CN102289714A - Method for controlling autonomous take-off and landing of small unmanned rotorcraft based on behavioral model - Google Patents

Abstract

A behavior model-based autonomous take-off and landing control method for a small unmanned rotorcraft involves learning flight controller hand behavior data, building a flight controller hand behavior model, and designing autonomous take-off and landing rules. Firstly, through the data acquisition system to collect the control behavior data of the flight control hand; secondly, according to the working characteristics of the small unmanned rotorcraft in the autonomous take-off and landing stage, the flight control hand behavior model is constructed through the fuzzy control method, and the neural network is optimized to improve the flight control hand. Behavior model performance; by analyzing the behavior characteristics of the flight control hand and the take-off and landing characteristics of the small unmanned rotorcraft, the autonomous take-off and landing behavior rules are constructed. The invention has the advantages of strong anti-interference, high stability, convenient design and the like, and can be used for autonomous take-off and landing control of small unmanned rotorcraft in complex environments and the like.

Description

A Behavioral Model-Based Autonomous Take-off and Landing Control Method for Small Unmanned Rotorcraft

technical field

The invention relates to a behavior model-based autonomous take-off and landing control method for a small unmanned rotorcraft, which is suitable for the autonomous control field of aerial unmanned robots.

Background technique

The small unmanned rotorcraft has the characteristics of vertical take-off and landing, hovering, etc., and can perform tasks in narrow spaces such as urban streets, and has a wide range of application prospects. With the expansion of the application field, the demand for the intelligence of small unmanned rotorcraft is also increasing, and the fully autonomous and highly intelligent small unmanned rotorcraft has become a research hotspot.

As a complex multiple-input and multiple-output control system, small unmanned rotorcraft has the characteristics of strong coupling, nonlinearity, and high control difficulty. At present, the take-off and landing of small unmanned rotorcraft are mainly controlled by manual remote control, which largely depends on the skills and proficiency of the flight controller. However, in practical applications, most of the military and civilian users do not have professional control skills, and the small unmanned rotorcraft R&D department needs to conduct long-term training for the users, thus limiting the promotion and application of small unmanned rotorcraft. application.

In order to improve the performance, various control methods such as PID control method, robust control method and intelligent control method are used for the autonomous take-off and landing control of small unmanned rotorcraft. The intelligent PID controller controls the attitude of the unmanned rotorcraft to improve the control accuracy of the system, but the anti-interference ability is poor, and the ground effect interference and wind disturbance are large during the take-off and landing phase of the small unmanned rotorcraft, and the intelligent PID control method is difficult to achieve stable autonomous start. down function. Robust control eliminates the perturbation of some model parameters when the UAV is flying, but it has the characteristics of real-time and slow response of dynamic parameters. The nonlinear adaptive control of the neural network overcomes the uncertainty, modellessness and potential dynamic nonlinearity of the parameters of the unmanned helicopter, and realizes the attitude control of the unmanned helicopter, but requires a large number of sample calculations.

Contents of the invention

The technical solution problem of the present invention is: aiming at the deficiency of the existing control method of small unmanned rotorcraft, a kind of behavioral model-based autonomous take-off and landing control method of small unmanned rotorcraft is proposed, which solves the problem of the autonomous take-off and landing controller of small unmanned rotorcraft design problem.

The technical solution of the present invention is: the present invention proposes an autonomous take-off and landing control method based on an adaptive pilot hand behavior model, collects flight controller hand behavior data, analyzes the take-off and landing characteristics of a small unmanned rotorcraft, and determines behavior rules based on The method of learning the flight control hand behavior data, constructing the flight control hand behavior model, and optimizing the flight control hand behavior model through the radial basis neural network, the specific steps are as follows:

(1) Collect flight controller hand behavior data

The flight controller controls the small unmanned rotorcraft for take-off and landing operations, and collects the control data of the main propeller pitch and throttle threshold during the take-off and landing phase of the small unmanned rotorcraft through the data acquisition system, and the corresponding horizontal position of the small unmanned rotorcraft , altitude, speed status information.

(2) Build the flight control hand behavior model

By learning the control behavior data of the flight control hand and the state information of the small unmanned rotorcraft, the self-adaptive fuzzy control method is used to construct the flight control hand behavior model; THEN (conditional conclusion) controls the process;

input layer output node $o_i^{\left(1\right)}=x_i,i=\mathrm{1,2}---\left(1\right)$

为输入层的输出；Among them, _Xi is the input of the input layer, which are the height error Δheight and the height error variation

is the output of the input layer;

Fuzzy layer input node $I_{\mathrm{ik}}^{\left(2\right)}=-{(x_i-a_{\mathrm{ik}})}^2/(k_i\&CenterDot;{{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA0000065776690000031.png"\; state="\{\{state\}\}">b</figure-callout>}}^2}_{\mathrm{ik}}),i=1,2,k=\mathrm{1,2}\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;N---\left(2\right)$

output node $o_{\mathrm{ik}}^{\left(2\right)}=\exp \left(I_{\mathrm{ik}}^{\left(2\right)}\right)---\left(3\right)$

表示i输入对应的k类高斯函数的均方差值，k_i为径向基神经网络对i输入的调控参数，N为输入对应的模糊变量数目；Among them, a _ik represents the central value of the k-type Gaussian function corresponding to the i input,

Indicates the mean square error value of the k-type Gaussian function corresponding to the i input, _ki is the control parameter of the radial basis neural network for the i input, and N is the number of fuzzy variables corresponding to the input;

The rule layer generates corresponding N ² fuzzy rules based on the MAX-MIN (maximum and minimum) product synthesis method, and its input node

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&cap;</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="2"\; label="I"\; filenames="HDA0000065776690000031.png"\; state="\{\{state\}\}">I</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>$

l=N (m-1)+n m=1, 2...N, n=1, 2...N (4)

output node $o_l^{\left(3\right)}=I_l^{\left(3\right)}---\left(5\right)$

Deblurring layer input node $I^{\left(4\right)}=\underset{j=1}{\overset{\mathrm{ll}}{\mathrm{\&Sigma;}}}{o}_j^{\left(3\right)}\&Center\; Dot;w_j,\mathrm{ll}=N^2---\left(6\right)$

output node $o^{\left(4\right)}=k_3\&CenterDot;I^{\left(4\right)}/\underset{j=1}{\overset{\mathrm{ll}}{\mathrm{\&Sigma;}}}{o}_j^{\left(3\right)},\mathrm{ll}=N^2---\left(7\right)$

Among them, the jth rule of w _j corresponds to the weight value, k ₃ is the control parameter of radial basis neural network input to i, and O ⁽⁴⁾ is the control amount of collective pitch and rudder.

Select 150 sets of training results of autonomous take-off and landing of small unmanned rotorcraft based on the flight control hand behavior model under different disturbance environments as samples, and build a radial basis neural network to optimize the Gaussian mean square error of the fuzzy layer of the flight control hand behavior model in real time value and the control amount of the collective pitch rudder; the radial basis neural network is a three-layer neural network, including the input layer, the hidden layer and the output layer constituted by the three-layer radial basis neural network; the output node of the input layer is

${\mathrm{<span\; class="notranslate">\; OO</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

Among them, XX _i is the input of the input layer, which are the horizontal position difference Δlateral and the speed difference Δspeed when the small unmanned rotorcraft is in the hovering state during the off-ground stage, is the output of the input layer;

Hidden layer input node: $I_c^{\left(2\right)}=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ic}}{\mathrm{XX}}_i,c=\mathrm{1,2}\&CenterDot;\&CenterDot;\&CenterDot;Z---\left(9\right)$

Output node: ${\mathrm{OO}}_c^{\left(2\right)}=\exp ({(I_c^{\left(2\right)}-{\mathrm{aa}}_c)}^2/{\mathrm{bb}}_c^2),c=\mathrm{1,2}\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;Z---\left(10\right)$

表示对应的c个高斯函数的均方差值，Z为隐含层神经元数目；Among them, aa _c represents the center value of the corresponding c Gaussian functions,

Indicates the mean square error value of the corresponding c Gaussian functions, Z is the number of neurons in the hidden layer;

output layer input node ${I_p}^{\left(3\right)}=\underset{e=1}{\overset{Z}{\mathrm{\&Sigma;}}}{\mathrm{oo}}_{\mathrm{pe}}^{\left(2\right)}\&CenterDot;w_{\mathrm{pe}},p=\mathrm{1,2,3}---\left(11\right)$

output node k _p ⁽³⁾ = I _p ⁽³⁾ (12)

The output k ₁ of the output layer is used to adjust the coefficient of the Gaussian function corresponding to the altitude error, k ₂ is used to adjust the coefficient of the Gaussian function corresponding to the altitude error, and k ₃ is used to adjust the coefficient of the output collective pitch rudder.

(3) Design autonomous takeoff and landing behavior rules

At the initial stage, the adhesion is obtained through the throttle and the collective distance threshold. During the lift-off stage, the longitudinal position and altitude control obtained by eliminating external disturbances through the adaptive flight control hand behavior model, and the PID controls the position, speed and attitude of the transverse plane. Use PID to control position, speed and attitude; the autonomous landing phase is divided into flight phase, lift-off phase, and landing phase. Attitude control is achieved through PID during the flight phase, longitudinal position and altitude control is implemented through the adaptive flight controller hand behavior model during the takeoff phase, horizontal position, speed and attitude control is achieved through PID, and position control is realized through the flight controller hand behavior model during the landing phase.

The advantage of the present invention compared with prior art is:

(1) The present invention learns the control behavior data of the flight control hand through the adaptive fuzzy control method, simulates the control behavior of the skilled flight control hand, has weak dependence on the model of the small unmanned rotorcraft, and can optimize the rules online in real time according to the actual flight situation, Strong anti-interference ability;

(2) The present invention is based on the well-trained flight control hand behavior model, combined with traditional control methods, can adjust the rudder amount according to the status information of the aircraft in real time, with small calculation amount, good real-time performance, and fast response speed of dynamic parameters;

(3) The flight control hand behavior model proposed by the present invention is based on the fuzzy control method, learns the flight control hand behavior data, and performs real-time performance optimization through the radial basis network. On the basis of reasonable rules, only fewer samples are needed Reasonable rules can be derived.

Description of drawings

Figure 1 is the autonomous control process of a small unmanned rotorcraft;

Figure 2 is a small unmanned rotorcraft adaptive flight control hand behavior model;

Figure 3 is a three-dimensional track map of the autonomous take-off of a small unmanned rotorcraft;

Figure 4 is a three-dimensional track map of the autonomous landing of a small unmanned rotorcraft.

Detailed ways

As shown in Figures 1 and 2, the specific implementation method of the present invention is as follows:

(1) Collect flight controller hand behavior data

The flight controller controls the small unmanned rotorcraft for take-off and landing operations, and the data acquisition system collects the main propeller collective pitch and the control of the throttle controller during the take-off and landing phase of the small unmanned rotorcraft by the data acquisition system according to the clock and a sampling period of 10Hz Input data, according to the flight state of the small unmanned rotorcraft, determine the take-off threshold of collective pitch and throttle, the control threshold of the hovering phase, the idle speed threshold of the landing phase, and the horizontal position information and height of the small unmanned rotorcraft during the take-off and landing phase The state information and speed information are used to construct the training samples of the flight control hand behavior model.

(2) Build the flight control hand behavior model

By studying the state information of the height error and height error variation of the small unmanned rotorcraft collected by the data acquisition system and the corresponding input of the main propulsion collective distance, the hand behavior model of the flight controller is constructed by using the adaptive fuzzy control method; the model consists of the input layer , fuzzy layer, rule layer and defuzzy layer, representing the IF-THEN control process;

input layer output node $o_i^{\left(1\right)}=x_i,i=\mathrm{1,2}---\left(1\right)$

为输入层的输出；Among them, _Xi is the input of the input layer, which are the height error Δheight and the height error variation between the actual height and the expected height collected by the data acquisition system

is the output of the input layer;

Fuzzy layer input node $I_{\mathrm{ik}}^{\left(2\right)}=-{(x_i-a_{\mathrm{ik}})}^2/(k_i\&Center\; Dot;{{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="HDA0000065776690000031.png"\; state="\{\{state\}\}">b</figure-callout>}}^2}_{\mathrm{ik}}),i=1,2,k=\mathrm{1,2}\&Center\; Dot;\&CenterDot;\&CenterDot;N---\left(2\right)$

output node $o_{\mathrm{ik}}^{\left(2\right)}=\exp \left(I_{\mathrm{ik}}^{\left(2\right)}\right)---\left(3\right)$

Among them, a _ik represents the central value of the k-type Gaussian function corresponding to the i input, Indicates the mean square error value of the k-type Gaussian function corresponding to the i input, _ki is the control parameter of the radial basis neural network for the i input, and N is the number of fuzzy variables corresponding to the input;

The rule layer generates corresponding N ² fuzzy rules based on the MAX-MIN (maximum and minimum) product synthesis method, and its input node

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="2"\; label="o"\; filenames="HDA0000065776690000031.png"\; state="\{\{state\}\}">o</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; m</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&cap;</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span>$

l=N (m-1)+n m=1, 2...N, n=1, 2...N (4)

output node $o_l^{\left(3\right)}=I_l^{\left(3\right)}---\left(5\right)$

Deblurring layer input node $I^{\left(4\right)}=\underset{j=1}{\overset{\mathrm{ll}}{\mathrm{\&Sigma;}}}{o}_j^{\left(3\right)}\&CenterDot;w_j,\mathrm{ll}=N^2---\left(6\right)$

output node $o^{\left(4\right)}=k_3\&CenterDot;I^{\left(4\right)}/\underset{j=1}{\overset{\mathrm{ll}}{\mathrm{\&Sigma;}}}{o}_j^{\left(3\right)},\mathrm{ll}=N^2---\left(7\right)$

和w_j通过遗传算法训练得到。Among them, the jth rule of w _j corresponds to the weight value, k ₃ is the control parameter input by the radial basis neural network to i, O ⁽⁴⁾ is the control amount of the collective pitch rudder, a _ik ,

and w _j are obtained through genetic algorithm training.

Select 150 sets of training results of autonomous take-off and landing of small unmanned rotorcraft based on the flight control hand behavior model under different disturbance environments as samples, and build a radial basis neural network to optimize the Gaussian mean square error of the fuzzy layer of the flight control hand behavior model in real time value and the control amount of the collective pitch rudder; the radial basis neural network is a three-layer neural network, including the input layer, the hidden layer and the output layer constituted by the three-layer radial basis neural network; the output node of the input layer is

${\mathrm{<span\; class="notranslate">\; OO</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

为输入层的输出；Among them, XX _i is the input of the input layer, which are respectively the horizontal position difference Δlateral and the speed difference Δspeed between the horizontal position and the expected position of the small unmanned rotorcraft collected by the data acquisition system during the stage of leaving the ground,

is the output of the input layer;

Hidden layer input node: $I_c^{\left(2\right)}=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ic}}{\mathrm{XX}}_i,c=\mathrm{1,2}\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;Z---\left(9\right)$

Output node: ${\mathrm{OO}}_c^{\left(2\right)}=\exp ({(I_c^{\left(2\right)}-{\mathrm{aa}}_c)}^2/{\mathrm{bb}}_c^2),c=\mathrm{1,2}\&Center\; Dot;\&CenterDot;\&CenterDot;Z---\left(10\right)$

表示对应的c个高斯函数的均方差值，Z为隐含层神经元数目；Among them, aa _c represents the center value of the corresponding c Gaussian functions,

Indicates the mean square error value of the corresponding c Gaussian functions, Z is the number of neurons in the hidden layer;

output layer input node ${I_p}^{\left(3\right)}=\underset{e=1}{\overset{Z}{\mathrm{\&Sigma;}}}{\mathrm{oo}}_{\mathrm{pe}}^{\left(2\right)}\&CenterDot;w_{\mathrm{pe}},\mathrm{<figure-callout\; id="2"\; label="p"\; filenames="HDA0000065776690000031.png"\; state="\{\{state\}\}">p</figure-callout>}\mathrm{1,2,3}---\left(11\right)$

output node k _p ⁽³⁾ = I _p ⁽³⁾ (12)

The output k ₁ of the output layer is used to adjust the coefficient of the Gaussian function corresponding to the altitude error, k ₂ is used to adjust the coefficient of the Gaussian function corresponding to the altitude error, and k ₃ is used to adjust the coefficient of the output collective pitch rudder.

(3) Design autonomous takeoff and landing behavior rules

Analyzing the flight characteristics of the small unmanned rotorcraft during the take-off stage, the autonomous take-off stage is mainly divided into the initial stage, the take-off stage, and the flight stage. In the initial stage, after receiving the autonomous take-off command, the small unmanned rotorcraft takes the current point as the take-off point, and through the learning of the flight controller’s hand behavior data, increases the throttle and collective distance control at a constant speed to the take-off value collected by the data acquisition system. Threshold, in order to obtain the desired speed and lift, when the throttle and collective distance are reached, the throttle will be saturated, and the small unmanned rotorcraft will enter the critical state of lift-off; during the lift-off phase, the longitudinal position and height will be obtained through the adaptive flight control hand behavior model Control, the PID method is used to control the pitch and roll of the horizontal plane, and the attitude lock control is performed based on the heading error calculated by the magnetic compass. When the barometric altitude measurement is greater than 4m, it enters the flight stage; in the flight stage, the target is 10m above the take-off point Hover point, through the coupling control of pitch, roll, heading and altitude loops, realizes autonomous hovering control.

The flight characteristics of the landing stage of the small unmanned rotorcraft are analyzed, and the autonomous landing stage is mainly divided into the flight stage, the lift-off stage, and the landing stage. In the flight phase, after the small unmanned rotorcraft receives the landing command, based on the current state, it takes 10 meters above the landing point as the target hovering point, and realizes autonomous hovering control through the coupling control of the pitch, roll, heading and altitude loops ; Then based on the horizontal position error, the stability control is realized through the speed and attitude, the attitude lock control is performed based on the heading error calculated by the magnetic compass, and the error information between the actual altitude and the expected altitude is obtained based on the air pressure information, and the mode of large scale and small limit is used Carry out autonomous height control. When the measured height is less than 2 meters, it enters the stage of lift-off; in the stage of lift-off, the small unmanned rotorcraft realizes the position and attitude control of the horizontal plane through the PID loop, and realizes the longitudinal control through the adaptive flight control hand behavior model. The height control of the plane, when the measured height is less than 0.2 meters, enters the landing stage; the horizontal attitude is the expected attitude angle, and the pitch, roll, and yaw channel control is realized based on the attitude loop, which is collected by the data acquisition system through the flight control hand behavior model The threshold value obtained in the landing stage is the expected value and enters the collective pitch lock, and the throttle is controlled to decrease at a constant speed within 4 seconds to the idle throttle in the landing stage of the small unmanned rotorcraft collected by the data acquisition system, and the lift is reduced to achieve a stable landing. After more than 4 seconds , the system directly controls the fixed value through the collective pitch and the throttle to obtain greater adhesion and improve system reliability.

flight example

Based on the Thunder Tiger 90 small unmanned aerial vehicle for flight verification, the behavior model-based autonomous take-off and landing control method for small unmanned rotorcraft, the small unmanned rotorcraft can achieve stable autonomous take-off and landing functions; the autonomous flight trajectory is shown in Figure 3, the small After receiving the autonomous take-off command, the unmanned rotorcraft takes (-7, -3, 0) as the take-off point, and completes the flight mission within 10 seconds. The position error of the final hovering point is less than 2m, the height error is less than 1m, and the hovering speed The error is less than 0.5m/s; the autonomous landing process of the small unmanned rotorcraft is shown in Figure 4. After receiving the landing command at any point, the small unmanned rotorcraft directly flies to the target hovering point (0, 0, 10) autonomously , The autonomous take-off and landing process is completed within 15 seconds. Due to the influence of ground effect, the actual landing point of the small unmanned rotorcraft is (1, 2, 0) and the position error is less than 2m.

The behavior model-based autonomous take-off and landing control method of the small unmanned rotorcraft of the present invention overcomes the shortcomings of the existing control methods, and can realize fully autonomous take-off and landing control of the small unmanned rotorcraft in a complex environment.

The contents not described in detail in the description of the present invention belong to the prior art known to those skilled in the art.

Claims (2)

1. autonomous landing control method of the miniature self-service giro based on behavior model is characterized in that performing step is as follows:

(1) gather fly to control hand control the main slurry in miniature self-service giro landing stage total apart from, throttle threshold values and miniature self-service giro at the horizontal level in landing stage, highly, speed state information;

(2) structure flies to control the hand behavior model, fly to control hand by study and control behavioral data and miniature self-service giro status information, utilize adaptive fuzzy control method to make up and fly to control the hand behavior model, this model is made of input layer, obscuring layer, rules layer and deblurring layer, expression condition conclusion IF-THEN control procedure; Input layer be input as height error Δ height and height error variable quantity Be output as height error and height error variable quantity through normalized; The output of input layer is as the input of obscuring layer, and the member function of the controller of obscuring layer utilizes the Gaussian function definition:

Q _ik＝exp(-(X _i-a _ik) ²/(k _i·b ² _ik))，i＝1，2，k＝1，2…N

Wherein, Q _IkFor the output of the fuzzy variable of correspondence, represent N kind situation respectively, x _iBe the input value after the process normalized, a _IkThe central value of the k class Gaussian function that expression i input is corresponding,

The mean square deviation of the k class Gaussian function that expression i input is corresponding, k _iBe the regulation and control parameter of radial base neural net to the i input, N is the corresponding fuzzy variable number of input; Obscuring layer is output as the fuzzy variable of height error and height error variable quantity correspondence; The output that is input as obscuring layer of rules layer generates corresponding N based on minimax MAX-MIN product synthetic method ²Individual fuzzy rule; At the deblurring layer, the numerical value that utilizes the membership function weighting average decision method that rules layer is drawn is converted into the accurate output of direct controlling object, promptly always apart from the controlled quentity controlled variable of rudder amount, and by radial base neural net input and the input and output that fly to control the hand behavior model are improved, improve and fly to control the adaptability of hand behavior model environment;

(3) determine the rule of conduct in autonomous landing stage, initial period, liftoff stage and mission phase will be divided into autonomous takeoff phase, initial period is by throttle and always obtain adhesion apart from threshold values, the liftoff stage flies to control the hand behavior model by self-adaptation and eliminates outer attitude control of disturbing acquisition, and mission phase is realized attitude control by conventional control method; The autonomous landing stage is divided into mission phase, liftoff stage and landing stage, mission phase is realized attitude control by conventional control method, the liftoff stage flies to control position and the attitude control that hand behavior model and conventional control method realize by self-adaptation, and the stage that lands is realized position control by flying to control the hand behavior model.

2. the autonomous landing control method of the miniature self-service giro based on behavior model according to claim 1, it is characterized in that: the described radial base neural net of step (2) carries out improved method to input and the input and output that fly to control the hand behavior model and is: come real-time optimization to fly to control the Gaussian function mean square deviation of hand behavior model obscuring layer and always apart from the controlled quentity controlled variable of rudder amount by the sample data train RBF Neural Network, radial base neural net is a three-layer neural network, comprise input layer, hidden layer and output layer constitute, horizontal level difference and the speed difference of the liftoff stage of miniature self-service giro that is input as of input layer floating state the time; The output that is input as input layer of hidden layer, hidden layer have Z Gaussian function to constitute, through the input of weights stack as output layer; Output layer is output as the coefficient k of the Gaussian function that is used to adjust the height error correspondence ₁, be used to adjust the coefficient k of the Gaussian function of height error correspondence ₂And be used to adjust output always apart from the coefficient k of rudder amount ₃

Images