A method for rapid maneuvering of aircraft based on online learning of neural networks

Abstract

The present invention discloses a method for rapid maneuvering of an aircraft based on online learning of a neural network, and relates to the technical field of control of hypersonic aircraft. Unknown system uncertainties are processed by a neural network based on online learning of historical data, and derivatives of virtual control and some necessary information are estimated by a fixed-time filter. Based on the information estimated by the neural network and a fixed-time backstepping controller, the fixed-time tracking problem of the hypersonic aircraft is solved. Since the neural network has a self-learning function and parallel computing capabilities, the optimal control effect can be quickly obtained. Therefore, the fixed-time backstepping controller of the present invention can ensure that the hypersonic aircraft control system with uncertainty stabilizes the system attitude and speed to the desired state within a fixed time.

Description

Technical Field

[0001] The present invention relates to the technical field of hypersonic aircraft control, and in particular to an aircraft rapid maneuvering method based on neural network online learning. Background Art

[0002] Hypersonic vehicles have the advantages of amazing flight speed, strong penetration ability and wide flight envelope. Designing the flight control system of hypersonic vehicles is crucial to accomplish many complex missions, such as high-speed transportation, near-space access and reconnaissance.

[0003] However, the dynamics of hypersonic vehicles exhibit highly nonlinear characteristics with strong coupling between propulsion and aerodynamic effects, which poses challenges to flight control design.

[0004] In order to deal with the uncertainty of hypersonic aircraft control systems, some professionals have proposed to provide robust tracking control of speed and altitude reference trajectories in the presence of system uncertainty through backstepping design and virtual control. However, the current method of dealing with the uncertainty of hypersonic aircraft systems through backstepping design and virtual control has many defects, such as the explosion of the complexity of backstepping design, the limitation of finite time convergence, and the need to assume that the derivative of the virtual control law is bounded.

[0005] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0006] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a method for rapid maneuvering of an aircraft based on online learning of a neural network is provided, aiming to solve the problem in the prior art that the uncertainty of a hypersonic aircraft system is handled by backstepping design and virtual control, and it is difficult to achieve convergence of the system within a specific limited time.

[0007] The technical solution adopted by the present invention to solve the problem is as follows:

[0008] In a first aspect, an embodiment of the present invention provides a method for rapid maneuvering of an aircraft based on online learning of a neural network, the method comprising:

[0009] Decompose the dynamic model of the target aircraft in advance to obtain the velocity subsystem and attitude subsystem;

[0010] For the attitude subsystem, an estimated value of the system uncertainty corresponding to the attitude subsystem is obtained through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; an estimated value of the derivative of the target virtual control amount is obtained through a preset fixed time filter according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle to be achieved; and control information of the elevator deflection angle is output through a preset first fixed time backstepping controller according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and the estimated value of the system uncertainty of the attitude subsystem;

[0011] For the speed subsystem, an estimated value of the system uncertainty corresponding to the speed subsystem is obtained through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and control information of the fuel equivalent ratio is output according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller.

[0012] In one embodiment, the system state corresponding to the dynamic model includes speed, flight altitude, flight path angle, angle of attack and pitch rate, and the control input information includes elevator deflection angle and fuel equivalent ratio; after decomposition, the system state corresponding to the speed subsystem includes speed, and the system state corresponding to the attitude subsystem includes flight altitude, flight path angle, angle of attack and pitch rate.

[0013] In one embodiment, the step of obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed time filter; and outputting control information of an elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed time backstepping controller, comprises:

[0014] Obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers;

[0015] generating a target continuous function according to the target virtual control amount and the pitch rate;

[0016] The control information of the elevator deflection angle is outputted through the first fixed-time backstepping controller according to the estimated value of the derivative of the target virtual control amount, the target continuous function and the estimated value of the system uncertainty of the attitude subsystem.

[0017] In one embodiment, obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers includes:

[0018] generating a first continuous function according to the flight path angle and the target flight path angle;

[0019] Outputting an estimated value of a derivative of the first continuous function according to the first continuous function through a preset first fixed time filter;

[0020] Outputting a first virtual control amount according to the first continuous function and an estimated value of a derivative of the first continuous function through a preset first virtual controller;

[0021] generating a second continuous function according to the first virtual control amount, the angle of attack, and the flight path angle;

[0022] Outputting an estimated value of a derivative of the second continuous function according to the second continuous function through a preset second fixed time filter;

[0023] Outputting the target virtual control amount according to the second continuous function and an estimated value of a derivative of the second continuous function through a preset second virtual controller;

[0024] An estimated value of a derivative of the target virtual controlled amount is output according to the target virtual controlled amount through a preset third fixed time filter.

[0025] In one embodiment, obtaining an estimated value of system uncertainty corresponding to the posture subsystem through a preset first neural network includes:

[0026] Calculating a first prediction error using a predictor of the first neural network and historical data of the posture subsystem at a previous time;

[0027] outputting an estimated value of a first optimal neural network weight vector according to the first prediction error by an update rate of the first neural network;

[0028] An estimated value of the system uncertainty corresponding to the posture subsystem is output according to the estimated value of the first optimal neural network weight vector through the neural network estimator of the first neural network.

[0029] In one embodiment, the outputting of the control information of the fuel equivalent ratio by the preset second fixed-time backstepping controller according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem includes:

[0030] Calculating a speed difference according to the speed of the target aircraft and the target speed;

[0031] The control information of the fuel equivalent ratio is outputted through the second fixed-time backstepping controller according to the speed difference and the estimated value of the system uncertainty corresponding to the speed subsystem.

[0032] In one embodiment, obtaining an estimated value of system uncertainty corresponding to the speed subsystem through a preset second neural network includes:

[0033] Calculating a second prediction error using the predictor of the second neural network and the historical data of the speed subsystem at a previous time;

[0034] outputting an estimated value of a second optimal neural network weight vector according to the second prediction error by an update rate of the second neural network;

[0035] An estimated value of the system uncertainty corresponding to the speed subsystem is output according to the estimated value of the second optimal neural network weight vector through the neural network estimator of the second neural network.

[0036] In a second aspect, an embodiment of the present invention further provides an aircraft rapid maneuvering device based on neural network online learning, the device comprising:

[0037] A pre-decomposition module is used to pre-decompose the dynamic model of the target aircraft to obtain a velocity subsystem and an attitude subsystem;

[0038] an elevator deflection angle control module, for obtaining an estimated value of system uncertainty corresponding to the attitude subsystem through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed-time filter; and outputting control information of the elevator deflection angle according to the target virtual control amount, an estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed-time backstepping controller;

[0039] A fuel equivalent ratio control module is used to obtain an estimated value of the system uncertainty corresponding to the speed subsystem through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and output control information of the fuel equivalent ratio according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller.

[0040] In a third aspect, an embodiment of the present invention further provides a terminal, comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing any of the above-described methods for rapid maneuvering of an aircraft based on online learning of a neural network; and the processor is used to execute the program.

[0041] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a plurality of instructions stored thereon, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the steps of the above-mentioned method for rapid maneuvering of an aircraft based on online learning of a neural network.

[0042] Beneficial effects of the present invention: The embodiment of the present invention processes unknown system uncertainties through a neural network based on online learning of historical data. And the derivatives of virtual control and some necessary information are estimated through a fixed time filter. Based on the information estimated by the neural network and the fixed time backstepping controller, the fixed time tracking problem of the hypersonic aircraft is solved. Since the neural network has a self-learning function and parallel computing ability, the optimal control effect can be quickly obtained. Therefore, the fixed time backstepping controller of the present invention can ensure that the control system of the hypersonic aircraft with uncertainty stabilizes the system attitude and speed to the desired state within a fixed time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 It is a flowchart of an aircraft rapid maneuvering method based on neural network online learning provided by an embodiment of the present invention.

[0045] Figure 2 It is a control scheme block diagram of an aircraft rapid maneuvering method based on neural network online learning provided by an embodiment of the present invention.

[0046] Figure 3 It is a tracking performance comparison curve chart of a hypersonic aircraft provided by an embodiment of the present invention.

[0047] Figure 4 is an estimation curve diagram of the uncertainty and the derivative of the virtual control law provided by the embodiment of the present invention.

[0048] Figure 5 It is a time response comparison curve diagram of the controller control input provided by an embodiment of the present invention.

[0049] Figure 6 It is a time response comparison curve diagram of the neural network weight update rate provided by an embodiment of the present invention.

[0050] Figure 7 It is a schematic diagram of the internal modules of an aircraft rapid maneuvering device based on neural network online learning provided by an embodiment of the present invention.

[0051] Figure 8 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention discloses a method for rapid maneuvering of an aircraft based on online learning of a neural network. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0054] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0055] In view of the above-mentioned defects of the prior art, the present invention provides a method for rapid maneuvering of an aircraft based on online learning of a neural network, which processes unknown system uncertainties through a neural network based on online learning of historical data. And the derivative of the virtual control and some necessary information are estimated through a fixed-time filter. Based on the information estimated by the neural network and the fixed-time backstepping controller, the fixed-time tracking problem of the hypersonic aircraft is solved. Since the neural network has a self-learning function and parallel computing capabilities, the optimal control effect can be quickly obtained. Therefore, the fixed-time backstepping controller of the present invention can ensure that the control system of a hypersonic aircraft with uncertainty stabilizes the system attitude and speed to the desired state within a fixed time.

[0056] like Figure 1 As shown, the method specifically includes:

[0057] Step S100: Decompose the dynamic model of the target aircraft in advance to obtain a velocity subsystem and an attitude subsystem.

[0058] Specifically, the control object of this embodiment is the target aircraft, and the type of the target aircraft includes but is not limited to a hypersonic aircraft. The goal of this embodiment is to control the attitude (which may also be the height) and speed of the target aircraft. In practical application, firstly, a dynamic model of the target aircraft is established in advance, for example, a control guidance model of the longitudinal dynamics of the target aircraft is established. Then, the dynamic model of the target aircraft is decomposed into two subsystems, namely, a speed subsystem and an attitude subsystem (which may also be an altitude subsystem).

[0059] In one implementation, the system state corresponding to the dynamic model includes speed, flight altitude, flight path angle, angle of attack, and pitch rate, and the control input information includes elevator deflection angle and fuel equivalent ratio; after decomposition, the system state corresponding to the speed subsystem includes speed, and the system state corresponding to the attitude subsystem includes flight altitude, flight path angle, angle of attack, and pitch rate.

[0060] Specifically, for the dynamic model of the target aircraft, the system state may include the speed , Flight altitude , flight path angle , Angle of Attack and pitch rate , the control input is the elevator deflection Fuel Equivalence Ratio After decomposing the dynamic model, Figure 2 As shown, the system state corresponding to the speed subsystem includes speed , the system state corresponding to the attitude subsystem includes the flight altitude , flight path angle , Angle of Attack and pitch rate .

[0061] For example, taking a hypersonic aircraft as the target aircraft, the longitudinal dynamics model of the hypersonic aircraft is:

[0062] ;

[0063] in, , and Represents gravitational acceleration, mass and moment of inertia respectively; thrust ,resistance , lift and pitch moment It is expressed as:

[0064] ;

[0065] in, ; ; ; ; ; ; represents thrust fitting parameters; express The i-th order coefficient contributing to the resistance; express The i-th order coefficient contributing to lift; express The i-th order coefficient contributing to the torque; express Contribution coefficient to torque; represents the thrust-torque coupling coefficient; represents the reference area; It indicates the nominal air density under the approximate value of air density; Indicates the nominal altitude at the approximate value of air density; Indicates the air density decay rate; represents the mean aerodynamic chord; and Both represent thrust coefficient; They represent constant terms respectively.

[0066] Decompose the dynamic model of hypersonic vehicle into velocity subsystem and attitude subsystem;

[0067] The speed subsystem is:

[0068] ;

[0069] in, , ;

[0070] After further arrangement, the speed subsystem is finally:

[0071] ;

[0072] in, .

[0073] definition , , , , represents transpose, , (pitch angle), the attitude subsystem is:

[0074] ;

[0075] in, , , , ;

[0076] After further arrangement, the attitude subsystem is finally as follows:

[0077] ;

[0078] in, , .

[0079] Step S200: for the attitude subsystem, obtaining an estimated value of the system uncertainty corresponding to the attitude subsystem through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; obtaining an estimated value of the derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack and the desired target flight path angle through a preset fixed-time filter; and outputting control information of the elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate and the estimated value of the system uncertainty of the attitude subsystem through a preset first fixed-time backstepping controller.

[0080] Specifically, before actual application, this embodiment will pre-construct a fixed-time filter and a neural network for estimating the derivative and system uncertainty of the virtual control amount. And according to the constructed fixed-time filter and neural network, a fixed-time dynamic surface controller of the attitude subsystem of the target aircraft is obtained through a backstepping design method, that is, a first fixed-time backstepping controller. In actual application, the first fixed-time backstepping controller controls the elevator deflection angle of the target aircraft through the estimated value of the target virtual control amount and its derivative, the pitch rate, and the system uncertainty of the system, thereby stabilizing the system to a desired state within a fixed time.

[0081] For example, the fixed time filter is designed as:

[0082] ;

[0083] in, , , is the control parameter of the design; a continuous function represents the input of the fixed-time filter, is the output of the fixed-time filter; yes The estimated value of represents the state variable of the fixed-time filter, ; is a function that ensures the continuity of the system. Defined as:

[0084] ;

[0085] in, ; ; ; ; , represents the error of the fixed-time filter, Represents a constant whose value is greater than 0 (which can be used as the dividing point of the corresponding piecewise function).

[0086] In one implementation, the step of obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed-time filter; and outputting control information of an elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed-time backstepping controller, includes:

[0087] Obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers;

[0088] generating a target continuous function according to the target virtual control amount and the pitch rate;

[0089] The control information of the elevator deflection angle is outputted through the first fixed-time backstepping controller according to the estimated value of the derivative of the target virtual control amount, the target continuous function and the estimated value of the system uncertainty of the attitude subsystem.

[0090] Specifically, Figure 2 As shown, this embodiment adopts an alternate cascade method to set the input-output relationship of multiple fixed-time filters and virtual controllers, that is, the first to receive and process data is the fixed-time filter, followed by the virtual controller, and then the fixed-time filter, and so on. Through the alternate cascade of multiple fixed-time filters and virtual controllers, the pitch rate, flight path angle, angle of attack of the target aircraft and the target flight path angle to be achieved are used to estimate the derivative of the target virtual control amount. Finally, the target virtual control amount and the pitch rate are used to calculate the target continuous function as the input of the first fixed-time backstepping controller, and the estimated value of the derivative of the target virtual control amount and the estimated value of the system uncertainty of the attitude subsystem are also used as the input of the first fixed-time backstepping controller, and the first fixed-time backstepping controller calculates and outputs the control information of the elevator deflection angle according to the input data.

[0091] In one implementation, obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers includes:

[0092] generating a first continuous function according to the flight path angle and the target flight path angle;

[0093] Outputting an estimated value of a derivative of the first continuous function according to the first continuous function through a preset first fixed time filter;

[0094] Outputting a first virtual control amount according to the first continuous function and an estimated value of a derivative of the first continuous function through a preset first virtual controller;

[0095] generating a second continuous function according to the first virtual control amount, the angle of attack, and the flight path angle;

[0096] Outputting an estimated value of a derivative of the second continuous function according to the second continuous function through a preset second fixed time filter;

[0097] Outputting the target virtual control amount according to the second continuous function and an estimated value of a derivative of the second continuous function through a preset second virtual controller;

[0098] An estimated value of a derivative of the target virtual controlled amount is output according to the target virtual controlled amount through a preset third fixed time filter.

[0099] Specifically, for the attitude subsystem, this embodiment illustrates the design process of each fixed time filter and virtual controller, so as to explain and / or verify the use process of each fixed time filter and virtual controller:

[0100] The first step is to design the first fixed time filter as:

[0101] ;

[0102] Among them, the continuous function represents the input of the first fixed-time filter, is the output of the first fixed-time filter; yes An estimated value of represents the state variable of the first fixed-time filter; , represents the error of the first fixed time filter; , , is the design parameter; It is a function corresponding to the first fixed time filter to ensure the continuity of the system.

[0103] The first virtual controller is:

[0104] ;

[0105] in, represents the output of the first virtual controller; , , is a positive control parameter, ; is a continuous piecewise function; The definition of can be found in the definition section of the attitude subsystem mentioned above.

[0106] Designing continuous piecewise functions for:

[0107] ;

[0108] in, ; ; ; ; , represents a constant with a value greater than 0 (which can be used as the dividing point of the corresponding piecewise function); An exponential expression representing the input to the fixed-time filter.

[0109] The second step is to design the second fixed time filter as:

[0110] ;

[0111] Among them, the continuous function represents the input of the second fixed-time filter, is the output of the second fixed-time filter; yes An estimated value of represents a state variable of a second fixed-time filter; , represents the error of the second fixed time filter; , , is the design parameter; It is a function corresponding to the second fixed time filter to ensure the continuity of the system.

[0112] The second virtual controller is:

[0113] ;

[0114] in, represents the output of the second virtual controller; , , is a positive design parameter, ; Design continuous piecewise functions for:

[0115] ;

[0116] in, ; ; ; ; , represents a constant with a value greater than 0 (which can be used as the dividing point of the corresponding piecewise function); An exponential expression representing the input to the fixed-time filter.

[0117] The third step is to design the third fixed time filter as:

[0118] ;

[0119] in, is the output of the third fixed-time filter; , represents the error of the third fixed-time filter, yes An estimated value of represents a state variable of a third fixed-time filter; , , is a positive design parameter; is a continuous piecewise function.

[0120] Designing continuous piecewise functions for:

[0121] ;

[0122] in, ; ; ; ; ; represents the error of the third fixed-time filter, Represents a constant with a value greater than 0, which is a small positive parameter and can be used as the dividing point of the corresponding piecewise function; Indicates time.

[0123] In one implementation, obtaining an estimated value of system uncertainty corresponding to the posture subsystem through a preset first neural network includes:

[0124] Calculating a first prediction error using a predictor of the first neural network and historical data of the posture subsystem at a previous time;

[0125] outputting an estimated value of a first optimal neural network weight vector according to the first prediction error by an update rate of the first neural network;

[0126] An estimated value of the system uncertainty corresponding to the posture subsystem is output according to the estimated value of the first optimal neural network weight vector through the neural network estimator of the first neural network.

[0127] Specifically, the predictor of the first neural network outputs an estimated value of the derivative of the pitch rate according to the pitch rate, flight path angle, and angle of attack of the target aircraft. Then, the prediction error is calculated using the historical data of the attitude subsystem (three historical state values: pitch rate, flight path angle, and angle of attack) and the estimated value of the derivative of the pitch rate output by the predictor. Through the prediction error or the corresponding integral signal, the historical data is introduced into the update law to estimate the optimal neural network weight vector. The estimated optimal neural network weight vector is introduced into the neural network estimator, and the system uncertainty corresponding to the attitude subsystem is estimated through the neural network estimator. In short, in order to reduce the adverse effects of control errors on neural network learning, this embodiment sets a predictor for the neural network as an auxiliary prediction system, which can not only improve the data utilization rate of the neural network, but also improve the estimation accuracy.

[0128] It should be noted that the previous neural network-based backstepping control method was mainly designed from the perspective of Lyapunov stability, and the weight update usually depends on the tracking error, and the performance of the neural network is not taken seriously. Moreover, the update law of the neural network is completely dependent on the data at the current moment, and the historical data of the hypersonic aircraft is not effectively utilized. The neural network estimator in this embodiment effectively improves the data utilization rate by incorporating historical data into the design of the update law. And an auxiliary predictor is introduced in the neural network estimator to further improve the approximate performance. In addition, the use of fuzzy logic systems to estimate uncertainty completely relies on artificially formulated control rules, and does not have learning ability, so it is difficult to obtain the optimal control effect. This embodiment uses a neural network to estimate uncertainty, and the neural network has a self-learning function and parallel computing ability, which can quickly obtain the optimal control effect.

[0129] For example, for the posture sub-series, this embodiment uses a new neural network estimator with online learning of historical data to reconstruct the matching uncertainty :

[0130] ;

[0131] in, is the optimal neural network weight vector; is the basis function vector; is the neural network estimation error; , , They represent the neural network estimation error and the upper bound of the neural network estimation error respectively; Indicates transpose.

[0132] In order to introduce historical data online learning, a predictor is introduced as an auxiliary system:

[0133] ;

[0134] in, ; , is a positive design parameter, ; yes An estimated value of represents transpose; Indicates pitch rate The error of , is the elevator deflection degree.

[0135] Define two historical data as follows:

[0136] ;

[0137] ;

[0138] in, is the integration interval.

[0139] The forecast error based on historical data is defined as:

[0140] ;

[0141] in, represents the prediction error; , .

[0142] The update rate of the first neural network is:

[0143] ;

[0144] in, ; express The derivative of Represents the pitch angle q and the virtual controller difference; represents a positive neural network design parameter; represents a positive neural network design parameter.

[0145] In one implementation, for the attitude subsystem, this embodiment uses the backstepping design technology to design a controller to obtain a first fixed-time backstepping controller corresponding to the attitude subsystem. Through the design process of the above-mentioned fixed-time filters, virtual controllers and neural networks, the final controller (i.e., the first fixed-time backstepping controller) is designed as follows:

[0146] ;

[0147] in, , and is the design parameter, design continuous piecewise function for:

[0148] ;

[0149] in, ; ; ; ; , represents a constant with a value greater than 0 (which can be used as the dividing point of the corresponding piecewise function); express The exponential expression of .

[0150] It should be noted that the previous fixed-time backstepping scheme requires a bounded derivative of the virtual control rate assumption. When the system state converges to zero, this assumption may be violated, making the designed control scheme invalid. The fixed-time backstepping controller in this embodiment uses a first-order fixed-time filter to estimate the necessary variables, and designs a fixed-time controller based on the output of the filter, which not only meets the assumption that the derivative of the virtual control rate is bounded, but also reduces the computational burden of the hypersonic aircraft control scheme and expands the applicability of the control strategy.

[0151] Step S300: for the speed subsystem, obtaining an estimated value of the system uncertainty corresponding to the speed subsystem through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and outputting control information of the fuel equivalent ratio according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller.

[0152] Specifically, similar to the attitude subsystem, the speed subsystem also uses a neural network to estimate the uncertainty of the system, and constructs a fixed-time dynamic surface controller of the speed subsystem of the target aircraft through the backstepping design technology, that is, the second fixed-time backstepping controller. In actual application, the second fixed-time backstepping controller outputs the control information of the fuel equivalent ratio through the speed of the target aircraft, the target speed and the estimated value of the system uncertainty, so as to stabilize the system to the desired state within a fixed time.

[0153] In one implementation, the outputting of the control information of the fuel equivalent ratio by the preset second fixed-time backstepping controller according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem includes:

[0154] Calculating a speed difference according to the speed of the target aircraft and the target speed;

[0155] The control information of the fuel equivalent ratio is outputted through the second fixed-time backstepping controller according to the speed difference and the estimated value of the system uncertainty corresponding to the speed subsystem.

[0156] Specifically, by comparing the current speed of the target aircraft with the desired target speed, the numerical difference between the two, i.e., the speed difference, is obtained. The speed difference and the system uncertainty estimated by the neural network corresponding to the speed subsystem are input into the pre-built second fixed-time backstepping controller, which calculates and outputs the control information of the fuel equivalent ratio based on the input data.

[0157] For example, for the speed subsystem, the controller (i.e., the second fixed-time backstepping controller corresponding to the speed subsystem) is designed as:

[0158] ;

[0159] in, yes The estimate, represents the optimal neural network weight vector; is the basis function vector; , , denote positive design parameters respectively; Indicates the speed tracking error; Represents vector ; Represent the derivative of the reference velocity; design a continuous piecewise function for:

[0160] ;

[0161] in, ; ; ; ; , represents a constant with a value greater than 0 (which can be used as the dividing point of the corresponding piecewise function); Indicates the speed tracking error The exponential expression of .

[0162] In one implementation, obtaining an estimated value of system uncertainty corresponding to the speed subsystem through a preset second neural network includes:

[0163] Calculating a second prediction error using the predictor of the second neural network and the historical data of the speed subsystem at a previous time;

[0164] outputting an estimated value of a second optimal neural network weight vector according to the second prediction error by an update rate of the second neural network;

[0165] An estimated value of the system uncertainty corresponding to the speed subsystem is output according to the estimated value of the second optimal neural network weight vector through the neural network estimator of the second neural network.

[0166] Specifically, the predictor of the second neural network outputs an estimated value of the derivative of the speed according to the speed, flight path angle, and angle of attack of the target aircraft. The prediction error is then calculated using the historical data of the speed subsystem and the estimated value of the derivative of the speed output by the predictor. The historical data is introduced into the update law through the prediction error or the corresponding integral signal to estimate the optimal neural network weight vector. The estimated optimal neural network weight vector is introduced into the neural network estimator, and the system uncertainty corresponding to the speed subsystem is estimated through the neural network estimator. In short, in order to reduce the adverse effects of control errors on neural network learning, this embodiment sets a predictor for the neural network as an auxiliary prediction system, which can not only improve the data utilization rate of the neural network, but also improve the estimation accuracy.

[0167] For example, the update rate of the second neural network is:

[0168] ;

[0169] in, , The error in speed, represents an estimate of the speed; represents a positive neural network design parameter; ; , , , , denotes a positive design parameter.

[0170] In order to introduce historical data online learning, a predictor is introduced as an auxiliary system:

[0171] .

[0172] The definitions of the letters are detailed in the update rate of the second neural network and the controller for the speed subsystem. (i.e., the second fixed-time backstepping controller corresponding to the speed subsystem).

[0173] The technical effect of the present invention proves that:

[0174] The test uses a hypersonic aircraft as the target aircraft. The initial state of the hypersonic aircraft control system is set to , , , and The command signal is selected as and For the attitude subsystem, the filter parameters are set to , , , , , , , , .for , , ,set up , , The control gain is set to , , The adaptive law is chosen as , the other parameters of the neural network are set to , , and For the speed subsystem, the control gain is , , .for , Set to The adaptive law is chosen as , the other parameters of the neural network are set to , , and .

[0175] Figure 3 For height and speed tracking performance. Figure 3 The tracking results shown show that the algorithm developed by the present invention performs better than the algorithms used for comparison. When the system is affected by model uncertainty, the method of the present invention achieves better accuracy in tracking performance and obtains faster convergence speed under fixed time design. Figure 4 is the filter estimation result. After a short transition process, the estimation error converges to near 0. Figure 5 A response comparison curve of the control input is given, and it can be observed that the control input has a faster response under the controller proposed in the present invention. Figure 6 Response contrast curve for neural network weight updates.

[0176] It should be noted that the previous control methods can only ensure that the system state converges near the origin within a limited time convergence rate. The invention discloses a method for rapid maneuvering of an aircraft based on online learning of a neural network. Under the proposed fixed time backstepping control scheme, the filtering error, the neural network estimation error and the tracking error can reach a compact set close to zero at a fixed time convergence rate.

[0177] Summary of beneficial effects:

[0178] 1. The present invention proposes a fixed-time neural network estimator with online learning of historical data. Compared with some traditional neural network estimators, the neural network estimator proposed in the present invention effectively improves data utilization by incorporating historical data into the design of the update law. In addition, an auxiliary predictor is introduced in the proposed neural network estimator to further improve the estimation accuracy.

[0179] 2. The present invention proposes a new fixed-time backstepping control scheme, which cleverly uses a first-order fixed-time filter to estimate the necessary variables. Based on the output of the filter, a fixed-time controller is designed. The method proposed by the present invention can not only meet the assumption that the derivative of the virtual control law is bounded, but also reduce the computational burden of the control scheme and expand the applicability of the existing control strategy.

[0180] 3. The fixed-time backstepping control scheme proposed in the present invention can make the filtering error, neural network estimation error and tracking error reach a tight set close to zero at a fixed convergence speed.

[0181] Based on the above embodiments, the present invention also provides an aircraft rapid maneuvering device based on neural network online learning, such as Figure 7 As shown, the device comprises:

[0182] Pre-decomposition module 01 is used to pre-decompose the dynamic model of the target aircraft to obtain the velocity subsystem and attitude subsystem;

[0183] The elevator deflection angle control module 02 is used for obtaining an estimated value of the system uncertainty corresponding to the attitude subsystem through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; obtaining an estimated value of the derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack and the target flight path angle to be achieved through a preset fixed time filter; and outputting control information of the elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate and the estimated value of the system uncertainty of the attitude subsystem through a preset first fixed time backstepping controller;

[0184] The fuel equivalent ratio control module 03 is used to obtain an estimated value of the system uncertainty corresponding to the speed subsystem through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and output control information of the fuel equivalent ratio according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller.

[0185] The aforementioned explanation of the embodiment of the method for rapid maneuvering of an aircraft based on online learning of a neural network is also applicable to the rapid maneuvering device for an aircraft based on online learning of a neural network in this embodiment, and will not be repeated here.

[0186] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 8 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for rapid maneuvering of an aircraft based on online learning of a neural network is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0187] Those skilled in the art will understand that Figure 8 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0188] In one implementation, the terminal has one or more programs stored in its memory and is configured to be executed by one or more processors. The one or more programs include instructions for performing an aircraft rapid maneuvering method based on neural network online learning.

[0189] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0190] In summary, the present invention discloses a method for rapid maneuvering of an aircraft based on online learning of a neural network, in which unknown system uncertainties are processed by a neural network based on online learning of historical data. The derivative of the virtual control and some necessary information are estimated by a fixed-time filter. Based on the information estimated by the neural network and the fixed-time backstepping controller, the fixed-time tracking problem of the hypersonic aircraft is solved. Since the neural network has a self-learning function and parallel computing capability, the optimal control effect can be quickly obtained. Therefore, the fixed-time backstepping controller of the present invention can ensure that the control system of a hypersonic aircraft with uncertainty stabilizes the system attitude and speed to the desired state within a fixed time.

[0191] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for rapid maneuvering of an aircraft based on online learning of a neural network, characterized in that: The method comprises: Decompose the dynamic model of the target aircraft in advance to obtain the velocity subsystem and attitude subsystem; For the attitude subsystem, an estimated value of the system uncertainty corresponding to the attitude subsystem is obtained through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; an estimated value of the derivative of the target virtual control amount is obtained through a preset fixed time filter according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle to be achieved; and control information of the elevator deflection angle is output through a preset first fixed time backstepping controller according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and the estimated value of the system uncertainty of the attitude subsystem; For the speed subsystem, an estimated value of the system uncertainty corresponding to the speed subsystem is obtained through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and control information of the fuel equivalent ratio is output according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller; The method comprises: obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed time filter; and outputting control information of an elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed time backstepping controller, including: Obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers; generating a target continuous function according to the target virtual control amount and the pitch rate; Outputting control information of the elevator deflection angle according to an estimated value of the derivative of the target virtual control amount, the target continuous function and an estimated value of system uncertainty of the attitude subsystem through the first fixed-time backstepping controller; The method of obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers comprises: generating a first continuous function according to the flight path angle and the target flight path angle; Outputting an estimated value of a derivative of the first continuous function according to the first continuous function through a preset first fixed time filter; Outputting a first virtual control amount according to the first continuous function and an estimated value of a derivative of the first continuous function through a preset first virtual controller; generating a second continuous function according to the first virtual control amount, the angle of attack, and the flight path angle; Outputting an estimated value of a derivative of the second continuous function according to the second continuous function through a preset second fixed time filter; Outputting the target virtual control amount according to the second continuous function and an estimated value of a derivative of the second continuous function through a preset second virtual controller; An estimated value of a derivative of the target virtual controlled amount is output according to the target virtual controlled amount through a preset third fixed time filter.

2. The method for rapid aircraft maneuvering based on neural network online learning according to claim 1, characterized in that: The system states corresponding to the dynamic model include speed, flight altitude, flight path angle, angle of attack and pitch rate, and the control input information includes elevator deflection angle and fuel equivalent ratio; after decomposition, the system states corresponding to the speed subsystem include speed, and the system states corresponding to the attitude subsystem include flight altitude, flight path angle, angle of attack and pitch rate.

3. The method for rapid aircraft maneuvering based on neural network online learning according to claim 1, characterized in that: The step of obtaining an estimated value of system uncertainty corresponding to the posture subsystem through a preset first neural network includes: Calculating a first prediction error using a predictor of the first neural network and historical data of the posture subsystem at a previous time; outputting an estimated value of a first optimal neural network weight vector according to the first prediction error by an update rate of the first neural network; An estimated value of the system uncertainty corresponding to the posture subsystem is output according to the estimated value of the first optimal neural network weight vector through the neural network estimator of the first neural network.

4. The method for rapid aircraft maneuvering based on neural network online learning according to claim 1, characterized in that: The outputting of control information of the fuel equivalent ratio by a preset second fixed-time backstepping controller according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of the system uncertainty corresponding to the speed subsystem includes: Calculating a speed difference according to the speed of the target aircraft and the target speed; The control information of the fuel equivalent ratio is outputted through the second fixed-time backstepping controller according to the speed difference and the estimated value of the system uncertainty corresponding to the speed subsystem.

5. The method for rapid aircraft maneuvering based on neural network online learning according to claim 1, characterized in that: The obtaining, by means of a preset second neural network, an estimated value of system uncertainty corresponding to the speed subsystem comprises: Calculating a second prediction error using the predictor of the second neural network and the historical data of the speed subsystem at a previous time; outputting an estimated value of a second optimal neural network weight vector according to the second prediction error by an update rate of the second neural network; An estimated value of the system uncertainty corresponding to the speed subsystem is output according to the estimated value of the second optimal neural network weight vector through the neural network estimator of the second neural network.

6. A rapid maneuvering device for aircraft based on online learning of neural networks, characterized in that: The device comprises: A pre-decomposition module is used to pre-decompose the dynamic model of the target aircraft to obtain a velocity subsystem and an attitude subsystem; an elevator deflection angle control module, for obtaining an estimated value of system uncertainty corresponding to the attitude subsystem through a preset first neural network, wherein the first neural network performs online learning based on historical data of the attitude subsystem; obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed-time filter; and outputting control information of the elevator deflection angle according to the target virtual control amount, an estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed-time backstepping controller; a fuel equivalent ratio control module, configured to obtain an estimated value of system uncertainty corresponding to the speed subsystem through a preset second neural network, wherein the second neural network performs online learning based on historical data of the speed subsystem; and output control information of the fuel equivalent ratio according to the speed of the target aircraft, the target speed to be achieved, and the estimated value of system uncertainty corresponding to the speed subsystem through a preset second fixed-time backstepping controller; The method comprises: obtaining an estimated value of a derivative of a target virtual control amount according to a pitch rate, a flight path angle, an angle of attack, and a desired target flight path angle through a preset fixed time filter; and outputting control information of an elevator deflection angle according to the target virtual control amount, the estimated value of the derivative of the target virtual control amount, the pitch rate, and an estimated value of system uncertainty of the attitude subsystem through a preset first fixed time backstepping controller, including: Obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack, and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers; generating a target continuous function according to the target virtual control amount and the pitch rate; Outputting control information of the elevator deflection angle according to an estimated value of the derivative of the target virtual control amount, the target continuous function and an estimated value of system uncertainty of the attitude subsystem through the first fixed-time backstepping controller; The method of obtaining an estimated value of a derivative of the target virtual control amount according to the pitch rate, the flight path angle, the angle of attack and the target flight path angle through a plurality of alternately cascaded fixed-time filters and virtual controllers comprises: generating a first continuous function according to the flight path angle and the target flight path angle; Outputting an estimated value of a derivative of the first continuous function according to the first continuous function through a preset first fixed time filter; Outputting a first virtual control amount according to the first continuous function and an estimated value of a derivative of the first continuous function through a preset first virtual controller; generating a second continuous function according to the first virtual control amount, the angle of attack, and the flight path angle; Outputting an estimated value of a derivative of the second continuous function according to the second continuous function through a preset second fixed time filter; Outputting the target virtual control amount according to the second continuous function and an estimated value of a derivative of the second continuous function through a preset second virtual controller; An estimated value of a derivative of the target virtual controlled amount is output according to the target virtual controlled amount through a preset third fixed time filter.

7. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the aircraft rapid maneuvering method based on neural network online learning as described in any one of claims 1-5; and the processor is used to execute the program.

8. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the method for rapid maneuvering of an aircraft based on online learning of a neural network as described in any one of claims 1 to 5 above.

Images