CN113985900B - Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method - Google Patents
Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method Download PDFInfo
- Publication number
- CN113985900B CN113985900B CN202110889990.9A CN202110889990A CN113985900B CN 113985900 B CN113985900 B CN 113985900B CN 202110889990 A CN202110889990 A CN 202110889990A CN 113985900 B CN113985900 B CN 113985900B
- Authority
- CN
- China
- Prior art keywords
- model
- aerial vehicle
- unmanned aerial
- output
- rotor unmanned
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000000694 effects Effects 0.000 claims abstract description 22
- 230000008034 disappearance Effects 0.000 claims abstract description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 12
- 238000012937 correction Methods 0.000 claims description 7
- 230000001419 dependent effect Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 5
- 230000006978 adaptation Effects 0.000 claims description 3
- 230000007306 turnover Effects 0.000 claims description 3
- 238000005315 distribution function Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000000630 rising effect Effects 0.000 claims description 2
- 238000012512 characterization method Methods 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000002708 enhancing effect Effects 0.000 abstract description 2
- 238000004422 calculation algorithm Methods 0.000 description 16
- 230000006870 function Effects 0.000 description 16
- 238000012549 training Methods 0.000 description 7
- 230000003044 adaptive effect Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 210000002569 neuron Anatomy 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005312 nonlinear dynamic Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000003723 Smelting Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- G05D1/08—Control of attitude, i.e. control of roll, pitch, or yaw
- G05D1/0808—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft
- G05D1/0816—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability
- G05D1/0825—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability using mathematical models
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
According to the four-rotor unmanned aerial vehicle attitude dynamic characteristic model, three attitude angles of the four-rotor unmanned aerial vehicle are used as output, four rotor motor voltages of the four-rotor unmanned aerial vehicle are used as input, and a ReCNN-ARX dynamic characteristic model of the four-rotor unmanned aerial vehicle is built. The method has the beneficial effects that the method is used for modeling the gesture dynamic characteristics of the four-rotor unmanned aerial vehicle, combining with two models of the ReCNN and the SD-ARX, a nonlinear model ReCNN-ARX with high characterization capacity for describing the gesture dynamic characteristics of the four-rotor unmanned aerial vehicle is obtained, wherein the ReCNN can effectively solve the gradient disappearance problem, and compared with a linear ARX model, the prediction precision of the model is improved; according to the invention, the residual convolution network ReCNN is used for fitting the nonlinear coefficient of the SD-ARX model, and compared with a common convolution network, the obtained model is more stable; and the problem of small gradient is effectively solved, so that the model has the capability of expanding the network layer number and enhancing the nonlinear fitting effect.
Description
Technical Field
The invention relates to the field of attitude control of a four-rotor unmanned aerial vehicle, and provides a ReCNN-ARX model for describing the nonlinear dynamic characteristics of the attitude of the four-rotor unmanned aerial vehicle and an identification method thereof, wherein the nonlinear fitting capacity of a residual convolution neural network and the system characterization capacity of an SD-ARX model are combined, and at any moment, the attitude of an aircraft at the next moment is predicted according to the current and historical states of the four-rotor unmanned aerial vehicle, so that a model for predicting the future dynamic behavior of an object is provided for a prediction control method based on the model. Meanwhile, a four-rotor unmanned aerial vehicle attitude prediction control method based on a ReCNN-ARX model and sampling a self-adaptive softening factor is provided, the softening factor is associated with a control error, and the softening factor is adjusted in different control stages so as to obtain a better control effect.
Background
Currently, the four-rotor unmanned aerial vehicle is increasingly widely applied to actual life, and the following scenes are as follows: in a large metal smelting plant area, due to the working and management requirements, environmental parameters such as temperature, humidity and the like of each point of the plant area need to be monitored; the traditional method consumes a large amount of manpower, material resources and financial resources, has poor effect, can realize the work fast, conveniently and accurately by matching the four-rotor unmanned aerial vehicle with each instrument, and has very high requirements on the gesture control precision of the four-rotor unmanned aerial vehicle. The four-rotor unmanned aerial vehicle is used as a complex system, has the characteristics of rapidness, strong nonlinearity, strong coupling, multiple inputs and multiple outputs and the like, and brings difficulty to the design of a control algorithm. Prior to conducting a control study, it is often necessary to build a reasonable model of the controlled object in order to design the controller. The modeling modes are generally classified into mechanism modeling and experimental modeling. For a four-rotor unmanned plane, a physical model of the four-rotor unmanned plane is usually established through dynamic analysis; or a basic control method is applied to control experiments, identification data are obtained, and experimental modeling of the four-rotor unmanned aerial vehicle object is carried out according to the data. For the four-rotor unmanned plane object, the physical model obtained by using the mechanism modeling mode has considerable limitation because the accurate expression of the mechanism model is difficult to obtain and the related parameters are difficult to obtain accurately. The experimental modeling mode can obtain the model of the system by the method of inputting and outputting data of the system and sampling the identification of the system, the obtained model has stronger anti-interference capability, but the modeling effect is quite different by adopting different identification models. After the model of the four-rotor unmanned aerial vehicle object is obtained, the model can be used as a prediction model, and a prediction control algorithm is designed based on the prediction model. Because the actual output of the system cannot realize rapid change like the expected output, a fixed softening factor is often added in the traditional predictive control algorithm to soften the reference track so as to obtain a better control effect. However, it is sometimes difficult to achieve satisfactory results with predictive control using the fixed softening factor method.
Disclosure of Invention
Aiming at the experimental modeling problem of the four-rotor unmanned aerial vehicle, the invention provides a ReCNN-ARX model combining a convolutional neural network and an SD-ARX model, which is used for describing the dynamic change characteristic of the gesture of the four-rotor unmanned aerial vehicle. The model combines the nonlinear parameter fitting capability of a residual convolution network and the nonlinear system characterization capability of an SD-ARX model, and is a model with local linearization and global nonlinearity. The residual convolution network structure is flexible and variable, and compared with the common convolution network, the residual convolution network structure can better solve the gradient vanishing problem in modeling; the SD-ARX model can comprehensively characterize the nonlinear dynamic characteristics of the system by selecting historical state variables related to the system.
Aiming at the problem of the design of the four-rotor unmanned aerial vehicle attitude prediction control algorithm, the invention also provides a four-rotor unmanned aerial vehicle attitude prediction control algorithm based on the flexible factor self-adaption of a ReCNN-ARX model, which is used for linking the flexible factor with the deviation between the actual value and the expected value of an output signal and carrying out online correction on the flexible factor according to the difference of the deviation, thereby obtaining better control effect than the traditional prediction control algorithm adopting a fixed flexible factor.
The invention provides a four-rotor unmanned aerial vehicle attitude dynamic characteristic model and an identification method thereof, and a four-rotor unmanned aerial vehicle attitude self-adaptive softening factor predictive control algorithm based on the constructed characteristic model, so as to improve the predictive accuracy of the four-rotor unmanned aerial vehicle attitude dynamic characteristic model and the effect of predictive control.
In order to solve the technical problems, the invention adopts the following technical scheme: a ReCNN-ARX model of four rotor unmanned aerial vehicle gesture dynamic characteristics, the model expression is:
wherein y (t) represents the actual attitude angle output of the four-rotor unmanned aerial vehicle at the moment t, and is a pitch angle phi (t), a turnover angle theta (t) and a yaw angle phi (t) respectively; u (t) represents the voltage of four motors of the quadrotor at time t;is the state dependent offset at time t; ny, nu are the output and input orders of the model, respectively, < >>Is a state dependent coefficient matrix.
The state dependency coefficient of the ReCNN-ARX model of the quadrotor unmanned plane is calculated by the following ReCNN (residual convolution network) model:
wherein ,is an element of a state dependency coefficient matrix in a four-rotor unmanned aerial vehicle ReCNN-ARX model; h is a i (t) represents the output of the full link layer of the residual convolution network,>offset term coefficients, output term coefficients and offset parameters in input term coefficients in a four-rotor unmanned aerial vehicle ReCNN-ARX model are respectively represented; w (W) i 、b i The weight parameter and the bias parameter of the full-connection layer of the residual convolution network are represented, and g (·) represents the activation function of the full-connection layer; />A j-th feature map representing a l-th convolutional layer in a r-th residual block in an i-th residual convolutional network, M 1 、M 2 Respectively representing the number of convolution kernels of the first convolution layer and the third convolution layer and the number of convolution kernels of the second convolution layer in the residual block; />Representing the t moment, the one-dimensional vector obtained by flattening all output feature graphs of the nth residual block, and f (·) represents the activation function of the convolution layer,>a j-th feature map representing a r-th residual block,>is the input vector of the first residual block, i.e., the input vector of the residual convolution network input= [ y (t-1) T y(t-2) T … y(t-d) T ] T D is the dimension of the input vector.
The ReCNN-ARX model of the quadrotor unmanned aerial vehicle has the characteristic of remarkable modularization, wherein the residual convolution neural network part comprises the following modules:
(1) An input layer for receiving an input state vector input;
(2) The residual blocks are formed by the convolution layer combination, are used for carrying out residual convolution operation on the input vector input, calculate an output feature map through an activation function and weight parameters of each convolution layer, and serve as the input of the next residual block; during counter propagation, the residual block provides a bypass for gradient rising, and the gradient of the output layer is uploaded to be close to the input layer, so that the parameters of the input layer can be updated normally, and the problem of gradient disappearance is avoided;
(3) The full-connection layer is used for processing the flattened characteristic diagram;
(4) The output layer, which can be regarded as a fully connected layer without an activation function, performs only linear operations. And the output layer linearly combines the characteristics output by the full-connection layer to obtain the state dependent coefficient of the ReCNN-ARX model.
The identification method of the ReCNN-ARX model of the quadrotor unmanned aerial vehicle comprises the following steps:
(a) Acquiring output/input data of the quadrotor unmanned aerial vehicle as identification data of a ReCNN-ARX model;
(b) The output/input variable orders ny, nu of the ReCNN-ARX model are selected,structural parameters M of the model 1 、 M 2 、n；
(c) Assigning an initial value to the model parameter;
(d) Performing forward operation on the model to obtain the predicted output of a four-rotor unmanned aerial vehicle ReCNN-ARX model, and calculating MSE (Mean Square Error ) between the predicted output and the expected output as a loss function;
(e) Calculating a counter-propagating gradient according to the loss function, and updating parameters reversely from an output layer to an input layer;
(f) Repeating steps (d) - (e) until optimal parameters of the model are found;
(g) Selecting other model input/output orders and model structure parameters, repeating the steps (b) - (f), and finding out the model order and the model structure parameters with better model prediction effect under the condition of meeting the real-time requirement of the system.
Compared with the traditional CNN-ARX model, the four-rotor unmanned aerial vehicle ReCNN-ARX model established by the invention can effectively solve the problem of gradient disappearance and avoid the phenomenon of model prediction precision reduction when the network layer number is deepened. The ReCNN-ARX model is not only more stable than the CNN-ARX model, but also provides an expansion capability to deepen the number of layers of the neural network within the allowable computational effort.
In order to solve the problem of attitude control of a four-rotor unmanned aerial vehicle, the invention adopts the following technical scheme: a four-rotor unmanned aerial vehicle attitude self-adaptive soft prediction control method. Compared with the traditional predictive control algorithm, the adaptive soft predictive control algorithm provided by the invention can enable the magnitude of the soft factor following error in the controller to be adaptively adjusted. When the error is larger, the system responds quickly, and the actual output of the system is quickly closed to the target output; when the error is small, the actual output smoothly approaches to the expected value, the system stability is maintained, and the overshoot of the system is reduced; when the actual output of the system is similar to the expected value, the softening factor of the system is further increased, the influence of micro-perturbation and interference on the system is avoided, and the anti-interference capability of the system is increased. Including the following control quantity optimization strategies and softening factor adaptation laws.
The four-rotor unmanned aerial vehicle attitude control amount optimizing strategy is as follows:
in the formula ,Q≥0,R 1 ＞0,R 2 > 0 is a weighting matrix; />Fixing regression coefficients of a four-rotor unmanned aerial vehicle ReCNN-ARX model at a time t, and recursively obtaining an output pre-sequencing column based on the four-rotor unmanned aerial vehicle ReCNN-ARX model; />The method is to output expected value sequences after being modified by the self-adaptive softening factors; u (t) is the control quantity sequence to be optimized, deltaU (t) is the control increment sequence, U * (t) is the control amount sequence obtained after optimization, u therein * (t) for control at time t; n (N) y Is the predicted time domain length, N u Is the control time domain length.
The adaptive law of the softening factors in the attitude prediction control law of the four-rotor unmanned aerial vehicle is as follows:
where y (t) is an actual output value of the quadrotor unmanned aerial vehicle at the current time t, e (t) represents a deviation between an output expected value and the actual value,is the expected output after self-adaptive softening, y r Is the desired output, α is the softening factor; alpha 0 Is the initial value of the softening factor, mu and sigma are the mean and variance of the normal distribution function, a is the scaling factor, b is the deviation term, f e (e (t)) represents a softening factor correction function based on the error variation.
The softening factor self-adaptive law parameter setting method comprises the following steps:
(1) Firstly, four-rotor unmanned aerial vehicle attitude prediction control with fixed softening factors is used, and an initial value alpha of the softening factors with good control effect is found between [0,1 ] 0 ；
(2) Then, the four-rotor unmanned aerial vehicle attitude prediction control of the self-adaptive softening factor is sampled, and proper correction directions and proper forces are selected according to the characteristics and control results of the system, so that proper parameters a and b are found;
(3) And (3) carrying out fine adjustment on the error tolerance range according to the gesture prediction control effect of the four-rotor unmanned aerial vehicle, namely searching for proper parameters mu and sigma.
The specific process of the self-adaptive law parameter setting comprises the following steps:
(1) Taking a=0, i.e. without taking adaptation into account, the initial value α of the softening factor is adjusted 0 A relatively good control effect is obtained;
(2) Mu=0 and sigma=1 are taken, and fine tuning is performed on the two parameters a and b on the basis of the mu=0 and the sigma=1. Taking a positive acting system as an example, if the dynamic response of the system is slower, a is larger than 0, b is smaller than 0, when the set value changes, the system error e (t) is increased, and the system deviates farther from the steady state. At this time, correction coefficient f e (e (k)). Apprxeq.ab, so α < α 0 The dynamic response speed of the system is accelerated; as the system approaches a steady state condition,if the parameters are chosen appropriately, then alpha > alpha 0 The control action of the system slows down to reduce overshoot. If the dynamic response of the system is faster, a > 0 and b > 0 can be selected, and analysis is the same as above;
(3) Finally, setting the parameters mu and sigmaMu determines f e (e (k)) if μ > 0, the positive error is more tolerant, and even if the positive error is slightly larger, the system is considered to be in steady state; if μ < 0, the contrary is usually the case, μ=0. The parameter σ determines the sensitivity range of the steady state error, and it can be considered that the error goes into steady state at μ±σ. The control effect can be finely adjusted through mu and sigma so as to achieve the optimal effect.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, aiming at the modeling of the gesture dynamic characteristics of the four-rotor unmanned aerial vehicle, a nonlinear model ReCNN-ARX with strong characterization capability for describing the gesture dynamic characteristics of the four-rotor unmanned aerial vehicle is obtained by combining the two models of ReCNN and SD-ARX, wherein the ReCNN can effectively solve the problem of gradient disappearance, and compared with a linear ARX model, the prediction precision of the model is improved.
2. According to the invention, the residual convolution network ReCNN is used for fitting the nonlinear coefficient of the SD-ARX model, and compared with a common convolution network, the obtained model is more stable; the problem of small gradient is effectively solved, so that the model has the capability of expanding the network layer number and enhancing the nonlinear fitting effect;
3. the ReCNN-ARX model and the identification method thereof provided by the invention can also be applied to modeling of other multi-input multi-output complex nonlinear systems similar to a four-rotor unmanned aerial vehicle;
4. the four-rotor unmanned aerial vehicle gesture self-adaptive flexible predictive control algorithm improves the traditional model-based predictive control algorithm, and better control effect can be obtained after parameters in the flexible factor self-adaptive law are set compared with the traditional predictive control algorithm using the fixed flexible factor.
Drawings
FIG. 1 is a block diagram of a ReCNN-ARX model of a quad-rotor unmanned helicopter system of the present invention;
FIG. 2 is a flow chart of the ReCNN-ARX model identification of the quad-rotor unmanned helicopter system of the present invention.
Detailed Description
Aiming at the four-rotor unmanned aerial vehicle, the voltage of four rotor motors of the four-rotor unmanned aerial vehicle is used as the model input, three attitude angles of a depression angle, a turnover angle and a yaw angle of an aircraft are used as the output of the model, and a four-input three-output ReCNN-ARX model is established.
The initial input order nu, output order ny, and state vector dimension d of the model are first determined. Each output dimension corresponds to one residual convolutional neural network, i.e., q networks in total. Each residual convolution neural network comprises an input layer, and the number of the nerve elements of the input layer is consistent with the length of an input state vector input and is 3 multiplied by d; can also be regarded as the input of the first residual block
Secondly, three cascaded residual blocks, namely the output of the last residual block is used as the input of the next residual block, each residual block is composed of three convolution layers, the number of convolution kernels of each layer is respectively 32, 64 and 32, the convolution kernels are 3 multiplied by 3, and the structure and the operation process can be represented by the following formula by adopting a post-activation calculation mode:
wherein ,the output of the r-1 th residual block at time t is also the input of the r-1 th residual block. The input is subjected to the operation of a first convolution layer to obtain an output characteristic diagram of the first convolution layer,/I>Representing a jth output characteristic diagram of a first convolution layer in an ith residual convolutional neural network (Rth residual block) at a t moment; />Parameters representing the mth convolution kernel corresponding to the jth feature map of the ith residual convolution neural network,/->Is the corresponding bias parameter; f (·) is the activation function of the convolutional layer, where the tanh activation function is chosen and can be represented by the following formula:
where x is an independent variable and f (x) is an independent variable.
The output profile of the last residual block is flattened and then serves as the input of the full-connected layer, the output of which serves as the input of the output layer, and the full-connected layer has 32 neurons in total. The calculation of the full connection layer can be expressed by the following formula:
wherein ,is a one-dimensional vector obtained by flattening the input of the last residual block, and the vector is matched with the parameter W of the full connection layer i Multiplying and adding bias b of full connection layer i The input of the last output layer can be obtained. g (·) is the activation function of the fully connected layer, and the tanh activation function is also chosen.
And finally, an output layer, wherein the output layer does not activate a function, only carries out linear combination operation on the input of the layer, and the number of neurons of the output layer is the number of state coefficients corresponding to the output dimension in the SD-ARX model, and the number of the neurons is 4 Xnu+3 Xny.
Corresponding model structures are built in the python programming language and the tensorflow+keras framework, an Xavier parameter initialization mode of generating random numbers according to a normal distribution mode is used for generating initial weight parameters and bias in a residual convolution neural network, input and output data are arranged into sample sets according to corresponding orders, and the sample sets are divided into training sets and test sets according to the proportion of 0.75:0.25.
The model is first trained. After initializing a model, performing forward operation, and calculating a current loss function; the parameters of each layer of neurons are then updated by back-propagation according to the gradient of the loss function. The training algorithm of the model selects Adam (Adaptive Moment Estimation, adaptive estimated) algorithm. The principle can be represented by the following formula:
in the formula ,Θk Is the kth iteration update of the parameter to be updated Θ, ζ is the set learning rate, ΔΘ represents the gradient of the parameter to be updated,represents the partial derivative of Θ, f (x i ；Θ k-1 ) Represents the kth update, the xth i Forward calculation of the samples, y i The label representing the sample, i.e. the actual output at the input, L (·) represents the loss function, where the loss function takes the MSE (Mean Square Error ), β 1 、β 2 Is a given attenuation parameter and satisfies beta 1 ,β 2 E (0, 1), s, r are two variables that control the cumulative change of the gradient, and δ is a small constant that prevents the denominator from being 0. Adam's algorithm calculates the moving average of the gradient by s and r variables, again by β 1 And beta 2 Both parameters control the decay rate of the moving average. In actual use, the learning rate ζ=0.0001 is taken, and the super parameter β is calculated 1 ＝0.9,β 2 =0.999. Model parameters are updated in back propagation by using an Adam algorithm until the model converges, i.e., the MSE value of the model training set is stable or the decrease amplitude is very small.
And predicting the test set by using the trained model, calculating whether the MSE value of the prediction error of the test set is close to the training set, and retraining after adjusting the super-parameters or the network structure if the MSE value is obviously larger than the training set or smaller than the training set and does not belong to the normal phenomenon. If the two are close, the model can be proved to be converged to a better solution, an error distribution histogram and a residual error map of the test set are further drawn, and whether the error distribution accords with normal distribution or not is observed; if the model does not accord with normal distribution, the model has a good effect on the training set, and each state of the quadrotor aircraft cannot be predicted; if the model basically accords with normal distribution, the description model has certain system characterization capability.
The input order nu, the output order ny and the state vector order d of the model are adjusted, and the steps are repeated to find the order with smaller fitting error and faster calculation speed.
After a trained ReCNN-ARX model is obtained, a corresponding self-adaptive soft prediction controller is designed, and parameter setting of a self-adaptive law is carried out through repeated tests.
Let a=0 and b=0 first, find the softening factor initial value alpha with better control effect without using adaptive law 0 ；
Secondly, mu=0 and sigma=1, selecting proper softening factor correction direction and strength according to actual control conditions, and searching proper values of a and b through repeated tests;
and finally, carrying out parameter fine adjustment on mu and sigma to obtain the optimal control effect.
By combining the advantages of the ReCNN and the SD-ARX, compared with a linear ARX model or a common CNN-ARX modeling method, the ReCNN-ARX model improves the prediction precision of the attitude dynamic characteristics of the four-rotor unmanned aerial vehicle, enhances the stability of the model, and gives the model the capability of expanding the network layer number of the model within a real-time allowed range; the four-rotor unmanned aerial vehicle attitude prediction control method based on the ReCNN-ARX model, which is adaptive to the error of the softening factor, achieves better control effect than the traditional prediction control method.
It should be understood by those skilled in the art that the protection scheme of the present invention is not limited to the above embodiments, and various arrangements and modifications can be made on the basis of the above embodiments, and various modifications of the present invention fall within the protection scope of the present invention without departing from the spirit of the present invention.
Claims (5)
1. The four-rotor unmanned aerial vehicle attitude dynamic characteristic model is characterized in that three attitude angles of the four-rotor unmanned aerial vehicle are used as output, four rotor motor voltages of the four-rotor unmanned aerial vehicle are used as input, and a ReCNN-ARX dynamic characteristic model of the four-rotor unmanned aerial vehicle is established, wherein the expression is as follows:
wherein y (t) represents the actual attitude angle output of the four-rotor unmanned aerial vehicle at the moment t, and is a pitch angle phi (t), a turnover angle theta (t) and a yaw angle phi (t) respectively; u (t) represents the voltage of four motors of the quadrotor at time t;is the state dependent offset at time t; ny, nu are the output and input orders of the model, respectively, < >>Is a state dependent coefficient matrix;
the state dependent coefficient of the ReCNN-ARX dynamic characteristic model is obtained through calculation of a ReCNN residual convolution network model, and the model is as follows:
wherein ,is the state dependent coefficient moment in the ReCNN-ARX modelMatrix element; h is a i (t) represents the output of the full link layer of the residual convolution network,>respectively representing offset term coefficients, output term coefficients and offset parameters in input term coefficients in a ReCNN-ARX model; w (W) i 、b i The weight parameter and the bias parameter of the full-connection layer of the residual convolution network are represented, and g (·) represents the activation function of the full-connection layer; />A j-th feature map representing a l-th convolutional layer in a r-th residual block in an i-th residual convolutional network, M 1 、M 2 Respectively representing the number of convolution kernels of the first convolution layer and the third convolution layer and the number of convolution kernels of the second convolution layer in the residual block; />Representing the t moment, the one-dimensional vector obtained by flattening all output feature graphs of the nth residual block, and f (·) represents the activation function of the convolution layer,>a j-th feature map representing a r-th residual block,>is the input vector of the first residual block, i.e., the input vector of the residual convolution network, input= [ y (t-1) T y(t-2) T … y(t-d) T ] T D is the dimension of the input vector.
2. The four-rotor unmanned aerial vehicle attitude dynamic characteristic model according to claim 1, wherein the residual convolution network comprises the following modules:
(1) An input layer for receiving an input state vector input;
(2) The residual blocks are formed by the convolution layer combination, are used for carrying out residual convolution operation on the input vector input, calculate an output feature map through an activation function and weight parameters of each convolution layer, and serve as the input of the next residual block; during counter propagation, the residual block provides a bypass for gradient rising, and the gradient of the output layer is uploaded to be close to the input layer, so that the parameters of the input layer can be updated normally, and the problem of gradient disappearance is avoided;
(3) The full-connection layer is used for processing the flattened characteristic diagram;
(4) The output layer can be regarded as a full-connection layer without an activation function and only carries out linear operation, and the output layer obtains the state dependence coefficient of the ReCNN-ARX model after the characteristics output by the full-connection layer are combined linearly.
3. A method for identifying a four-rotor unmanned aerial vehicle attitude dynamic characteristic model comprises the following steps:
(S1) acquiring output/input data of the quadrotor unmanned aerial vehicle as identification data of the ReCNN-ARX model as set forth in claim 1;
(S2) selecting the output/input variable orders ny, nu of the ReCNN-ARX model and the structural parameters M of the model 1 、M 2 、n；
(S3) giving an initial value to the model parameter;
(S4) performing forward operation on the model to obtain the predicted output of the ReCNN-ARX model of the quadrotor unmanned aerial vehicle, and calculating MSE mean square error between the predicted output and the expected output as a loss function;
(S5) calculating a counter-propagating gradient according to the loss function, and updating parameters reversely from an output layer to an input layer;
(S6) repeating the steps (S4) - (S5) until the optimal parameters of the model are found;
and (S7) selecting other model input/output orders and model structure parameters, and repeating the steps (S2) - (S6) to find out the model orders and the model structure parameters with better model prediction effect under the condition of meeting the real-time requirement of the system.
4. The four-rotor unmanned aerial vehicle gesture self-adaptive softening prediction control method is characterized by comprising a control quantity optimization strategy and a softening factor self-adaptive law, wherein the control quantity optimization strategy is as follows:
in the formula ,Q≥0,R 1 ＞0,R 2 > 0 is a weighting matrix; />Is the regression coefficient of the ReCNN-ARX model according to claim 1 at time t, the output prediction sequence obtained based on recursion of the ReCNN-ARX model of a quadrotor unmanned aerial vehicle, ">The method is to output expected value sequences after being modified by the self-adaptive softening factors; u (t) is the control quantity sequence to be optimized, deltaU (t) is the control increment sequence, U * (t) is the control amount sequence obtained after optimization, u therein * (t) control for time t, N y Is the predicted time domain length, N u Is to control the time domain length;
the flexibility factor adaptation law is as follows:
where y (t) is an actual output value of the quadrotor unmanned aerial vehicle at the current time t, e (t) represents a deviation between an output expected value and the actual value,is the expected output after self-adaptive softening, y r Is the desired output, α is the softening factor; alpha 0 Is the initial value of the softening factor, mu and sigma are the mean and variance of the normal distribution function, a is the scaling factor, b is the deviation term, f e (e (t)) represents a softening factor correction function based on the error variation.
5. The four-rotor unmanned aerial vehicle gesture self-adaptive flexible prediction control method as claimed in claim 4, wherein the parameter setting method of the flexible factor self-adaptive law comprises the following steps:
(1) Firstly, four-rotor unmanned aerial vehicle attitude prediction control with fixed softening factors is used, and an initial value alpha of the softening factors with good control effect is found between [0,1 ] 0 ；
(2) Then, the four-rotor unmanned aerial vehicle attitude prediction control of the self-adaptive softening factor is sampled, and proper correction directions and proper forces are selected according to the characteristics and control results of the system, so that proper parameters a and b are found;
(3) And (3) carrying out fine adjustment on the error tolerance range according to the gesture prediction control effect of the four-rotor unmanned aerial vehicle, namely searching for proper parameters mu and sigma.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110889990.9A CN113985900B (en) | 2021-08-04 | 2021-08-04 | Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110889990.9A CN113985900B (en) | 2021-08-04 | 2021-08-04 | Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113985900A CN113985900A (en) | 2022-01-28 |
| CN113985900B true CN113985900B (en) | 2023-09-08 |
Family
ID=79735137
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110889990.9A Active CN113985900B (en) | 2021-08-04 | 2021-08-04 | Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113985900B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013025521A (en) * | 2011-07-20 | 2013-02-04 | Kyushu Univ | Operation support device for vehicle |
| CN107065902A (en) * | 2017-01-18 | 2017-08-18 | 中南大学 | UAV Attitude fuzzy adaptive predictive control method and system based on nonlinear model |
| CN109885077A (en) * | 2019-03-21 | 2019-06-14 | 中南大学 | A kind of quadrotor attitude control method and controller |
| CN112036513A (en) * | 2020-11-04 | 2020-12-04 | 成都考拉悠然科技有限公司 | Image anomaly detection method based on memory-enhanced potential spatial autoregression |
| CN112947532A (en) * | 2021-04-13 | 2021-06-11 | 中国计量大学 | Height control method of plant protection quad-rotor unmanned aerial vehicle based on liquid medicine measurement |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2924832B1 (en) * | 2007-12-11 | 2010-11-19 | Airbus France | METHOD AND APPARATUS FOR GENERATING A CONTROLLED SPEED FOR AN AIRCRAFT RUNNING ON THE GROUND |
| US11705226B2 (en) * | 2019-09-19 | 2023-07-18 | Tempus Labs, Inc. | Data based cancer research and treatment systems and methods |
-
2021
- 2021-08-04 CN CN202110889990.9A patent/CN113985900B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013025521A (en) * | 2011-07-20 | 2013-02-04 | Kyushu Univ | Operation support device for vehicle |
| CN107065902A (en) * | 2017-01-18 | 2017-08-18 | 中南大学 | UAV Attitude fuzzy adaptive predictive control method and system based on nonlinear model |
| CN109885077A (en) * | 2019-03-21 | 2019-06-14 | 中南大学 | A kind of quadrotor attitude control method and controller |
| CN112036513A (en) * | 2020-11-04 | 2020-12-04 | 成都考拉悠然科技有限公司 | Image anomaly detection method based on memory-enhanced potential spatial autoregression |
| CN112947532A (en) * | 2021-04-13 | 2021-06-11 | 中国计量大学 | Height control method of plant protection quad-rotor unmanned aerial vehicle based on liquid medicine measurement |
Non-Patent Citations (1)
| Title |
|---|
| 一种基于级联神经网络的飞机检测方法;王晓林，等;《智能系统学报》;第15卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113985900A (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110647042B (en) | Robot robust learning prediction control method based on data driving | |
| Wu et al. | Adaptive fuzzy control for full states constrained systems with nonstrict-feedback form and unknown nonlinear dead zone | |
| Zhou et al. | Deep convolutional neural network based fractional-order terminal sliding-mode control for robotic manipulators | |
| Shirzadeh et al. | Trajectory tracking of a quadrotor using a robust adaptive type-2 fuzzy neural controller optimized by cuckoo algorithm | |
| CN110009528B (en) | Parameter self-adaptive updating method based on optimal structure multi-dimensional Taylor network | |
| Rego et al. | Lyapunov-based continuous-time nonlinear control using deep neural network applied to underactuated systems | |
| Çetin et al. | Adaptive uncertainty compensation-based nonlinear model predictive control with real-time applications | |
| Kusumoputro et al. | System identification and control of pressure process rig system using Backpropagation Neural Networks | |
| Emami et al. | Multimodel ELM-based identification of an aircraft dynamics in the entire flight envelope | |
| CN111930010A (en) | LSTM network-based general MFA controller design method | |
| Moumouh et al. | A Novel Tuning approach for MPC parameters based on Artificial Neural Network | |
| Chen et al. | Novel adaptive neural networks control with event-triggered for uncertain nonlinear system | |
| Jumanov et al. | Mechanisms for optimizing the error control of micro-object images based on hybrid neural network models | |
| CN113985900B (en) | Four-rotor unmanned aerial vehicle attitude dynamic characteristic model, identification method and self-adaptive soft prediction control method | |
| CN105677936B (en) | The adaptive recurrence multistep forecasting method of electromechanical combined transmission system demand torque | |
| Zhou et al. | Launch vehicle adaptive flight control with incremental model based heuristic dynamic programming | |
| CN116880184A (en) | Unmanned ship track tracking prediction control method, unmanned ship track tracking prediction control system and storage medium | |
| CN111624872A (en) | PID controller parameter setting method and system based on self-adaptive dynamic programming | |
| CN111522240B (en) | MTMLP-ARX model of four-rotor aircraft, identification method, system and storage medium | |
| CN112346342B (en) | Single-network self-adaptive evaluation design method of non-affine dynamic system | |
| Yuqi et al. | Time-varying parameters estimation with adaptive neural network EKF for missile-dual control system | |
| Ye et al. | Online optimal control of discrete-time systems based on globalized dual heuristic programming with eligibility traces | |
| Inanc et al. | Long short-term memory for improved transients in neural network adaptive control | |
| CN112379601A (en) | MFA control system design method based on industrial process | |
| Alshareefi et al. | Real-time implementation of adaptive neural controller by LabVIEW software |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |