uses the constructed fast and high-precision state prediction model to build a predictive guidance method. Under the conditions of satisfying heat flow and control amplitude constraints, the probe is accurately descended to the predetermined orbit, while reducing the probe's fuel consumption.

[0005] The objective of this invention is achieved through the following technical solution.

[0006] This invention discloses an intelligent predictive guidance method for Mars aerodynamic-assisted descent. It constructs a high-precision dynamic model for Mars aerodynamic-assisted descent. Considering that control is mainly applied at the far Mars point, and that the control inputs during the near Mars point state and attachment process must satisfy heat flux and control amplitude constraints, a neural network is used to construct intelligent prediction models for near Mars point state (predicting the near Mars point state from the initial state) and far Mars point state (predicting the far Mars point state from the initial state). The dataset used for neural network training is generated from the constructed high-precision dynamic model for Mars aerodynamic-assisted descent. During dataset generation, the range of initial state values ​​is defined, and sampling is performed based on a uniform distribution. The high-precision dynamic model for Mars aerodynamic-assisted descent is used to recursively calculate the near Mars point state and far Mars point state at the next moment. The near Mars point state correction is obtained by subtracting the near Mars point state obtained from the two-body dynamic model from the high-precision dynamic model, and the far Mars point state correction is obtained by subtracting the far Mars point state obtained from the two-body dynamic model. The initial state of the probe is used as input, and the near-fire point state correction is used as output to form the training dataset for the near-fire point state correction prediction network. The initial state of the probe is used as input, and the far-fire point state correction is used as output to form the training dataset for the far-fire point state correction prediction network. After training the near-fire point state correction prediction networks, the near-fire point and far-fire point states obtained from two-body dynamics are corrected to construct intelligent prediction models for the near-fire point state and far-fire point state of Mars aerodynamically assisted descent. The intelligent prediction model for the near-fire point state is used to construct the near-fire point heat flow constraint equation and the control amplitude constraint equation. At the same time, the intelligent prediction model for the far-fire point state is used as the state prediction equation. With orbital altitude deviation and control input as optimization performance indicators, a predictive guidance method is designed. The current state of the probe is used as input to optimize the solution of the control quantity of the probe at the far-fire point. Under the condition of satisfying the heat flow and control amplitude constraints, the probe is accurately descended to the predetermined orbit, while reducing the probe's fuel consumption. This is the intelligent predictive guidance method for Mars aerodynamically assisted descent.

[0007] The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method disclosed in this invention includes the following steps:

[0008] Step 1: Construct a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Mars inertial frame. In the subsequent Step 2, the constructed high-precision dynamic model will be used to generate a dataset for neural network learning.

[0009] Constructing a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Martian inertial frame.

[0010]

[0011] Where r is the distance of the probe from the center of Mars, θ and φ represent the latitude and longitude of the probe in the Martian inertial frame, respectively, V is the magnitude of the probe's velocity, γ and ψ represent the probe's track angle and heading angle, respectively, D and L represent the magnitudes of aerodynamic drag and aerodynamic lift acceleration, respectively, and g, l, and f represent the radial, longitude, and latitudinal components of the non-spherical gravity of Mars, respectively. The Martian full-gradient gravity model...

[0012]

[0013] The Martian atmospheric model is represented by the Martian atmospheric correction index model.

[0014]

[0015] Where h represents the orbital altitude of the probe. Given a Martian atmospheric model, the aerodynamic drag and aerodynamic lift acceleration in the probe's dynamic equations are expressed as follows:

[0016]

[0017] Among them, C D and C L ρ represents the drag coefficient and lift coefficient of the probe, respectively; S is the effective frontal area of ​​the probe; V is the velocity of the probe; m is the mass of the probe; and ρ is the atmospheric density determined by the Martian atmospheric model.

[0018] Because the aerodynamic descent process is relatively long, the influence of the irregular spherical gravitational perturbation of Mars needs to be considered when designing the predictive guidance method. Using a low-cost gravitational model in the probe's dynamic equations will result in a significant deviation between the predicted trajectory and the actual trajectory, failing to meet accuracy requirements. Conversely, using a high-order gravitational model would severely reduce computational efficiency, making on-orbit application difficult. To characterize the impact of the irregular spherical gravitational perturbation of Mars on the probe's state in a high-precision dynamic model for aerodynamic descent, while simultaneously improving model accuracy and computational efficiency, step 2 utilizes the high-precision dynamic model for aerodynamic descent to generate training data. A neural network is then used to learn the mapping relationship between input and output, constructing a fast and high-precision intelligent state prediction model.

[0019] Step 2: Generate the dataset needed for neural network training using the high-precision dynamic model constructed in Step 1. During dataset generation, the range of initial state values ​​is defined, and sampling is performed based on a uniform distribution. The high-precision dynamic model for Mars aerodynamic descent is used to recursively calculate the near-horizon and far-horizon states at the next moment. The near-horizon state obtained from the high-precision dynamic model is subtracted from the near-horizon state obtained from the two-body dynamic model to obtain the near-horizon state correction. The far-horizon state obtained from the high-precision dynamic model is subtracted from the far-horizon state obtained from the two-body dynamic model to obtain the far-horizon state correction. The initial state of the probe is used as input, and the near-horizon state correction is used as output to form the dataset for training the near-horizon state correction prediction network.

[0020] During the Mars aerodynamic assisted orbit descent process, control is primarily applied at the apogee, while the perigee state must meet thermal flow constraints. Therefore, when using neural networks, intelligent prediction models for the perigee state and the apogee state, based on the initial state, will be constructed. In step 2, the high-precision dynamic model for Mars aerodynamic assisted orbit descent constructed in step 1 will be used to collect the initial state of the probe as input and the next perigee state as output; similarly, the high-precision dynamic model for Mars aerodynamic assisted orbit descent constructed in step 1 will be used to collect the initial state of the probe as input and the next apogee state as output.

[0021] The range of values ​​for the detector's initial state is determined using the detector's near-fire point altitude, far-fire point altitude, and the six roots of the orbit. h is defined as follows. p and h a Let i, Ω, w, and f be the near-fire altitude and far-fire altitude of the detector, respectively. Define i, Ω, w, and f as the detector's orbital inclination, right ascension of the ascending node, argument of the pericenter, and true anomaly, respectively. Then, the initial state range of the detector is:

[0022]

[0023] Among them, h p,min h a,min i min Ω min w min f min These are the initial near-fire altitude, far-fire altitude, orbital inclination, right ascension of the ascending node, argument of the pericenter, and lower bound of the true pericenter angle, h. p,max h a,max i max Ω max w max f maxThese are the detector's initial near-fire point altitude, far-fire point altitude, orbital inclination, right ascension of the ascending node, argument of the pericenter, and upper limit of the true pericenter angle, respectively.

[0024] After determining the initial state range of the detector, samples are taken within this range according to a uniform distribution, and the sampling results are converted into a state represented by latitude, longitude, and altitude, defined as x0=[r0,θ0,φ0,V0,γ0,ψ0]. T Given the selected initial state of the detector, the state of the detector at the near-fire point and the state of the detector at the far-fire point can be obtained by using high-precision dynamic integration.

[0025]

[0026] Where f represents the constructed high-precision dynamic model for Mars aerodynamic-assisted orbit descent, and t p and t a These represent the time from the initial moment to the next near point and the time from the initial moment to the next far point, respectively.

[0027] The time t for the detector to reach the next near point of fire from the initial moment p and the time t from the initial moment to the next distant firing point a It is difficult to estimate precisely, so its orbital period is estimated by using the semi-major axis determined by the initial state of the probe.

[0028]

[0029] Where T is the orbital period of the detector under simple two-body dynamics, a0 is the semi-major axis of the orbit, and μ m is the standard gravitational constant for Mars. After estimating the orbital period, a high-precision dynamic model for Mars aerodynamic-assisted orbit descent was used to integrate the probe's initial state for one revolution, and data on the next perihelion and next apohelion states corresponding to the initial state were collected.

[0030] Considering that under two-body dynamics, the next perihelion and apogee states can be analytically solved based on the probe's initial state, the mappings from the initial state to the next perihelion state and from the initial state to the next apogee state under Mars aerodynamic-assisted orbit descent high-precision dynamics can be divided into analytical quantities obtained from two-body dynamics and correction quantities caused by non-spherical gravitational perturbations and atmospheric drag. Definitions and To analyze the near-fire and far-fire states obtained using the simplified two-body dynamics model, the corrected near-fire state δx... p and the state of the far-field firing point δx a Represented as

[0031]

[0032] Based on the established range of detector initial state values, K samples are collected to construct a dataset that maps the detector initial state to near-fire point state corrections and far-fire point state corrections.

[0033]

[0034] Where X is the set of initial states of the detector, Y p and Y a These represent the sets of near-fire point state corrections and the sets of far-fire point state corrections, respectively.

[0035] Step 3: Construct a near-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to near-fire point state corrections for network training. Construct a far-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to far-fire point state corrections for network training. After training the near-fire point and far-fire point state correction prediction networks, correct the near-fire point and far-fire point states obtained from two-body dynamics to construct intelligent prediction models for Mars aerodynamically assisted descent near-fire point and far-fire point states. Compared to methods that directly construct networks to predict near-fire point and far-fire point states, this method solves for the near-fire point and far-fire point states under the two-body dynamics model, constructing the network to predict the deviation between the near-fire point and far-fire point states under the high-precision dynamics model and the two-body model prediction results, reducing the training cost of the neural network and improving the prediction accuracy of the neural network.

[0036] Constructing a near-fire state correction prediction network (DNN) p The dataset constructed in step 2, which maps the detector's initial state to near-fire point state corrections, is used for network training. A distant-fire point state correction prediction network (DNN) is then constructed. a The dataset constructed in step 2, which maps the initial state of the probe to the correction values ​​of the far-horse point state, is used for network training. After training the near-horse point state correction prediction network and the far-horse point state correction prediction network, the near-horse point state and the far-horse point state obtained from two-body dynamics are corrected to construct intelligent prediction models for the near-horse point state and the far-horse point state of Mars aerodynamically assisted descent orbit.

[0037]

[0038] Where, x k This indicates the state of the flexible lander at time k. and The near-fire point state and far-fire point state at time k+1 under two-body dynamics, respectively, x p,k+1 and x a,k+1This provides high-precision near-fire point and far-fire point status obtained through intelligent prediction.

[0039] Step 4: Construct the near-fire point heat flow constraint equation and the control amplitude constraint equation using the near-fire point state intelligent prediction model. At the same time, use the far-fire point state intelligent prediction model as the state prediction equation. With orbital altitude deviation and control input as optimization performance indicators, design a predictive guidance method. Use the current state of the probe as the input of the predictive guidance law, optimize the solution of the probe's control quantity at the far-fire point, and accurately lower the probe to the predetermined orbit under the conditions of satisfying heat flow and control amplitude constraints, while reducing the probe's fuel consumption. This is the Mars aerodynamic-assisted orbit lowering intelligent predictive guidance.

[0040] During the aerodynamically assisted descent to Mars, to protect the probe's safety, it is necessary to limit the heat flux density at the periapsis. This is based on the probe's apoapsis state x at time k+i. a,k+i and control input ΔV a,k+i The near-fire point heat flow constraint equation, constructed from the near-fire point state intelligent prediction model, is expressed as follows:

[0041]

[0042] Where, ρ p,k+i+1 Let γ be the atmospheric density near the fire point at time k+i+1, and let γ = [0,0,0,1,0,0]. T , The maximum permissible heat flux density.

[0043] Meanwhile, the thrust that the detector can apply has an upper limit, and the velocity change at the far-point of fire is constrained by the control amplitude.

[0044] |ΔV a,k+i |≤ΔV max (12)

[0045] Where, ΔV max This represents the upper bound of the velocity change at the far-field firing point.

[0046] When the probe's initial position is at its apogee, the Mars aerodynamic assisted descent begins. The Mars aerodynamic assisted descent prediction guidance, at each time k, is based on the current apogee state x. a,k To predict the input to the guidance law, the control input ΔV of the far-fire point is obtained by solving an optimization problem. a,k The optimization problem to be solved uses orbital altitude deviation and control input as optimization performance indicators, and a remote firing point state intelligent prediction model as the state prediction equation, while also considering heat flux constraints and control amplitude constraints.

[0047]

[0048] Where N is the number of predicted laps, Rm The radius of Mars. The desired orbital altitude of the far-orbiting flank. The far-orbiting flank control input ΔV within the predicted orbital number. a,k+i After optimization, ΔV is selected. a,k As the control input at time k.

[0049] When the probe is at its farthest point from Mars, the probe is controlled by optimizing the control input until its orbital altitude is reduced to near the predetermined orbital altitude. This achieves precise descent of the probe to the predetermined orbit while meeting the constraints of heat flow and control amplitude, and at the same time reduces the probe's fuel consumption. This is known as Mars aerodynamic-assisted orbit descent intelligent prediction guidance.

[0050] Beneficial effects:

[0051] 1. The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method disclosed in this invention constructs a high-precision dynamic model for Mars aerodynamic-assisted orbit descent, generates a network training dataset using the high-precision model, constructs intelligent prediction models for the near-Mars point and far-Mars point states, and simultaneously constructs near-Mars heat flow constraints. Using orbital altitude deviation and control input as optimization performance indicators, a prediction guidance method is designed. The prediction guidance method optimizes and solves the control input to control the probe until the probe's orbital altitude is reduced to near the predetermined orbital altitude. This achieves precise descent of the probe to the predetermined orbit under the conditions of satisfying heat flow and control amplitude constraints, while reducing the probe's fuel consumption, i.e., Mars aerodynamic-assisted orbit descent intelligent prediction guidance.

[0052] 2. Compared with the method of directly constructing a network to predict the near-horizon and far-horizon states, the Mars aerodynamic-assisted descent intelligent prediction guidance method disclosed in this invention solves the near-horizon and far-horizon states under the two-body dynamics model, constructs a network to predict the deviation between the near-horizon and far-horizon states under the high-precision dynamics model and the prediction results of the two-body model, reduces the training cost of the neural network, and improves the prediction accuracy of the constructed Mars aerodynamic-assisted descent near-horizon and far-horizon state intelligent prediction models. Attached Figure Description

[0053] Figure 1 Flowchart of intelligent predictive guidance method for Mars aerodynamic-assisted orbit descent;

[0054] Figure 2 A schematic diagram illustrating the training effect of the network for predicting near-fire point state corrections.

[0055] Figure 3 A schematic diagram illustrating the training effect of the network for predicting the state correction of distant firing points;

[0056] Figure 4 The far-field maneuver curve is given for the predictive guidance method. Detailed Implementation

[0057] To better illustrate the purpose and advantages of the present invention, the following description, in conjunction with an example and corresponding drawings, further explains the invention.

[0058] like Figure 1 As shown in the figure, the specific implementation steps of the Mars aerodynamic-assisted orbit descent intelligent prediction guidance method disclosed in this embodiment are as follows:

[0059] Step 1: Construct a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Mars inertial frame. In the subsequent Step 2, the constructed high-precision dynamic model will be used to generate a dataset for neural network learning.

[0060] Constructing a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Martian inertial frame.

[0061]

[0062] Where r is the distance of the probe from the center of Mars, θ and φ represent the latitude and longitude of the probe in the Martian inertial frame, respectively, V is the magnitude of the probe's velocity, γ and ψ represent the probe's track angle and heading angle, respectively, D and L represent the magnitudes of aerodynamic drag and aerodynamic lift acceleration, respectively, and g, l, and f represent the radial, longitude, and latitudinal components of the non-spherical gravity of Mars, respectively. The Martian full-gradient gravity model...

[0063]

[0064] The Martian atmospheric model is represented by the Martian atmospheric correction index model.

[0065]

[0066] Where h represents the orbital altitude of the probe. Given a Martian atmospheric model, the aerodynamic drag and aerodynamic lift acceleration in the probe's dynamic equations are expressed as follows:

[0067]

[0068] Among them, C D =2.2 and C L =0.31 are the drag coefficient and lift coefficient of the probe, respectively; S=36 is the effective frontal area of ​​the probe; V is the velocity of the probe; m=1225 is the mass of the probe; and ρ is the atmospheric density determined by the Martian atmospheric model.

[0069] Because the aerodynamic descent process is relatively long, the influence of the irregular spherical gravitational perturbation of Mars needs to be considered when designing the predictive guidance method. Using a low-cost gravitational model in the probe's dynamic equations will result in a significant deviation between the predicted trajectory and the actual trajectory, failing to meet accuracy requirements. Conversely, using a high-order gravitational model would severely reduce computational efficiency, making on-orbit application difficult. To characterize the impact of the irregular spherical gravitational perturbation of Mars on the probe's state in a high-precision dynamic model for aerodynamic descent, while simultaneously improving model accuracy and computational efficiency, step 2 utilizes the high-precision dynamic model for aerodynamic descent to generate training data. A neural network is then used to learn the mapping relationship between input and output, constructing a fast and high-precision intelligent state prediction model.

[0070] Step 2: Generate the dataset needed for neural network training using the high-precision dynamic model constructed in Step 1. During dataset generation, the range of initial state values ​​is defined, and sampling is performed based on a uniform distribution. The high-precision dynamic model for Mars aerodynamic descent is used to recursively calculate the near-horizon and far-horizon states at the next moment. The near-horizon state obtained from the high-precision dynamic model is subtracted from the near-horizon state obtained from the two-body dynamic model to obtain the near-horizon state correction. The far-horizon state obtained from the high-precision dynamic model is subtracted from the far-horizon state obtained from the two-body dynamic model to obtain the far-horizon state correction. The initial state of the probe is used as input, and the near-horizon state correction is used as output to form the dataset for training the near-horizon state correction prediction network.

[0071] During the Mars aerodynamic assisted orbit descent process, control is primarily applied at the apogee, while the perigee state must meet thermal flow constraints. Therefore, when using neural networks, intelligent prediction models for the perigee state and the apogee state, based on the initial state, will be constructed. In step 2, the high-precision dynamic model for Mars aerodynamic assisted orbit descent constructed in step 1 will be used to collect the initial state of the probe as input and the next perigee state as output; similarly, the high-precision dynamic model for Mars aerodynamic assisted orbit descent constructed in step 1 will be used to collect the initial state of the probe as input and the next apogee state as output.

[0072] The range of values ​​for the detector's initial state is determined using the detector's near-fire point altitude, far-fire point altitude, and the six roots of the orbit. h is defined as follows. p and h a Let i, Ω, w, and f be the near-fire altitude and far-fire altitude of the detector, respectively. Define i, Ω, w, and f as the detector's orbital inclination, right ascension of the ascending node, argument of the pericenter, and true anomaly, respectively. Then, the initial state range of the detector is:

[0073]

[0074] Among them, h p,min =8×10 4 m, h a,min =4.5×10 5 m, i min =89deg, Ω min =0deg, w min =0deg,f min =180deg represents the initial near-fire altitude, far-fire altitude, orbital inclination, right ascension of the ascending node, argument of the perihelion, and lower limit of the true perihelion angle, respectively. p,max =1.25×10 5 m, h a,max =1×10 6 m, i max =93deg, Ω max =360deg, w max =360deg, f max =180deg represents the initial near-fire point altitude, far-fire point altitude, orbital inclination, right ascension of the ascending node, argument of the pericenter, and upper limit of the true pericenter angle, respectively.

[0075] After determining the initial state range of the detector, samples are taken within this range according to a uniform distribution, and the sampling results are converted into a state represented by latitude, longitude, and altitude, defined as x0=[r0,θ0,φ0,V0,γ0,ψ0]. T Given the selected initial state of the detector, the state of the detector at the near-fire point and the state of the detector at the far-fire point can be obtained by using high-precision dynamic integration.

[0076]

[0077] Where f represents the constructed high-precision dynamic model for Mars aerodynamic-assisted orbit descent, and t p and t a These represent the time from the initial moment to the next near point and the time from the initial moment to the next far point, respectively.

[0078] The time t for the detector to reach the next near point of fire from the initial moment p and the time t from the initial moment to the next distant firing point a It is difficult to estimate precisely, so its orbital period is estimated by using the semi-major axis determined by the initial state of the probe.

[0079]

[0080] Where T is the orbital period of the detector under simple two-body dynamics, a0 is the semi-major axis of the orbit, and μm is the standard gravitational constant for Mars. After estimating the orbital period, a high-precision dynamic model for Mars aerodynamic-assisted orbit descent was used to integrate the probe's initial state for one revolution, and data on the next perihelion and next apohelion states corresponding to the initial state were collected.

[0081] Considering that under two-body dynamics, the next perihelion and apogee states can be analytically solved based on the probe's initial state, the mappings from the initial state to the next perihelion state and from the initial state to the next apogee state under Mars aerodynamic-assisted orbit descent high-precision dynamics can be divided into analytical quantities obtained from two-body dynamics and correction quantities caused by non-spherical gravitational perturbations and atmospheric drag. Definitions and To analyze the near-fire and far-fire states obtained using the simplified two-body dynamics model, the corrected near-fire state δx... p and the state of the far-field firing point δx a It can be represented as

[0082]

[0083] Based on the determined range of initial state values ​​for the detector, K = 2 × 10⁻⁶ is collected. 5 A dataset was constructed using samples, mapping the detector's initial state to near-fire point state corrections and far-fire point state corrections.

[0084]

[0085] Where X is the set of initial states of the detector, Y p and Y a These represent the sets of near-fire point state corrections and the sets of far-fire point state corrections, respectively.

[0086] Step 3: Construct a near-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to near-fire point state corrections for network training. Construct a far-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to far-fire point state corrections for network training. After training the near-fire point and far-fire point state correction prediction networks, correct the near-fire point and far-fire point states obtained from two-body dynamics to construct intelligent prediction models for Mars aerodynamically assisted descent near-fire point and far-fire point states. Compared to the method of directly constructing networks to predict near-fire point and far-fire point states, this method solves for the near-fire point and far-fire point states under the two-body dynamics model, constructs the network to predict the deviation between the near-fire point and far-fire point states under the high-precision dynamics model and the two-body model prediction results, reduces the training cost of the neural network, and improves the prediction accuracy of the neural network.

[0087] Constructing a near-fire state correction prediction network (DNN) p The dataset constructed in step 2, which maps the detector's initial state to near-fire point state corrections, is used for network training. A distant-fire point state correction prediction network (DNN) is then constructed. a The dataset constructed in step 2, which maps the initial state of the probe to the correction values ​​of the far-horse point state, is used for network training. After training the near-horse point state correction prediction network and the far-horse point state correction prediction network, the near-horse point state and the far-horse point state obtained from two-body dynamics are corrected to construct intelligent prediction models for the near-horse point state and the far-horse point state of Mars aerodynamically assisted descent orbit.

[0088]

[0089] Where, x k This indicates the state of the flexible lander at time k. and The near-fire point state and far-fire point state at time k+1 under two-body dynamics, respectively, x p,k+1 and x a,k+1 This is for the high-precision near-fire and far-fire state obtained through intelligent prediction. The constructed network has four hidden layers, each with 32 neurons, and the activation function is tanh. Figure 2 and Figure 3 The training results of the near-fire point state correction prediction network and the far-fire point state correction prediction network are shown respectively.

[0090] Step 5: Construct the near-fire point heat flow constraint equation and control amplitude constraint equation using the near-fire point state intelligent prediction model. At the same time, use the far-fire point state intelligent prediction model as the state prediction equation. With orbital altitude deviation and control input as optimization performance indicators, design a predictive guidance method. Use the current state of the probe as the input of the predictive guidance law, optimize the solution of the probe's control quantity at the far-fire point, and accurately lower the probe to the predetermined orbit under the conditions of satisfying heat flow and control amplitude constraints, while reducing the probe's fuel consumption. This is the Mars aerodynamic-assisted orbit lowering intelligent predictive guidance.

[0091] During the aerodynamically assisted descent to Mars, to protect the probe's safety, it is necessary to limit the heat flux density at the periapsis. This is based on the probe's apoapsis state x at time k+i. a,k+i and control input ΔV a,k+i The near-fire point heat flow constraint equation, constructed from the near-fire point state intelligent prediction model, is expressed as follows:

[0092]

[0093] Where, ρ p,k+i+1Let γ be the atmospheric density near the fire point at time k+i+1, and let γ = [0,0,0,1,0,0]. T , The maximum permissible heat flux density.

[0094] Meanwhile, the thrust that the probe can apply has an upper limit, and the velocity change at the far-field point is constrained.

[0095] |ΔV a,k+i |≤ΔV max (25)

[0096] Where, ΔV max =50m / s is the upper limit of the velocity change at the far-field firing point.

[0097] When the probe's initial position is at its apogee, the Mars aerodynamic assisted descent begins. The Mars aerodynamic assisted descent prediction guidance, at each time k, is based on the current apogee state x. a,k To predict the input of the guidance law, the control input ΔV of the far-fire point is obtained by solving an optimization problem. a,k The optimization problem to be solved uses orbital altitude deviation and control input as optimization performance indicators, and a remote firing point state intelligent prediction model as the state prediction equation, while also considering heat flux constraints and control amplitude constraints.

[0098]

[0099] Where N = 20 is the number of predicted laps, R m =3.397×10 6 The radius of Mars. The desired orbital altitude of the far-orbiting flank. The far-orbiting flank control input ΔV within the predicted orbital number. a,k+i After optimization, ΔV is selected. a,k As the control input at time k.

[0100] When the detector is at the far firing point, the detector is controlled by optimizing the control input. The optimization result is as follows: Figure 4 As shown, the probe's orbital altitude is lowered until it is near the predetermined orbital altitude, thus achieving precise descent of the probe to the predetermined orbit while meeting the constraints of heat flow and control amplitude, and reducing the probe's fuel consumption. This is known as Mars aerodynamic-assisted orbit descent intelligent predictive guidance.

[0101] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A Mars aerodynamic-assisted orbit descent intelligent predictive guidance method, characterized in that: Includes the following steps: Step 1: Construct a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Mars inertial frame. In the subsequent Step 2, the constructed high-precision dynamic model will be used to generate a dataset for neural network learning. Step 2: Use the high-precision dynamic model constructed in Step 1 to generate the dataset required for neural network training; during the dataset generation process, define the value range of the initial state, sample according to the uniform distribution, use the Mars aerodynamic-assisted descent high-precision dynamic model to recursively calculate the initial state to collect the near-fire point state and far-fire point state at the next moment, subtract the near-fire point state obtained by the high-precision dynamic model from the near-fire point state obtained by the two-body dynamic model to obtain the near-fire point state correction, and subtract the far-fire point state obtained by the high-precision dynamic model from the far-fire point state obtained by the two-body dynamic model to obtain the far-fire point state correction. The dataset for training the near-fire point state correction prediction network is formed by taking the detector's initial state as input and the near-fire point state correction as output. The dataset for training the far-fire point state correction prediction network is formed by taking the detector's initial state as input and the far-fire point state correction as output. Step 3: Construct a near-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to the near-fire point state correction value for network training; construct a far-fire point state correction prediction network, using the dataset constructed in Step 2 that maps the probe's initial state to the far-fire point state correction value for network training; after completing the training of the near-fire point state correction prediction networks and far-fire point state correction prediction networks, correct the near-fire point state and far-fire point state obtained from two-body dynamics to construct a Mars aerodynamically assisted descent near-fire point state intelligent prediction model and a far-fire point state intelligent prediction model; the method solves for the near-fire point and far-fire point states under the two-body dynamics model, constructs the network to predict the deviation between the near-fire point state and far-fire point state under the high-precision dynamics model and the two-body model prediction results, reduces the training cost of the neural network, and improves the prediction accuracy of the neural network. Where, x k This indicates the state of the flexible lander at time k. and The near-fire point state and far-fire point state at time k+1 under two-body dynamics, respectively, x p,k+1 and x a,k+1 This provides high-precision near-fire point and far-fire point status obtained through intelligent prediction. Step 4: Construct the near-fire point heat flow constraint equation and control amplitude constraint equation using the near-fire point state intelligent prediction model. At the same time, use the far-fire point state intelligent prediction model as the state prediction equation. With orbital altitude deviation and control input as optimization performance indicators, design a predictive guidance method. Use the current state of the probe as the predictive guidance law input, optimize the solution of the probe's control quantity at the far-fire point, and accurately lower the probe to the predetermined orbit under the conditions of satisfying heat flow and control amplitude constraints, while reducing the probe's fuel consumption. This is the Mars aerodynamic-assisted orbit lowering intelligent predictive guidance.

2. The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method as described in claim 1, characterized in that: In step 1, Constructing a high-precision dynamic model for Mars aerodynamic-assisted orbit descent in the Martian inertial frame. Where r is the distance of the probe from the center of Mars, θ and φ represent the latitude and longitude of the probe in the Martian inertial frame, respectively, V is the magnitude of the probe's velocity, γ and ψ represent the probe's track angle and heading angle, respectively, D and L represent the magnitudes of aerodynamic drag and aerodynamic lift acceleration, respectively, and g, l, and f represent the radial, longitude, and latitudinal components of the non-spherical gravity of Mars, respectively. The Martian full-gradient gravity model... The Martian atmospheric model is represented by the Martian atmospheric correction index model. Where h represents the orbital altitude of the probe; given the Martian atmospheric model, the aerodynamic drag and aerodynamic lift acceleration in the probe's dynamic equations are respectively expressed as: Among them, C D and C L ρ represents the drag coefficient and lift coefficient of the probe, respectively; S is the effective frontal area of ​​the probe; V is the velocity of the probe; m is the mass of the probe; and ρ is the atmospheric density determined by the Martian atmospheric model.

3. The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method as described in claim 2, characterized in that: Step 2 is implemented as follows: During the Mars aerodynamic-assisted descent process, control is mainly applied at the far-horizon, while the near-horizon state must meet thermal flow constraints. Therefore, when using neural networks, intelligent prediction models of the near-horizon state and the far-horizon state, predicted from the initial state, will be constructed. In step 2, the high-precision dynamic model of Mars aerodynamic-assisted descent constructed in step 1 will be used to collect the initial state of the probe and the output will be the dataset of the next near-horizon state. The high-precision dynamic model of Mars aerodynamic-assisted descent constructed in step 1 will be used to collect the initial state of the probe and the output will be the dataset of the next far-horizon state. The range of values ​​for the detector's initial state is determined using the detector's near-fire point altitude, far-fire point altitude, and the six roots of the orbit. h is defined as follows. p and h a Let i, Ω, w, and f be the near-fire altitude and far-fire altitude of the detector, respectively. Define i, Ω, w, and f as the detector's orbital inclination, right ascension of the ascending node, argument of the pericenter, and true anomaly, respectively. Then, the initial state range of the detector is: Among them, h p,min h a,min i min Ω min w min f min These are the initial near-fire altitude, far-fire altitude, orbital inclination, right ascension of the ascending node, argument of the perihelion, and lower bound of the true perihelion angle, h. p,max h a,max i max Ω max w max f max These are the detector's initial near-fire point altitude, far-fire point altitude, orbital inclination, right ascension of the ascending node, argument of the pericenter, and upper limit of the true pericenter angle, respectively. After determining the initial state range of the detector, samples are taken within this range according to a uniform distribution, and the sampling results are converted into a state represented by latitude, longitude, and altitude, defined as x0=[r0,θ0,φ0,V0,γ0,ψ0]. T Given the selected initial state of the detector, the state of the detector at the near-fire point and the state of the detector at the far-fire point can be obtained by using high-precision dynamic integration. Where f represents the constructed high-precision dynamic model for Mars aerodynamic-assisted orbit descent, and t p and t a These represent the time from the initial moment to the next near point and the time from the initial moment to the next far point, respectively. The time t for the detector to reach the next near point of fire from the initial moment p and the time t from the initial moment to the next distant firing point a It is difficult to estimate precisely, so its orbital period is estimated by using the semi-major axis determined by the initial state of the probe. Where T is the orbital period of the detector under simple two-body dynamics, a0 is the semi-major axis of the orbit, and μ m The standard gravitational constant of Mars is used; after estimating the orbital period, the high-precision dynamic model of Mars aerodynamic-assisted orbit descent is used to integrate the probe's initial state once, and data on the next perihelion state and the next apohelion state corresponding to the initial state are collected. Considering that under two-body dynamics, the next perihelion and apohelion states can be analytically solved based on the probe's initial state; under Mars aerodynamic-assisted orbit descent high-precision dynamics, the mapping from the initial state to the next perihelion state and the mapping from the initial state to the next apohelion state can be divided into analytical quantities obtained from two-body dynamics and correction quantities caused by non-spherical gravitational perturbations and atmospheric drag; [Definition] and To analyze the near-fire and far-fire states obtained using the simplified two-body dynamics model, the corrected near-fire state δx... p and the state of the far-field firing point δx a Represented as Based on the established range of detector initial state values, K samples are collected to construct a dataset that maps the detector initial state to near-fire point state corrections and far-fire point state corrections. Where X is the set of initial states of the detector, Y p and Y a These represent the sets of near-fire point state corrections and the sets of far-fire point state corrections, respectively.

4. The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method as described in claim 3, characterized in that: In step 3, Constructing a near-fire state correction prediction network (DNN) p The dataset constructed in step 2, which maps the detector's initial state to near-fire point state corrections, is used for network training; a distant-fire point state correction prediction network (DNN) is constructed. a The dataset constructed in step 2, which maps the initial state of the probe to the correction values ​​of the far-horse point state, is used for network training. After training the near-horse point state correction prediction network and the far-horse point state correction prediction network, the near-horse point state and the far-horse point state obtained from two-body dynamics are corrected to construct intelligent prediction models for the near-horse point state and the far-horse point state of Mars aerodynamically assisted descent orbit. .

5. The Mars aerodynamic-assisted orbit descent intelligent prediction guidance method as described in claim 4, characterized in that: Step 4 is implemented as follows: During the aerodynamically assisted descent to Mars, to protect the probe's safety, it is necessary to limit the heat flux density at the periapsis. This is based on the probe's apoapsis state x at time k+i. a,k+i and control input ΔV a,k+i The near-fire point heat flow constraint equation, constructed from the near-fire point state intelligent prediction model, is expressed as follows: Where, ρ p,k+i+1 Let γ be the atmospheric density near the fire point at time k+i+1, and let γ = [0,0,0,1,0,0]. T Q max The maximum permissible heat flux density; Meanwhile, the thrust that the detector can apply has an upper limit, and the velocity change at the far-point of fire is constrained by the control amplitude. |ΔV a,k+i |≤ΔV max (12) Where, ΔV max This represents the upper bound of the velocity change at the far-field firing point. When the probe's initial position is at the apogee, it begins aerodynamically assisted descent to Mars; the aerodynamically assisted descent prediction guidance is applied at every time k, based on the current apogee state x. a,k To predict the input of the guidance law, the control input ΔV of the far-fire point is obtained by solving an optimization problem. a,k The optimization problem to be solved uses orbital altitude deviation and control input as optimization performance indicators, and a remote firing point state intelligent prediction model as the state prediction equation, while also considering heat flux constraints and control amplitude constraints. Where N is the number of predicted laps, R m The radius of Mars. The desired orbital altitude of the far-fire point; the control input ΔV of the far-fire point within the predicted orbital number. a,k+i After optimization, ΔV is selected. a,k As the control input at time k; When the probe is at its farthest point from Mars, the probe is controlled by optimizing the control input until its orbital altitude is reduced to near the predetermined orbital altitude. This achieves precise descent of the probe to the predetermined orbit while meeting the constraints of heat flow and control amplitude, and at the same time reduces the probe's fuel consumption. This is known as Mars aerodynamic-assisted orbit descent intelligent prediction guidance.

Images