Time angle multi-constraint guidance method based on prediction correction
By introducing angle and time prediction correctors into the guidance method of unmanned aerial vehicles, an optimized guidance law is obtained, which solves the problem of increasing angle errors caused by time constraints in the prior art, and improves guidance accuracy and energy efficiency.
Patent Information
- Application Number
- CN202311233171.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-22
- Publication Date
- 2025-04-01
AI Technical Summary
When implementing time constraints, the existing unmanned aerial vehicle guidance methods may lead to an increase in the error of the angle constraints, and fail to effectively consider factors such as speed changes and altitude changes during missile flight, resulting in a decrease in guidance accuracy and an increase in energy loss.
The time angle multi-constrained guidance method based on prediction correction is adopted. By setting an angle predictor and angle corrector, a residual flight time predictor and a residual flight time corrector, the footing bias term and the time bias term are obtained, and these bias terms are expanded on the basis of the proportional guidance law to form an optimized guidance law.
It realizes energy use optimization under the premise of satisfying trajectory accuracy, improves the energy efficiency of flight missions, can accurately control the terminal state of the aircraft in real time, and maintains excellent performance under different environments and conditions, enhancing the stability and reliability of the system.
Smart Images
Figure CN119739176A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a time angle multi-constraint guidance method based on prediction correction, belonging to the technical field of guidance. Background Art
[0002] The simultaneous realization of time constraint and angle constraint guidance technology is the key technology to realize cooperative guidance of unmanned aerial vehicles. However, there is a certain coupling relationship between multiple constraint items during the flight process. When realizing the time constraint, the error of the angle constraint may increase. Therefore, the appropriate time angle constraint guidance law is usually selected according to different decoupling design methods.
[0003] However, the current multi-constraint analytical guidance laws are generally derived based on the constant velocity model, without considering factors such as speed changes and altitude changes during the missile's flight. The actual mission environment of the missile does not meet the constant velocity assumption, which may increase the energy loss of the analytical guidance law and reduce the guidance accuracy, thereby reducing the guidance performance.
[0004] Therefore, it is necessary to conduct in-depth research on the existing unmanned aerial vehicle guidance methods to solve the above problems. Summary of the invention
[0005] In order to overcome the above problems, the inventors have conducted in-depth research and provided a time angle multi-constraint guidance method based on prediction correction, comprising the following steps:
[0006] Set the angle predictor and angle corrector to constrain the angle of the aircraft and obtain the landing offset item;
[0007] Setting a remaining flight time predictor and a remaining flight time corrector to constrain the remaining flight time of the aircraft and obtain a time offset term;
[0008] On the basis of the proportional guidance law, the landing bias term and the time bias term are expanded to form an optimized guidance law, and the aircraft flies under the control of the optimized guidance law;
[0009] The angle predictor is used to predict the landing angle of the aircraft terminal according to the flight state, and the angle corrector is used to optimize the error between the expected landing angle and the predicted actual landing angle according to the flight environment state to obtain the landing bias item;
[0010] The remaining flight time predictor is used to predict the remaining flight time of the aircraft according to the flight status and the expected terminal angle, and the remaining flight time corrector is used to optimize the error between the predicted remaining flight time of the aircraft and the expected remaining flight time according to the flight environment status to obtain a time offset term.
[0011] In a preferred embodiment, the input of the terminal angle predictor is set to the flight state, and the output is set to the predicted terminal angle.
[0012] In a preferred embodiment, the terminal angle predictor adopts a residual neural network.
[0013] In a preferred embodiment, the flight state includes the aircraft speed v, the aircraft trajectory inclination angle θ, the lateral position x of the aircraft in the horizontal plane, and the longitudinal height y of the aircraft.
[0014] In a preferred embodiment, the flight environment status includes aircraft speed and remaining flight time.
[0015] In a preferred embodiment, the angle corrector is a reinforcement learning network, and the input of the angle corrector is set as:
[0016]
[0017] Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε a Indicates the terminal angle error.
[0018] In a preferred embodiment, the reward function of the angle corrector includes an angle constraint reward term, a landing point constraint reward term, and an energy consumption reward term.
[0019] In a preferred embodiment, the remaining flight time predictor adopts a residual neural network.
[0020] In a preferred embodiment, the residual flight time corrector is a reinforcement learning network, and the input of the angle corrector is set as:
[0021]
[0022] Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε t Indicates the remaining flight time error.
[0023] In a preferred embodiment, the reward function of the remaining flight time includes an angle constraint reward item, a landing point constraint reward item, an energy consumption reward item, and a flight altitude reward item.
[0024] The beneficial effects of the present invention include:
[0025] 1. In the unmanned aerial vehicle guidance problem, energy consumption is introduced as an optimization target in an unprecedented way. This not only meets the guidance accuracy requirements, but also significantly improves the energy efficiency of the flight mission. This method optimizes the energy use of the aircraft while meeting the trajectory accuracy.
[0026] 2. Considering the dynamic characteristics of the aircraft and the real-time flight state changes of the missile target, the terminal state of the aircraft can be precisely controlled in real time;
[0027] 3. Dynamically adjust the guidance strategy in real time during flight to optimize energy consumption based on the current state of the aircraft and target information, which makes the guidance process more flexible and efficient and can adapt to different flight conditions and target requirements;
[0028] 4. Ability to maintain excellent performance in different actual environments and flight conditions, thereby enhancing the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 A schematic flow chart of a time angle multi-constraint guidance method based on prediction correction according to a preferred embodiment of the present invention is shown;
[0030] Figure 2 A simulation diagram of the flight trajectory of the aircraft under the same expected terminal angle and different expected flight times in Example 1 is shown;
[0031] Figure 3 A simulation diagram of the remaining flight time of the aircraft under the same expected terminal angle and different expected flight times in Example 1 is shown;
[0032] Figure 4 A simulation diagram of the flight speed of the aircraft under the same expected terminal angle and different expected flight times in Example 1 is shown;
[0033] Figure 5 The simulation diagram of the guidance instruction under the same expected terminal angle and different expected flight time in Example 1 is shown;
[0034] Figure 6 The simulation diagram of the flight trajectory of the aircraft under different expected terminal angles and the same expected flight time in Example 1 is shown;
[0035] Figure 7 A simulation diagram of the remaining flight time of the aircraft under different expected terminal angles and the same expected flight time in Example 1 is shown;
[0036] Figure 8 A simulation diagram of the flight speed of the aircraft under different expected terminal angles and the same expected flight time in Example 1 is shown;
[0037] Fig. 9 A simulation diagram of guidance instructions under different expected terminal angles and the same expected flight time in Example 1 is shown. DETAILED DESCRIPTION
[0038] The present invention will be further described in detail below through the accompanying drawings and embodiments. Through these descriptions, the characteristics and advantages of the present invention will become more clear and distinct.
[0039] The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise noted.
[0040] According to the present invention, a time angle multi-constraint guidance method based on prediction correction is provided. Figure 1 As shown, the following steps are included:
[0041] Set the angle predictor and angle corrector to constrain the angle of the aircraft and obtain the landing offset item;
[0042] Setting a remaining flight time predictor and a remaining flight time corrector to constrain the remaining flight time of the aircraft and obtain a time offset term;
[0043] On the basis of the proportional guidance law, the landing bias term and the time bias term are expanded to form an optimized guidance law, and the aircraft flies under the control of the optimized guidance law.
[0044] The angle predictor is used to predict the landing angle of the aircraft terminal according to the flight state, and the angle corrector is used to optimize the error between the expected landing angle and the predicted actual landing angle according to the flight environment state to obtain the landing bias item;
[0045] The remaining flight time predictor is used to predict the remaining flight time of the aircraft according to the flight status and the expected terminal angle, and the remaining flight time corrector is used to optimize the error between the predicted remaining flight time of the aircraft and the expected remaining flight time according to the flight environment status to obtain a time offset term.
[0046] Furthermore, the optimized guidance law a M It is expressed as:
[0047]
[0048] in, is the proportional guidance law, N is the guidance coefficient, v is the aircraft speed, λ is the sight angle between the missile and the target, and b IA represents the footing bias term, b IT Represents the time offset term.
[0049] In the present invention, the foothold bias term is used to constrain the angle of the aircraft, and the angle constraint is a unilateral decoupling. The time bias term is used to constrain the remaining flight time of the aircraft, which is a dual-parameter decoupling involving the angle of the aircraft. In the present invention, an inner and outer double-layer nested structure is set to obtain the foothold bias term and the time bias term, wherein the inner layer decomposes the design of the extended angle constraint term into an angle predictor design based on a residual neural network (ResNet) and an angle corrector design based on reinforcement learning, and the outer layer, on the basis of the angle constraint, decomposes the design of the extended time constraint term into a remaining flight time predictor design based on ResNet and a remaining flight time corrector design based on reinforcement learning.
[0050] Preferably, the reinforcement learning is a proximal optimization strategy (PPO). PPO has good adaptability and robustness to noise, uncertainty and changing flight conditions. This advantage enables the guidance strategy to maintain excellent performance in different actual environments and flight conditions, thereby enhancing the stability and reliability of the system.
[0051] In the inner structure, the input of the terminal angle predictor is set as the flight state, and the flight state includes the aircraft speed v, the aircraft trajectory inclination angle θ, the lateral position x of the aircraft in the horizontal plane, and the longitudinal height y of the aircraft;
[0052] The output is set to the predicted terminal angle.
[0053] In a preferred embodiment, the terminal angle predictor is a residual neural network.
[0054] When the flight state vector of the aircraft and the relative motion relationship between the missile and the target are determined, the flight trajectory of the aircraft starting from this state is unique, expressed as λ f =f(v,θ,x,y)
[0055] Among them, f() represents the mapping relationship, λ f Represents the predicted terminal angle.
[0056] Therefore, based on the above uniqueness, sufficient training samples can be obtained through the simulation environment to train the residual neural network, so that the residual neural network can achieve high-precision prediction.
[0057] In a preferred embodiment, the residual neural network is constructed using a fully connected neural network with three hidden layers, and residual blocks are arranged between the hidden layers for jump connections. By introducing residual connections, deep networks can be trained more effectively, avoiding the gradient vanishing problem.
[0058] More preferably, the number of hidden layers of the residual block is 10, and the number of neurons in each hidden layer is set to 100.
[0059] Further preferably, in the residual neural network, ReLU is used as the activation function of the input layer and the hidden layer, and no activation function is set for the output layer.
[0060] According to the present invention, after the terminal angle of the aircraft is obtained by a terminal angle predictor, the error between the desired landing angle and the predicted landing angle is optimized by setting a terminal angle corrector.
[0061] Furthermore, the flight environment status includes aircraft speed and remaining flight time.
[0062] In a preferred embodiment, the angle corrector is a reinforcement learning network, and the input of the angle corrector is set as:
[0063]
[0064] Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε a Indicates the terminal angle error.
[0065] Furthermore, the terminal angle error is:
[0066] ε a =θ d -θ f
[0067] Among them, θ d represents the desired terminal location, θ f Indicates the predicted terminal location.
[0068] According to the present invention, the output of the angle corrector is set to the guidance law formed by expanding the foothold bias term of the proportional guidance law, which is expressed as:
[0069]
[0070] in, is the proportional guidance law, b IA is the footing offset term.
[0071] In the Proximal Optimization (PPO) strategy, the reward indicates the goal of instruction optimization. Reasonable rewards can ensure the convergence of the learning process and improve learning efficiency.
[0072] In a preferred embodiment, the reward function of the angle corrector includes an angle constraint reward term, a landing point constraint reward term and an energy consumption reward term.
[0073] Among them, the reward term r1 of the angle constraint is set as:
[0074]
[0075] Among them, σ1 is the scaling factor of the drop angle constraint reward term.
[0076] In the present invention, on the basis that the proportional guidance law has been able to make the landing point constraint converge, the amplitude of the bias term can be limited by setting the landing point constraint reward term to avoid missing the target.
[0077] The landing point constraint reward item r2 is set to:
[0078]
[0079]
[0080] Among them, ε ZEM represents the zero-control miss distance, and σ2 represents the scaling factor of the landing point constraint reward item.
[0081] The energy consumption reward item r3 is set as:
[0082]
[0083]
[0084] Among them, σ3 is the scaling factor of the energy consumption bonus term, Indicates the remaining energy of the aircraft.
[0085] According to the present invention, the setting of the energy consumption reward item takes into account the mutual conversion relationship between gravitational potential energy and kinetic energy, and can maximize the remaining energy.
[0086] Furthermore, the reward function r of the terminal angle corrector is set as the weighted average of the angle constraint reward item, the landing point constraint reward item and the energy consumption reward item, expressed as:
[0087]
[0088] Among them, b1, b2, and b3 are weight coefficients, satisfying b1+b2+b3=1.
[0089] According to the present invention, during the process of training the angle predictor and the angle corrector, the aircraft state and the flight environment state are obtained through a simulation environment.
[0090] In the outer structure, the input of the remaining flight time predictor is set to the flight state and the expected terminal angle, wherein the flight state includes the aircraft speed v, the aircraft trajectory inclination angle θ, the lateral position x of the aircraft in the horizontal plane, and the longitudinal height y of the aircraft;
[0091] The output is set to the predicted remaining flight time, denoted as t go=f(v,θ,λ d ,x,y)
[0092] Among them, λ d Indicates the desired terminal angle.
[0093] Similar to the terminal angle predictor, the remaining flight time predictor adopts a residual neural network. In the remaining time prediction, the use of the residual network can better capture the nonlinear characteristics of time changes, thereby improving the accuracy of the terminal angle and remaining flight time prediction.
[0094] In a preferred embodiment, the number of hidden layers of the remaining flight time predictor is 10, and the number of neurons in each hidden layer is set to 100.
[0095] Further preferably, ReLU is used as the activation function of the input layer and the hidden layer, and no activation function is set for the output layer.
[0096] In a preferred embodiment, the residual flight time corrector is a reinforcement learning network, and the input of the residual flight time corrector is set as:
[0097]
[0098] Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε t Indicates the remaining flight time error.
[0099] Further, the remaining flight time error is:
[0100] ε t =t d -(t+t go )
[0101] Among them, t d represents the expected flight time, t represents the current flight time, and t go Indicates the remaining flight time.
[0102] According to the invention, the output of the residual flight time corrector is arranged to optimize the guidance law.
[0103] In a preferred embodiment, the reward function of the remaining flight time corrector includes an angle constraint reward item, a landing point constraint reward item, an energy consumption reward item and a flight altitude reward item, wherein the angle constraint reward item, the landing point constraint reward item and the energy consumption reward item are the same as those in the angle corrector.
[0104] The flight altitude reward item r4 is set as:
[0105]
[0106] Among them, R is the relative distance between the projectile and the target, and σ4 is the scaling factor of the energy consumption bonus item.
[0107] The principle of the time-constrained guidance law to extend the flight time is to increase the curvature of the trajectory to extend the remaining trajectory length. However, under the same initial conditions, there is no unique trajectory that can meet the time constraint. Some trajectories may cause the missile to fall before hitting the target. In the present invention, by setting a flight altitude reward item, the aircraft can maintain a certain altitude when extending the trajectory.
[0108] In a preferred embodiment, the reward function of the remaining flight time corrector also includes an inner layer reward term, and by setting the inner layer reward term, the outer layer remaining flight time corrector further optimizes the terminal angle error ε in the interactive optimization learning stage. a , improving the control accuracy under the coupling of two constraints: angle and remaining flight time.
[0109] The inner layer reward item is expressed as:
[0110]
[0111] Among them, σ5 is the scaling factor of the inner layer reward term.
[0112] In a preferred embodiment, the reward function r′ of the remaining flight time corrector is set as the weighted average of the angle constraint reward item, the landing point constraint reward item, the energy consumption reward item, the flight height reward item and the inner layer reward item, expressed as:
[0113]
[0114] Among them, b1′, b2′, b3′, b4′, and b5′ are weight coefficients, satisfying b1′+b2′+b3′+b4′+b5′=1.
[0115] According to the present invention, in the process of training the remaining flight time predictor and the remaining flight time corrector, the aircraft state and the flight environment state are obtained through a simulation environment.
[0116] In the present invention, the training process of the angle corrector and the residual flight time corrector is not described in detail, and those skilled in the art can perform the training process based on actual experience.
[0117] According to the present invention, the trained angle predictor, angle corrector, remaining flight time predictor and remaining flight time corrector can output extended landing bias items and time bias items after obtaining the actual flight state and flight environment state, and then obtain the optimized guidance law based on the proportional guidance law, and control the flight of the aircraft by optimizing the guidance law.
[0118] Example
[0119] Example 1
[0120] The simulation experiment was conducted, and the initial state was set as: x0 = -20 km, y0 = 10 km, v0 = 600 m / s, θ0 = 0°. The multi-constraint guidance based on the predicted and corrected time angle included the following steps:
[0121] Set the angle predictor and angle corrector to constrain the angle of the aircraft and obtain the landing offset item;
[0122] Setting a remaining flight time predictor and a remaining flight time corrector to constrain the remaining flight time of the aircraft and obtain a time offset term;
[0123] On the basis of the proportional guidance law, the landing bias term and the time bias term are expanded to form an optimized guidance law, and the aircraft flies under the control of the optimized guidance law.
[0124] An inner and outer double-layer nested structure is set up to obtain the foothold bias term and the time bias term. The inner layer decomposes the design of the extended angle constraint term into the design of an angle predictor based on a residual neural network and the design of an angle corrector based on reinforcement learning. The outer layer, based on the angle constraint, decomposes the design of the extended time constraint term into the design of a residual flight time predictor based on a residual neural network and the design of a residual flight time corrector based on reinforcement learning.
[0125] In the inner structure, the input of the terminal angle predictor is set to the flight state, which includes the aircraft speed v, the aircraft ballistic inclination angle θ, the aircraft's lateral position x in the horizontal plane, and the aircraft's longitudinal height y; the output is set to the predicted terminal angle.
[0126] The residual neural network is constructed by a fully connected neural network with three hidden layers. Residual blocks are set between the hidden layers for jump connection. The number of hidden layers of the residual block is 10, and the number of neurons in each hidden layer is set to 100. In the residual neural network, ReLU is used as the activation function of the input layer and the hidden layer, and no activation function is set for the output layer.
[0127] The angle corrector is a reinforcement learning network, and the input of the angle corrector is set as:
[0128]
[0129] ε a =θ d -θ f
[0130] ε a =θ d -θ f
[0131] The reward function of the terminal angle corrector is set as:
[0132]
[0133] In the outer structure, the input of the remaining flight time predictor is set to the flight state and the expected terminal angle, the flight state includes the aircraft speed v, the aircraft ballistic inclination angle θ, the aircraft's lateral position x in the horizontal plane, and the aircraft's longitudinal height y; the output is set to the predicted remaining flight time.
[0134] The remaining flight time predictor adopts the same residual neural network as the terminal angle predictor.
[0135] The remaining flight time corrector is a reinforcement learning network, and the input of the remaining flight time corrector is set as:
[0136]
[0137] ε t =t d -(t+t go )
[0138] The output of the residual flight time corrector is set to the optimized guidance law, expressed as:
[0139]
[0140] The reward function of the remaining flight time corrector is set as:
[0141]
[0142] Among them, different expected flight times and the same expected terminal angle are set, and the expected flight time is set to t d =60s,70s,80s, the expected terminal angle is set to λ d =-90°, the simulation results are as follows Figure 2-5 As shown, Figure 2 The simulation diagram of the aircraft flight trajectory is shown. Figure 3 The remaining flight time simulation diagram of the aircraft is shown. Figure 4 The figure shows the aircraft flight speed simulation diagram. Figure 5 A guidance instruction simulation diagram is shown.
[0143] Set different expected terminal angles and the same expected flight time, and set the expected terminal angles to λ d =-45°,-60°,-90°, the expected flight time is set to t d = 80s, the simulation results are as follows Figure 6-9 As shown, Figure 6 The simulation diagram of the aircraft flight trajectory is shown. Figure 7 The remaining flight time simulation diagram of the aircraft is shown. Figure 8 The figure shows the aircraft flight speed simulation diagram. Fig. 9 A guidance instruction simulation diagram is shown.
[0144] from Figure 2-Figure 9 It can be seen that the above method can simultaneously meet different flight time constraints and terminal angle constraints.
[0145] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front", "rear", etc. indicate positions or positional relationships based on the working state of the present invention, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", and "fourth" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0146] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0147] The present invention has been described above in conjunction with preferred embodiments, but these embodiments are only exemplary and serve only as an illustration. On this basis, the present invention may be subjected to a variety of substitutions and improvements, all of which fall within the scope of protection of the present invention.
Claims
1. A time angle multi-constraint guidance method based on prediction and correction, characterized in that: The following steps are involved: Set the angle predictor and angle corrector to constrain the angle of the aircraft and obtain the landing offset item; Setting a remaining flight time predictor and a remaining flight time corrector to constrain the remaining flight time of the aircraft and obtain a time offset term; On the basis of the proportional guidance law, the landing bias term and the time bias term are expanded to form an optimized guidance law, and the aircraft flies under the control of the optimized guidance law; The angle predictor is used to predict the landing angle of the aircraft terminal according to the flight state, and the angle corrector is used to optimize the error between the expected landing angle and the predicted actual landing angle according to the flight environment state to obtain the landing bias item; The remaining flight time predictor is used to predict the remaining flight time of the aircraft according to the flight status and the expected terminal angle, and the remaining flight time corrector is used to optimize the error between the predicted remaining flight time of the aircraft and the expected remaining flight time according to the flight environment status to obtain a time offset term.
2. The time angle multi-constraint guidance method based on prediction correction according to claim 1 is characterized in that: The input of the terminal angle predictor is set to the flight state, and the output is set to the predicted terminal angle.
3. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The terminal angle predictor adopts a residual neural network.
4. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The flight state includes the aircraft speed v, the aircraft trajectory inclination angle θ, the lateral position x of the aircraft in the horizontal plane, and the longitudinal height y of the aircraft.
5. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The flight environment status includes aircraft speed and remaining flight time.
6. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The angle corrector is a reinforcement learning network, and the input of the angle corrector is set as: Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε a Indicates the terminal angle error.
7. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The reward function of the angle corrector includes an angle constraint reward term, a landing point constraint reward term, and an energy consumption reward term.
8. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The remaining flight time predictor adopts a residual neural network.
9. The time angle multi-constraint guidance method based on prediction and correction according to claim 1 is characterized in that: The remaining flight time corrector is a reinforcement learning network, and the input of the angle corrector is set as: Among them, s represents the flight environment state, v represents the aircraft speed, t go represents the remaining flight time, ε t Indicates the remaining flight time error.
10. The time angle multi-constraint guidance method based on prediction and correction according to claim 1, characterized in that: The reward function for the remaining flight time includes angle constraint reward, landing point constraint reward, energy consumption reward and flight altitude reward.