CN106406340A - Quad-rotor unmanned aerial vehicle and control method thereof - Google Patents

Abstract

The invention provides a quad-rotor unmanned aerial vehicle and a control method thereof. The quad-rotor unmanned aerial vehicle comprises a body and a main controller module and further comprises a navigation module, a sensor module, a communication module and a motor control module which are respectively connected with the main controller module, wherein the main controller module is used for comprehensively calculating the real-time flight attitude information fed back by the sensor module and the navigation module and the control information of the communication module and then outputting a motor control signal to control the motor control module. The quad-rotor unmanned aerial vehicle is advantaged in that automatic adaptation to change of the external environment can be realized to modulate flight parameters to realize the pre-determined control effect. According to the control method, the main controller module employs a control scheme combining a back-propagation artificial neural network with inertia term (BPNNI) neural network and based on weight adjustment with proportional integral derivative (PID) control to calculate an actual output control variable.

Description

Four rotor unmanned aircrafts and its control method

Technical field

The present invention relates to unmanned vehicle technical field is and in particular to four rotor unmanned aircrafts and its control method.

Background technology

Four rotor unmanned aircrafts are that by wireless remote control equipment, autonomous flight realized by self-sensor device to one kind in addition Not manned vehicle, it has 6 frees degree, 4 control inputs, drives, by 4 brshless DC motors, the differential moment producing Realize its elevating movement and tumbling motion, the anti-twisted moment of generation realizes yawing rotation, is Nonlinear Underactuated System.This kind of winged Row device is widely used in military and civilian field.Four rotor unmanned aircrafts compare fixed-wing unmanned vehicle, because energy is vertical Landing, the requirement of take-off and landing is relatively low, and flexibility is high, has higher adaptability under complicated physical features.At present using at most Control method be PID control, it passes through to identify target, then detects the gap of present situation and target, then with action elimination it. PID control structure is simple, and control technology is ripe, and robustness is preferable.But, in the middle of four rotor unmanned aircraft flight courses, Parameter in the middle of controller is difficult to automatically adjust to adapt to extraneous change, thus is extremely difficult to predetermined target, and impact controls Effect.

Content of the invention

Present invention seek to address that technical problem present in prior art.

For this reason, the present invention provides a kind of four rotor unmanned aircrafts (Quadrotor Unmanned Aerial Vehicle, QUAV) and its control method, it is based on the BP neural network (Back- that weighed value adjusting amount adds " Inertia " Propagation Artificial Neural Network with Inertia Term, BPNNI) and PID control The control method that (Proportional Integral Derivative Control, PID) combines is controlling four rotors no So that unmanned vehicle antijamming capability is strengthened, itself robustness is improved people's aircraft.

A kind of four rotor unmanned aircrafts, including body and main controller module, also include respectively with described master controller Navigation module, sensing module, communication module and motor control module that module connects are it is characterised in that described navigation module adopts Satellite navigation system is positioned to aircraft and provides positional information to described main controller module and navigate, and is navigating through Change in journey and solidification baud rate；Described sensor assembly includes the inertia measurement list being connected respectively with described main control module Unit, baroceptor, electronic compass and air velocity transducer, three axis that described Inertial Measurement Unit is used for sense aircraft add Speed, rolling angular speed, pitch rate, yawrate and course information, described baroceptor is used for sense aircraft Height, described electronic compass is used for measuring the course information of aircraft, and described air velocity transducer is to aircraft location Wind speed be monitored；Described wireless communication module includes remote control, PPM decoder and PPM receiver, described PPM encoder It is connected with described remote control, four channel control signals of remote control will be wirelessly transmitted to described after PPM encoder encodes PPM receiver, described PPM receiver is connected with described main controller module；Described motor control module includes winged for driving Four motors of four rotors of row device and the electron speed regulator controlling described four motors work respectively, described electron speed regulator It is connected with described main controller module to receive motor control signal.

In four rotor unmanned aircrafts that the present invention provides, main controller module is the core of control system, its work With being responsible for gathering three axis accelerometers, rolling angular speed, pitch rate and the yawrate that sensor assembly detects Deng the attitude angular rate of composition, course information and wind speed real-time resolving, sent by remote control and through PPM further according to detecting Flight information after encoder coding, the BP neural network adding " Inertia " in conjunction with weighed value adjusting amount is combined with PID control Flight Control Scheme, calculates the rotating speed to control four motors for the actual output control amount, thus realizing the ginseng in the middle of controller Number automatically adjustment, to adapt to extraneous change, reaches preferable flight control effect, completes predetermined target.Main controller module After real-time flight attitude information that sensor assembly described in COMPREHENSIVE CALCULATING and navigation module are fed back and the control information of remote control Output motor control signal is with controlled motor control module, so that four rotor unmanned aircrafts can adapt to extraneous ring automatically The change in border to reach predetermined control effect to modulate flight parameter.

Further, four rotor unmanned aircrafts also include water-cooled-air cooling module, described water-cooled-air cooling module respectively with Described main controller module and described motor control module connect, with to described main controller module and described motor control module Cooling, and accept the control of described main controller module to realize the regulation to cooling power simultaneously.

Further, described main controller module and described motor control module are respectively arranged with temperature element, described Temperature element is connected to realize temperature acquisition with described main controller module, and described main controller module is according to collected temperature Information adjusts the cooling power of described water-cooled-air cooling module.

Further, described sensor assembly includes the Smoke Sensor being connected with described main control module, described smog Sensor arrangement close on carry-on circuit board annex with detect described circuit board break down generation smog and by cigarette Mist feedback of the information is to described main controller module.

The flight control method of described four rotor unmanned aircrafts, mainly to be executed by described main controller module, specifically Comprise the following steps：

S10：Set up the kinetic model of four rotor unmanned aircrafts, the kinetics equation of unmanned vehicle is；

Wherein, if φ, θ, ψ are respectively roll angle, the angle of pitch and the yaw angle of four rotor unmanned aircrafts, l is its barycenter To the distance of rotor centers, Ix, Iy, Iz are inertia master away from Ω i is i-th rotor rotating speed, and Fi is the liter of i-th rotor generation Power, the lift that rotor produces is directly proportional to rotary-wing transmission velocity squared, and IR is rotary inertia, and n1 is its lift coefficient, and n2 is anti- Torque coefficient；

In order to the kinetics equation of four rotor unmanned aircrafts is converted into four independent control passages, define four rotors The control input of aircraft is

S20：The controlling party that design is combined with PID control based on the BP neural network that weighed value adjusting amount adds " Inertia " Method；

Wherein led to by four independent controls that the kinetics equation of four rotor unmanned aircrafts is converted in step slo Road is respectively by controller control, and four passages are respectively by height BPNNIPID, rolling BPNNIPID, pitching BPNNIPID, partially Boat BPNNIPID composition, the rotating speed adjusting four rotors through the conversion and control of controlled quentity controlled variable to reach gesture stability,

S21:If BP neural network haves three layers, including input layer, hidden layer and output layer, and wherein input layer j contains 4 neurons, hidden layer i contains 5 neurons, and output layer k contains 3 neurons；OrderRepresent connect l layer j-th Weights between neuron and i-th neuron of l+1 layer,

The then input of the input layer of BP neural network is：

O_j ⁽¹⁾=x (j), (j=1,2,3,4) (3)

It is the biasing of neuron elements including r (k), y (k), e (k), 1, Bias=1；

The input of hidden layer i is：

As i=1, have：

net₁ ⁽²⁾(k)=r (k) w₁₁ ⁽¹⁾+y(k)w₁₂ ⁽¹⁾+e(k)w₁₃ ⁽¹⁾+1·w₁₄ ⁽¹⁾(5)

Hidden layer i is output as：

The excitation function of hidden layer neuron uses the sigmoid function of Symmetrical：

The input of output layer k：

Output layer k is output as:

O_k ⁽³⁾(k)=g [net_k ⁽³⁾(k)] (9)

Wherein, k=1 in (8) (9), 2,3；

In the same manner,

Due to K in PID control_p, K_i, K_dNeed negated negative value, therefore output layer neuron excitation function uses non-negative Sigmoid function,

Desired output r (k) and reality output y (k) calculate performance indications：

S22:According to gradient descent method modified weight coefficient, by E (k), the negative gradient direction of weight coefficient is searched for and adjusts, Accelerate convergence by increasing Inertia, and calculate modified weight amount Δ w, revise weights：

α is referred to as inertia coeffeicent, and η is learning rate；Then

Unknown, useReplace, the error causing is compensated with regularized learning algorithm speed η, then by (1), (2), (8), (9), (10), (11), (12) can get：

So, partial gradient：

G ' ()=g (x) (1-g (x)) (22)

Learning method by output layer k weight coefficient：

Δw_ki ⁽²⁾(k)=a Δ w_ki ⁽²⁾(k-1)+ηδ_k ⁽³⁾(k)O_i ⁽²⁾(k) (23)

The learning method of hidden layer i weight coefficient：

Δw_ij ⁽¹⁾(k)=a Δ w_ij ⁽¹⁾(k-1)+ηδ_i ⁽²⁾(k)O_j ⁽¹⁾(k) (24)

So partial gradient

Wherein,

The optimal K required for PID control is so determined by above method_p、K_i、K_dParameter, it is achieved thereby that parameter is certainly Adjust.

By optimal K_p、K_i、K_dParameter is transported to the rotating speed that electron speed regulator adjusts four rotors with controlled motor.

The present invention four rotor unmanned aircrafts using based on weighed value adjusting amount add " Inertia " BP neural network with The control method that PID control combines can be adjusted K with the impact of external interference change, real-time update_p、K_i、K_dParameter, realizes Parameter self-tuning, solves the defect that traditional PID control is unable to real-time adaptive parameter adjustment, unmanned vehicle is better achieved Gesture stability under being in-flight interfered, improves vulnerability to jamming and the robustness of system.

Brief description

The above-mentioned and/or additional aspect of the present invention and advantage will become from reference to the description to embodiment for the accompanying drawings below Substantially and easy to understand, wherein：

Fig. 1 is the four rotor unmanned aircraft overall structure diagrams of the present invention.

Fig. 2 is that the four rotor unmanned aircraft main modular of the present invention constitute schematic diagram.

Fig. 3 is four rotor unmanned aircraft body axis systems and the inertial coodinate system schematic diagram of the present invention.

Fig. 4 is the BP neural network portion of additional " Inertia " in the four rotor unmanned aircraft attitude control methods of the present invention Separation structure schematic diagram.

Fig. 5 is that the BP neural network of four rotor unmanned aircrafts additional " Inertia " of the present invention is combined with PID control Control system architecture schematic diagram.

Fig. 6 is the four rotor unmanned aircraft control system architecture schematic diagrams of the present invention.

Fig. 7 is the four rotor unmanned aircraft height tracing figures of the present invention.

Fig. 8 is four rotor unmanned aircraft control methods and BP neural network PID control and the existing PID control of the present invention Method vulnerability to jamming test result comparison diagram.

Fig. 9 is the BP neural network PID control of additional " Inertia " in the four rotor unmanned aircraft control methods of the present invention Robustness test comparison figure processed.

Figure 10 is BP neural network PID control robustness test comparison in existing four rotor unmanned aircraft control methods Figure.

Figure 11 is traditional PID control robustness test comparison figure in existing four rotor unmanned aircraft control methods.

Specific embodiment

In order to be more clearly understood that the above objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and specifically real Mode of applying is further described in detail to the present invention.It should be noted that in the case of not conflicting, the enforcement of the application Feature in example and embodiment can be mutually combined.

Elaborate a lot of details in the following description in order to fully understand the present invention, but, the present invention also may be used To be implemented different from mode described here using other, therefore, protection scope of the present invention is not subject to following public tool The restriction of body embodiment.

Referring to Fig. 1-2, four rotor unmanned aircrafts of the embodiment of the present invention are further described.

As depicted in figs. 1 and 2, four rotor unmanned aircrafts 100 include body 10 and the main control being fixed on body 10 Device module 20, also includes four brushless electric machine control modules 60 being fixed on 10 4 cantilevers of body and is driven by brushless electric machine Rotor 70, in addition, as shown in Fig. 2 four rotor unmanned aircrafts also include being arranged on described body 10 and respectively with described Sensor assembly 40, navigation module 50 and water-cooled-air cooling module 80 that main controller module 20 connects, also include and master controller The wireless communication module 30 of 20 communication connections；Navigation module 50 adopts high-precision GPS satellite navigation system unmanned to four rotors Aircraft 100 is tracked positioning, and provides positional information to described main controller module 20 and navigate, and in navigation procedure Modification and solidification baud rate, it can in addition contain preserve the setting up procedure of baud rate；Described sensor assembly 40 include respectively with institute State Inertial Measurement Unit, baroceptor, electronic compass, Smoke Sensor and the air velocity transducer of main control module 20 connection, Described Inertial Measurement Unit is used for three axis accelerometers of sense aircraft, rolling angular speed, pitch rate, yawrate And course information, described baroceptor is for the height of sense aircraft, the heading device of described electronic compass measurement aircraft Breath, described air velocity transducer is monitored to the wind speed of aircraft location, and described Smoke Sensor is arranged on aircraft On PCB on for detect described circuit board break down generation smog and by smog feedback of the information to described master control Device module 20 processed；Described wireless communication module 30 includes remote control, PPM decoder and PPM receiver, described PPM encoder with Described remote control connects, and four channel control signals of remote control will be wirelessly transmitted to described PPM after PPM encoder encodes Receiver, described PPM receiver is connected with described main controller module；Described motor control module 60 is included for driving flight Four motors of four rotors 70 and the electron speed regulator controlling described four motors work respectively, described electronic speed regulation on device Device is connected to receive motor control signal with described main controller module, described Smoke Sensor be arranged in close on carry-on Circuit board annex with detect described circuit board break down generation smog and by smog feedback of the information to described master controller mould Block；Described water-cooled-air cooling module 80 can effectively reduce the heat producing when main controller module 20 and motor control module 60 work Amount.

Navigation module 50 can provide four rotor wing unmanned aerial vehicles current positional information, main controller module 20 be four rotors no The core of man-machine 100 control systems, its effect is responsible for gathering three axis accelerometers, the roll angle speed that sensor detects The attitude angular rate of the compositions such as rate, pitch rate and yawrate and course information real-time resolving, further according to detecting The flight information being sent by remote control, calculates actual output motor control signal to electron speed regulator, then electronic speed regulation Device controls the rotating speed of 4 motors according to the control signal obtaining, thus realizing the control of the lift and torque that 4 rotors are produced System, brushless electric machine can control its rotating speed by PWM and carry out thus reaching the size to power and moment produced by each rotor Control.

Specifically, described main controller module 20 and described motor control module 60 are respectively arranged with temperature element (figure In for illustrating), described temperature element is connected to realize temperature acquisition with described main controller module 20, described main controller module 20 adjust the cooling power of described water-cooled-air cooling module 80 according to collected temperature information.Both when temperature element detects Need during the temperature drift of main controller module 20 and described motor control module 60 to lift the cooling work(of water-cooled-air cooling module 80 Rate is to accelerate to cool, if reducing cooling work(when the temperature of main controller module 20 and described motor control module 60 is low Rate, so can ensure that main controller module 20 and described motor control module 60 are operated within the scope of suitable temperature.

Specifically, water-cooled-air cooling module is lowered the temperature first with water-cooling system, reaches and effective object temperature after water temperature raises When spending close, discharge all of water, now will be lowered the temperature using air cooling system.So it is effectively reduced main controller module 20 The temperature rise caused by heat producing when working with motor control module 60.

Four rotor unmanned aircrafts of the present embodiment, control signal is wirelessly sent to PPM by PWM mode and connects by remote control Receipts machine, PPM encoder transports to main controller module 20 by after the control signal decoding received by PPM receipts machine, meanwhile, constitutes four The height of rotor unmanned aircraft real-time attitude information, rolling, pitching, driftage by etc. recorded and transmitted to master by sensor assembly Controller module 20, output motor control signal after main controller module COMPREHENSIVE CALCULATING real-time attitude information and control signal information To electron speed regulator, then electron speed regulator controls the rotating speed of 4 motors according to the motor control signal obtaining, thus realize right The lift of 4 rotor generations and the control of torque, brushless electric machine can control its rotating speed by PWM thus reaching to each rotor The size of produced power and moment is controlled, thus realizing automatically adapting to external environment change, reaching and preferably controlling effect Really.

After the main control module COMPREHENSIVE CALCULATING real-time attitude information of described four rotor unmanned aircrafts and control signal information Output motor control signal is comprised the following steps with controlling the method for unmanned vehicle：

S10：Set up the kinetic model of four rotor unmanned aircrafts, the machine of four rotor unmanned aircrafts as shown in Figure 3 Body coordinate-system figure, the kinetics equation of unmanned vehicle is；

Wherein, if φ, θ, ψ are respectively roll angle, the angle of pitch and the yaw angle of four rotor unmanned aircrafts, l is its barycenter To the distance of rotor centers, Ix, Iy, Iz are inertia master away from Ω i is i-th rotor rotating speed, and Fi is the liter of i-th rotor generation Power, the lift that rotor produces is directly proportional to rotary-wing transmission velocity squared, and IR is rotary inertia, and n1 is its lift coefficient, and n2 is anti- Torque coefficient；

In order to the kinetics equation of four rotor unmanned aircrafts is converted into four independent control passages, define four rotors The control input of unmanned vehicle is

S20：The controlling party that design is combined with PID control based on the BP neural network that weighed value adjusting amount adds " Inertia " Method；

Wherein led to by four independent controls that the kinetics equation of four rotor unmanned aircrafts is converted in step slo Road is respectively by controller control, and four passages are respectively by height BPNNIPID, rolling BPNNIPID, pitching BPNNIPID, partially Boat BPNNIPID composition, the rotating speed adjusting four rotors through the conversion and control of controlled quentity controlled variable to reach gesture stability,

S21:The BP neural network part-structure schematic diagram of additional " Inertia " in attitude control method as shown in Figure 4, If BP neural network haves three layers, including input layer (input layer), hidden layer (hidden layer) and output layer (output layer), and wherein input layer j contains 4 neurons, hidden layer i contains 5 neurons, and output layer k contains 3 Neuron；OrderRepresent the weights between j-th neuron and i-th neuron of l+1 layer connecting l layer,

The then input of the input layer of BP neural network is：

O_j ⁽¹⁾=x (j), (j=1,2,3,4) (3)

It is the biasing of neuron elements including r (k), y (k), e (k), 1, Bias=1；

The input of hidden layer i is：

As i=1, have：

net₁ ⁽²⁾(k)=r (k) w₁₁ ⁽¹⁾+y(k)w₁₂ ⁽¹⁾+e(k)w₁₃ ⁽¹⁾+1·w₁₄ ⁽¹⁾(5)

Hidden layer i is output as：

The excitation function of hidden layer neuron uses the sigmoid function of Symmetrical：

The input of output layer k：

Output layer k is output as:

O_k ⁽³⁾(k)=g [net_k ⁽³⁾(k)] (9)

Wherein, k=1 in (8) (9), 2,3；

In the same manner,

Due to K in PID control_p, K_i, K_dNeed negated negative value, therefore output layer neuron excitation function uses non-negative Sigmoid function,

Desired output r (k) and reality output y (k) calculate performance indications：

S22:Control system control process schematic diagram as shown in Figure 5, the BP nerve net of additional " Inertia " shown in Fig. 6 Network is combined with PID control system control structure schematic diagram, and the control signal that remote control is sent passes according to unmanned vehicle Real-time attitude information detected by sensor module according to gradient descent method modified weight coefficient, by the negative ladder to weight coefficient for the E (k) The search of degree direction and adjustment, accelerate convergence by increasing Inertia, and calculate modified weight amount Δ w, revise weights：

α is referred to as inertia coeffeicent, and η is learning rate；Then

Unknown, useReplace, the error causing is compensated with regularized learning algorithm speed η, then by (1), (2), (8), (9), (10), (11), (12) can get：

So, partial gradient：

G ' ()=g (x) (1-g (x)) (22)

Learning method by output layer k weight coefficient：

Δw_ki ⁽²⁾(k)=a Δ w_ki ⁽²⁾(k-1)+ηδ_k ⁽³⁾(k)O_i ⁽²⁾(k) (23)

The learning method of hidden layer i weight coefficient：

Δw_ij ⁽¹⁾(k)=a Δ w_ij ⁽¹⁾(k-1)+ηδ_i ⁽²⁾(k)O_j ⁽¹⁾(k) (24)

So partial gradient

Wherein,

The optimal K required for PID control is so determined by above method_p、K_i、K_dParameter, it is achieved thereby that parameter is certainly Adjust.

In order to verify the control effect of four rotor unmanned aircrafts proposed by the present invention and its control method, using build Four rotor unmanned aircraft model machines are tested.Carry out multiple scheme experiments respectively, specific as follows：

Control performance contrast experiment：

Devise the control method that corresponding PID control method and BP neural network are combined with PID control, and prominent having In the environment of sending out crosswind, and BP neural network and the PID control adding " Inertia " based on weighed value adjusting amount proposed by the present invention Four rotor unmanned aircrafts under the control method combining controls carry out vulnerability to jamming and robustness contrast experiment.In an experiment, Complete first in the case of calm, the BP neural network of " Inertia " is added based on weighed value adjusting amount and PID control phase is tied The control method closed controls four lower rotor unmanned aircraft height tracings to test, and corresponding flight effect is as shown in fig. 7, wherein K during following the tracks of_p、K_i、K_dParameter automatic optimal；Subsequently in t=3s, t=6s and t=9s adds burst side fitful wind, base The control method combining with PID control in the BP neural network that weighed value adjusting amount adds " Inertia ", based on BP neural network Four rotor unmanned aircraft Immunity Performances under the control method and traditional PID control method that combine with PID control and robust Performance comparison, such as Fig. 8, shown in Fig. 9, Figure 10 and Figure 11.

From experiments it is evident that in the case of being not required to very important person's work PID, adding the BP nerve of " Inertia " Network can be with self-adaptative adjustment K_p、K_i、K_dParameter, searches out optimized parameter, and controlled device quickly reaches tracking desired value, realizes Parameter self-tuning；And be based on weighed value adjusting amount and add the control method that the BP neural network of " Inertia " is combined with PID control For comparing traditional PID control method, improve dynamic performance comprehensively.From figure 8, it is seen that the BP of additional " Inertia " Combine with the PID control vulnerability to jamming of control method of neutral net is better than traditional PID control method, and overshoot is less, adjusts The section time is shorter, and compared with BP neural network PID control, its rise time is shorter.Fig. 9, Figure 10 and Figure 11 show, in QUAV certainly In the case of body Parameters variation, under the BP neural network PID control of additional " Inertia ", robustness is marginally better than BP neural network PID Control, be better than traditional PID control.

To sum up, the control method being combined with PID control based on the BP neural network that weighed value adjusting amount adds " Inertia " Under QUAV gesture stability effect be better than BP neural network PID control, better than traditional PID control.

These are only the preferred embodiments of the present invention, be not limited to the present invention, for those skilled in the art For member, the present invention can have various modifications and variations.Within all creative spirit in the present invention and principle, that is made is any Modification, equivalent, improvement etc., should be included within the scope of the present invention.

Claims (5)

1. a kind of four rotor unmanned aircrafts, including body and main controller module, also include respectively with described master controller mould Navigation module that block connects, sensing module, communication module and motor control module are it is characterised in that described navigation module is using defending Star navigation system is positioned to aircraft and provides positional information to described main controller module and navigate, and in navigation procedure Middle modification and solidification baud rate；Inertial Measurement Unit that described sensor assembly includes being connected with described main control module respectively, Baroceptor, electronic compass and air velocity transducer, three axis that described Inertial Measurement Unit is used for sense aircraft accelerate Degree, rolling angular speed, pitch rate, yawrate and course information, described baroceptor is used for sense aircraft Highly, described electronic compass is used for measuring the course information of aircraft, and described air velocity transducer is to aircraft location Wind speed is monitored；Described wireless communication module includes remote control, PPM decoder and PPM receiver, described PPM encoder with Described remote control connects, and four channel control signals of remote control will be wirelessly transmitted to described PPM after PPM encoder encodes Receiver, described PPM receiver is connected with described main controller module；Described motor control module is included for driving aircraft Four motors of four rotors and the electron speed regulator, described electron speed regulator and the institute that control described four motors work respectively State main controller module to connect to receive motor control signal.

2. four rotor unmanned aircrafts as claimed in claim 1 are it is characterised in that also include water-cooled-air cooling module, described water Cold-air cooling module is connected with described main controller module and described motor control module respectively, with to described main controller module With described motor control module cooling, and accept simultaneously described main controller module control to realize to cooling power Adjust.

3. four rotor unmanned aircrafts as claimed in claim 2 are it is characterised in that described main controller module and described motor It is respectively arranged with temperature element, described temperature element is connected with described main controller module to be adopted to realize temperature in control module Collection, described main controller module adjusts the cooling power of described water-cooled-air cooling module according to collected temperature information.

4. four rotor unmanned aircrafts as claimed in claim 1 are it is characterised in that described sensor assembly includes and described master The Smoke Sensor module that control module connects, described Smoke Sensor is arranged in and closes on carry-on circuit board annex to visit Survey described circuit board break down generation smog and by smog feedback of the information to described main controller module.

5. a kind of flight control method of four rotor unmanned aircrafts, comprises the following steps：

S10：Set up the kinetic model of quadrotor, the kinetics equation of QUAV is；

Wherein, if φ, θ, ψ are respectively roll angle, the angle of pitch and the yaw angle of four rotor unmanned aircrafts, l is its barycenter to rotation The distance at wing center, Ix, Iy, Iz are inertia master away from Ω i is i-th rotor rotating speed, and Fi is the lift of i-th rotor generation, rotation The lift that the wing produces is directly proportional to rotary-wing transmission velocity squared, and IR is rotary inertia, and n1 is its lift coefficient, and n2 is reaction torque system Number；

In order to the kinetics equation of quadrotor is converted into four independent control passages, define quadrotor Control input is

$\left[\begin{array}{c}{U}_1\\ U_2\\ U_3\\ U_4\end{array}\right]=\left[\begin{array}{c}{F}_1+F_2+F_3+F_4\\ F_4-F_2\\ F_3-F_1\\ F_2+F_4-F_1-F_3\end{array}\right]=\left[\begin{array}{c}{n}_1\underset{i=1}{\overset{4}{\Σ}}{{\mathrm{\Ω}}_i}^2\\ n_1({{\mathrm{\Ω}}_4}^2-{{\mathrm{\Ω}}_2}^2)\\ n_1({{\mathrm{\Ω}}_3}^2-{{\mathrm{\Ω}}_1}^2)\\ n_2({{\mathrm{\Ω}}_1}^2-{{\mathrm{\Ω}}_2}^2+{{\mathrm{\Ω}}_3}^2-{{\mathrm{\Ω}}_4}^2)\end{array}\right]---\left(2\right)$

S20：The control method that design is combined with PID control based on the BP neural network that weighed value adjusting amount adds " Inertia "；

Wherein divided by four independent control passages that the kinetics equation of four rotor unmanned aircrafts is converted in step slo Not by controller control, and four passages are respectively by height BPNNIPID, rolling BPNNIPID, pitching BPNNIPID, driftage BPNNIPID forms, and the rotating speed adjusting four rotors through the conversion and control of controlled quentity controlled variable to reach gesture stability,

S21:If BP neural network haves three layers, including input layer, hidden layer and output layer, and wherein input layer j contains 4 Neuron, hidden layer i contains 5 neurons, and output layer k contains 3 neurons；OrderRepresent j-th nerve connecting l layer Weights between unit and i-th neuron of l+1 layer,

The then input of the input layer of BP neural network is：

O_j ⁽¹⁾=x (j), (j=1,2,3,4) (3)

It is the biasing of neuron elements including r (k), y (k), e (k), 1, Bias=1；

The input of hidden layer i is：

$\begin{array}{cc}{{\mathrm{net}}_i}^{\left(2\right)}\left(k\right)=\underset{j=1}{\overset{4}{\Σ}}{w_{ij}}^{\left(1\right)}{x}_j& (i=1,2,3,4,5)\end{array}---\left(4\right)$

As i=1, have：

net₁ ⁽²⁾(k)=r (k) w₁₁ ⁽¹⁾+y(k)w₁₂ ⁽¹⁾+e(k)w₁₃ ⁽¹⁾+1·w₁₄ ⁽¹⁾(5)

Hidden layer i is output as：

The excitation function of hidden layer neuron uses the sigmoid function of Symmetrical：

$f\left(x\right)=\tanh \left(x\right)=\frac{e^x-e^{-x}}{e^x+e^{-x}}\∈(-1,1)---\left(7\right)$

The input of output layer k：

${{\mathrm{net}}_k}^{\left(3\right)}\left(k\right)=\underset{i=1}{\overset{5}{\Σ}}{w_{ki}}^{\left(3\right)}{O_i}^{\left(2\right)}\left(k\right)---\left(8\right)$

Output layer k is output as:

O_k ⁽³⁾(k)=g [net_k ⁽³⁾(k)] (9)

Wherein, k=1 in (8) (9), 2,3；

$\begin{array}{c}{K}_p={O_1}^{\left(3\right)}\left(k\right)={O_1}^{\left(2\right)}\left(k\right){w_{11}}^{\left(2\right)}+{O_2}^{\left(2\right)}\left(k\right){w_{12}}^{\left(2\right)}+{O_3}^{\left(2\right)}\left(k\right){w_{13}}^{\left(2\right)}+{O_4}^{\left(2\right)}\left(k\right){w_{14}}^{\left(2\right)}\\ +{O_5}^{\left(2\right)}\left(k\right){w_{15}}^{\left(2\right)}=\underset{i=1}{\overset{5}{\Σ}}{O_i}^{\left(2\right)}\left(k\right){w_{1i}}^{\left(2\right)}\end{array}---\left(10\right)$

In the same manner,

$K_i={O_2}^{\left(3\right)}\left(k\right)=\underset{i=1}{\overset{5}{\Σ}}{O_i}^{\left(2\right)}\left(k\right){w_{2i}}^{\left(2\right)}---\left(11\right)$

$K_d={O_3}^{\left(3\right)}\left(k\right)=\underset{i=1}{\overset{5}{\Σ}}{O_i}^{\left(2\right)}\left(k\right){w_{3i}}^{\left(2\right)}---\left(12\right)$

Due to K in PID control_p, K_i, K_dNeed negated negative value, therefore output layer neuron excitation function uses the Sigmoi of non-negative Function,

$g\left(x\right)=\frac{1}{2}(1+\tanh (x\left)\right)=\frac{e^x}{e^x+e^{-x}}\∈(0,1)---\left(13\right)$

Desired output r (k) and reality output y (k) calculate performance indications：

$E\left(k\right)=\frac{1}{2}{\left(r\right(k)-y(k\left)\right)}^2=\frac{1}{2}{e}^2\left(k\right)---\left(14\right)$

S22:According to gradient descent method modified weight coefficient, by E (k), the negative gradient direction of weight coefficient is searched for and adjust, pass through Increase Inertia and accelerate convergence, and calculate modified weight amount Δ w, revise weights：

${{\mathrm{\Δw}}_{ki}}^{\left(2\right)}\left(k\right)={{\mathrm{\α\Δw}}_{ki}}^{\left(2\right)}(k-1)-\mathrm{\η}\frac{\∂E\left(k\right)}{\∂{w_{ki}}^{\left(2\right)}}---\left(15\right)$

α is referred to as inertia coeffeicent, and η is learning rate；Then

$\frac{\∂E\left(k\right)}{\∂{w_{ki}}^{\left(2\right)}}=\frac{\∂E\left(k\right)}{\∂y\left(k\right)}\·\frac{\∂y\left(k\right)}{\∂\mathrm{\Δ}u\left(k\right)}\·\frac{\∂\mathrm{\Δ}u\left(k\right)}{\∂{O_k}^{\left(3\right)}\left(k\right)}\·\frac{\∂{O_k}^{\left(3\right)}\left(k\right)}{\∂{{\mathrm{net}}_k}^{\left(3\right)}\left(k\right)}\·\frac{\∂{{\mathrm{net}}_k}^{\left(3\right)}\left(k\right)}{\∂{w_{ki}}^{\left(2\right)}\left(k\right)}---\left(16\right)$

$\frac{\∂{{\mathrm{net}}_k}^{\left(3\right)}\left(k\right)}{\∂{w_{ki}}^{\left(2\right)}\left(k\right)}={O_i}^{\left(2\right)}\left(k\right)---\left(17\right)$

Unknown, useReplace, the error causing compensates with regularized learning algorithm speed η, then by (1), (2), (8), (9), (10), (11), (12) can get：

$\frac{\∂\mathrm{\Δ}u\left(k\right)}{\∂{O_1}^{\left(3\right)}\left(k\right)}=e\left(k\right)-e(k-1)---\left(18\right)$

$\frac{\∂\mathrm{\Δ}u\left(k\right)}{\∂{O_2}^{\left(3\right)}\left(k\right)}=e\left(k\right)---\left(19\right)$

$\frac{\∂\mathrm{\Δ}u\left(k\right)}{\∂{O_3}^{\left(3\right)}\left(k\right)}=e\left(k\right)-2e(k-1)+e(k-2)---\left(20\right)$

So, partial gradient：

${{\mathrm{\δ}}_k}^{\left(3\right)}\left(k\right)=e\left(k\right)\mathrm{sgn}\left(\frac{\∂y\left(k\right)}{\∂\mathrm{\Δ}u\left(k\right)}\right)\frac{\∂\mathrm{\Δ}u\left(k\right)}{\∂{O_k}^{\left(3\right)}\left(k\right)}{g}^{\′}\left({{\mathrm{net}}_k}^{\left(3\right)}\right(k\left)\right),(k=1,2,3)---\left(21\right)$

G ' ()=g (x) (1-g (x)) (22)

Learning method by output layer k weight coefficient：

Δw_ki ⁽²⁾(k)=α Δ w_ki ⁽²⁾(k-1)+ηδ_k ⁽³⁾(k)O_i ⁽²⁾(k) (23)

The learning method of hidden layer i weight coefficient：

Δw_ij ⁽¹⁾(k)=α Δ w_ij ⁽¹⁾(k-1)+ηδ_i ⁽²⁾(k)O_j ⁽¹⁾(k) (24)

So partial gradient

${{\mathrm{\δ}}_i}^{\left(2\right)}\left(k\right)=f^{\′}\left({{\mathrm{net}}_i}^{\left(2\right)}\right(k\left)\right)\underset{k=1}{\overset{3}{\Σ}}{{\mathrm{\δ}}_k}^{\left(3\right)}\left(k\right){w_{ki}}^{\left(2\right)}\left(k\right),(i=1,2,3,4,5)---\left(25\right)$

Wherein,

$f^{\′}(\·)=\frac{1}{2}(1-f^2(x\left)\right)---\left(26\right)$

The optimal K required for PID control is so determined by above method_p, K_i, K_dParameter, it is achieved thereby that parameter is from whole Fixed.