Unmanned ship parameter online identification method based on adaptive unscented Kalman filter

Abstract

The application discloses an unmanned ship parameter online identification method based on adaptive unscented Kalman filtering, which comprises the following steps: S1, constructing an unmanned ship model to be identified according to an unmanned ship dynamic equation; S2, taking the unmanned ship model parameters to be identified as augmented states, and expanding the unmanned ship model to be identified into an augmented model; S3, using adaptive unscented Kalman filtering to online estimate the augmented states, and obtaining the unmanned ship model parameters. Step S3 comprises the following steps: S31, initializing the adaptive unscented Kalman filtering; S32, obtaining input and output data of the augmented model; S33, updating covariance matrices of the augmented states and augmented state errors; S34, adaptively estimating a process noise covariance matrix; S35, predicting the covariance matrices of the augmented states and augmented state errors; S36, judging whether to terminate the online identification, and returning to step S32 if not. The application can solve the state estimation problem of the unmanned ship under the interference of unmodeled dynamics and unknown noise, and the unknown parameter identification precision is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic control, and in particular to an unmanned ship parameter online identification method based on adaptive unscented Kalman filtering technology. BACKGROUND

[0002] The process of extracting the parameters required by the control system from the data containing noise is called parameter estimation, and online estimation refers to the process of completing parameter estimation based on real-time observation data. Online estimation can update system parameters in real time during system operation, and is particularly suitable for controller design problems under uncertain system models, and has become an important technology in the field of automatic control.

[0003] With the development and utilization of the ocean, unmanned ships have become an important tool for exploring the ocean, and have been widely used in military and civilian fields, such as ocean resource exploration, environmental protection, search, and rescue. In recent years, researchers have done a lot of research on parameter estimation of unmanned ships, and have good results when applied to simplified linear models. However, the use of simplified models results in fewer estimated parameters and larger model errors. Some methods using neural networks cannot explicitly obtain the parameters of the nonlinear part due to the non-interpretable nature of neural networks, and can only obtain the function mapping relationship fitted by the neural network. SUMMARY

[0004] The present application aims to provide an unmanned ship parameter online identification method based on adaptive unscented Kalman filtering technology to solve the problem of poor parameter identification accuracy of existing methods for unmanned ships.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] An unmanned ship parameter online identification method based on adaptive unscented Kalman filtering, comprising the following steps:

[0007] S1, constructing an unmanned ship model to be identified according to the dynamics equation of the unmanned ship;

[0008] S2, expanding the unmanned ship model to be identified into an augmented model by taking the parameters of the unmanned ship model to be identified as augmented states;

[0009] S3, using adaptive unscented Kalman filtering to online estimate the augmented states and obtain the parameters of the unmanned ship model.

[0010] Further, the step S3 comprises:

[0011] S31, initializing the adaptive unscented Kalman filter;

[0012] S32, obtaining the input and output data of the augmented model;

[0013] S33. Update the covariance matrix of the augmented state and the augmented state error;

[0014] S34. Adaptive estimation of process noise covariance matrix;

[0015] S35, the covariance matrix of the predicted augmented state and the error of the augmented state;

[0016] S36. Determine whether to terminate online identification. If not, return to step S32.

[0017] 10. Further, step S1 includes:

[0018] S11. Based on the unmanned vessel dynamics equations, the unmanned vessel model can be expressed as:

[0019]

[0020]

[0021] Where, η=[x w ,y w ,ψ] T x w ,y w Let ψ represent the position of the unmanned surface vessel's center of mass, ψ represent the yaw angle of the unmanned surface vessel, and R(ψ) represent the rotation matrix. Let M represent the water flow velocity, assumed to be constant, C(υ) be the inertia matrix, D be the Coriolis force matrix, B be the damping matrix, and τ = [τ u ,τ r ] T These are the inputs to the unmanned surface vessel system, consisting of forward thrust and rudder angle.

[0022] The rotation matrix R(ψ) and the inertia matrix M are expressed as follows:

[0023]

[0024] The Coriolis force matrix C(υ), damping matrix D, and thrust distribution matrix B are respectively expressed as:

[0025]

[0026] S12. The unmanned vessel model is further represented as follows:

[0027]

[0028] Where a = [a1, a2, ..., a8] T This represents the parameter to be identified.

[0029] Further, step S2 includes:

[0030] S21, selecting an augmented state The unmanned ship augmented model is represented as

[0031] wherein v c = [V x , V y ] T represents the water flow velocity;

[0032]

[0033] S22, considering the process noise w(k) and the measurement noise v(k) at the kth moment, discretizing the unmanned ship augmented model, represented as:

[0034] x(k+1) = F(x(k), τ(k)) + w(k);

[0035] y(x) = h(x) + v(k);

[0036] wherein:

[0037] F(x(k), τ(k)) = x(k) + (f0(x(k)) + [f1(x(k)) f2(x(k))]τ(k))Δt; Δt is the sampling period.

[0038] Further, the step S31 comprises: given the initial predicted augmented state and the initial augmented state error covariance matrix P(0|-1).

[0039] Further, the step S32 comprises: inputting data as the input signal data τ(k) of the unmanned ship system obtained in real time in the periodic motion of the unmanned ship; outputting data as the real-time position x w , y w and the yaw angle ψ of the unmanned ship obtained by the real-time differential positioning, and the real-time acceleration and the yaw angular velocity r of the unmanned ship obtained by the inertial measurement unit as the measurement values, i.e.

[0040] Further, the step S33 comprises:

[0041] (1) generating the σ sample point χ (i) at the kth moment:

[0042]

[0043]

[0044]

[0045] Where n represents the dimension of the state vector, P(k|k-1) and P(k|k-1) are the covariance matrices of the augmented state and the augmented state error predicted at time k using the observations at time k-1, respectively, where c = α 2 (n+κ) represents the coefficient. Let i represent the i-th column of the lower triangular decomposition of cP(k|k-1);

[0046] (2) Calculate the measurement value for each σ sample point:

[0047]

[0048] (3) Calculate the predicted observation at time k.

[0049]

[0050]

[0051]

[0052] (4) Calculate the measurement prediction covariance matrix P yy :

[0053]

[0054]

[0055]

[0056] Where R(k) is the observation noise covariance matrix at time k;

[0057] (5) Calculation and The cross-covariance matrix P between xy :

[0058]

[0059] (6) Calculate the Kalman gain:

[0060]

[0061] (7) Update augmentation status The covariance matrix P(k|k) of the augmented state error:

[0062]

[0063] P(k|k)=P(k|k-1)+K(k)P yy K(k) T.

[0064] Further, the step S34 comprises:

[0065]

[0066] wherein the weight d(k) = 0.9e L-k , L represents an adaptive window size, and E(k) is an approximation of the covariance matrix of the measurement residual:

[0067]

[0068] wherein y(i) is the actual measurement value at the i-th moment.

[0069] Further, the step S35 comprises:

[0070] a. Regenerate the sigma sample point χ (i) (k) of the k-th moment:

[0071]

[0072]

[0073]

[0074] b. Calculate the predicted augmented state of each sigma sample point

[0075] χ (i) (k+1|k) = F(χ (i) (k|k), u(k)), i = 0,..., 2n;

[0076] c. Predict the augmented state

[0077]

[0078]

[0079]

[0080] d. Predict the covariance matrix P(k+1|k) of the augmented state error:

[0081]

[0082]

[0083]

[0084] Compared with the prior art, the present application has the following beneficial effects:

[0085] This invention first extends the parameters of the unmanned surface vessel (USV) model to be identified into an augmented state. Then, it acquires the input signal data of the USV system in real time during the USV's periodic motion. Simultaneously, it uses real-time differential positioning to obtain the USV's real-time position and yaw angle, and an inertial measurement unit (IMU) to obtain the USV's real-time acceleration and yaw acceleration. Using these measurement data, an online parameter estimation strategy is designed. Unmodeled dynamics and unknown disturbances are treated as Gaussian white noise with unknown intensity, and the noise variance is adaptively estimated. This approach solves the state estimation problem of USVs under unmodeled dynamics and unknown noise interference, improving the accuracy of identifying unknown parameters of the USV. Attached Figure Description

[0086] Figure 1 A flowchart illustrating the online parameter estimation method for unmanned surface vessels based on adaptive unscented Kalman filtering provided in this embodiment of the invention;

[0087] Figure 2 The diagram shows the effect of estimating the inertial parameters of the unmanned surface vessel provided in the embodiment of the present invention.

[0088] Figure 3 The diagram shows the effect of estimating the resistance parameters of the unmanned vessel provided in the embodiment of the present invention.

[0089] Figure 4 A diagram illustrating the effect of estimating the thrust distribution parameters of an unmanned surface vessel provided in an embodiment of the present invention;

[0090] Figure 5 The diagram shows the effect of the parameters for estimating water flow velocity provided in the embodiments of the present invention. Detailed Implementation

[0091] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments:

[0092] like Figure 1 The method shown includes the following steps for online parameter identification of unmanned surface vessels based on adaptive unscented Kalman filtering:

[0093] S1. Construct the unmanned vessel model to be identified based on the unmanned vessel dynamics equations;

[0094] Specifically, the unmanned vessel model can be represented by the following two equations:

[0095]

[0096]

[0097] Where, η=[x w ,y w ,ψ] T x w ,y wrepresents the position of the center of mass of the unmanned ship, represents the yaw angle of the unmanned ship, R(ψ) represents a rotation matrix, V = [V x , y , 0] T represents the water flow velocity, which is assumed to be constant, M is an inertia matrix, C(v) is a Coriolis matrix, D is a damping matrix, B is a thrust distribution matrix, τ = [τ u , τ r ] T is an input quantity of the unmanned ship system, which is composed of forward thrust and rudder angle respectively;

[0098] The rotation matrix R(ψ) and the inertia matrix M are respectively represented as:

[0099]

[0100] The Coriolis matrix C(v), the damping matrix D and the thrust distribution matrix B are respectively represented as:

[0101]

[0102] The unmanned ship model can be further represented as:

[0103]

[0104] wherein a = [a1, a2,..., a8] T represents the parameters to be identified, which reduces the parameters to be identified, while not affecting the final purpose of parameter identification, that is, the design of the controller.

[0105] S2, taking the to-be-identified unmanned ship model parameters as augmented states, extending the to-be-identified unmanned ship model into an augmented model;

[0106] Specifically, the following augmented states are selected:

[0107]

[0108] wherein v c = [V x , V y ] T represents the water flow velocity, so the unmanned ship augmented model can be represented as:

[0109]

[0110] wherein:

[0111]

[0112] Considering the process noise w(k) and the measurement noise v(k) at the k th moment, the discretized unmanned ship augmented system is written as the following discrete form:

[0113] x(k+1)=F(x(k),τ(k))+w(k);

[0114] y(x)=h(x)+v(k);

[0115] wherein:

[0116] F(x(k),τ(k))=x(k)+(f0(x(k))+[f1(x(k))f2(x(k))]τ(k))Δt;

[0117] wherein, Δt is a sampling period.

[0118] S3, the augmented state is estimated online by adaptive unscented Kalman filtering to obtain the unmanned ship model parameters.

[0119] The step S3 comprises:

[0120] S31, initializing adaptive unscented Kalman filtering;

[0121] Specifically, given the initial predicted augmented state as close as possible to the actual initial state and parameters of the unmanned ship, the initial augmented state error covariance matrix P(0|-1) is given, which is generally taken as a unit matrix.

[0122] S32, obtaining the augmented model input and output data;

[0123] Specifically, the input signal data τ(k) of the unmanned ship system is obtained in real time in the periodic motion of the unmanned ship; at the same time, the real-time position x w ,y w and the yaw angle ψ of the unmanned ship are obtained by real-time differential positioning, the real-time acceleration and the yaw rate r of the unmanned ship are obtained by using the inertial measurement unit, and the position, yaw angle, acceleration and angular acceleration are taken as the measured values, i.e. the output data:

[0124]

[0125] At the same time, discretization is performed:

[0126] y(k)=h(x(k))。

[0127] S33, updating the augmented state and the augmented state error covariance matrix;

[0128] Specifically, the σ sample point χ (i) at the k th moment is generated:

[0129]

[0130]

[0131]

[0132] Where n represents the dimension of the state vector, P(k|k-1) and P(k|k-1) are the covariance matrices of the augmented state and the augmented state error predicted at time k using the observations at time k-1, respectively, where c = α 2 (n+κ) represents the coefficient. Let i represent the i-th column of the lower triangular decomposition of cP(k|k-1);

[0133] Calculate the measurement value for each σ sample point:

[0134]

[0135] Calculate the predicted observation at time k.

[0136]

[0137]

[0138]

[0139] Calculate the measurement prediction covariance matrix P yy :

[0140]

[0141]

[0142]

[0143] Where R(k) is the observation noise covariance matrix at time k;

[0144] calculate and The cross-covariance matrix P between xy :

[0145]

[0146] Calculate the Kalman gain:

[0147]

[0148] Update augmentation status The covariance matrix P(k|k) of the augmented state error:

[0149]

[0150] P(k|k) = P(k|k-1) + K(k)P yy K(k) T ;

[0151] The updated augmented state is the estimation of the state and the parameters at the kth moment.

[0152] S34, adaptively estimate the process noise covariance matrix;

[0153] Specifically, estimate the process noise covariance

[0154]

[0155] wherein the weight d(k) = 0.9e L-k , L represents the adaptive window size, and E(k) is an approximation of the covariance matrix of the measurement residual:

[0156]

[0157] wherein y(i) is the actual measurement value at the ith moment.

[0158] S35, predict the augmented state and the covariance matrix of the augmented state error;

[0159] Specifically, regenerate the kth moment σ sample point χ (i) :

[0160]

[0161]

[0162]

[0163] Calculate the predicted augmented state of each σ sample point

[0164] χ (i) (k+1|k) = F(χ (i) (k|k),u(k)), i = 0,...,2n;

[0165] Predict the augmented state

[0166]

[0167]

[0168]

[0169] Predicted augmented state The covariance matrix for updating the augmented state and the augmented state error next time;

[0170] The covariance matrix P(k+1|k) of the predicted augmented state error is:

[0171]

[0172]

[0173]

[0174] S36, determine whether to terminate online identification, if not, return to step S32.

[0175] In this embodiment, MATLAB 2020b is used as simulation software, and the adaptive unscented Kalman filter-based unmanned ship parameter online estimation method of the application is compared with the unscented Kalman filter method without adaptive link. The unmanned ship is periodically moved through the controller.

[0176] The parameters of the unmanned ship model used in the simulation are shown in Table 1:

[0177] Table 1: Unmanned ship model parameter settings

[0178]

[0179]

[0180] The measurement noise used in the simulation is white noise, and the covariance matrix is:

[0181] R = diag(10, 10, 7.6154, 1000, 1000, 30.462) x 10 -6 ;

[0182] The process noise is time-varying white noise with covariance, and the covariance matrix is:

[0183]

[0184] The initial process noise covariance estimation of adaptive unscented Kalman and unscented Kalman filter is:

[0185] Q0 = diag(5, 5, 5, 5, 5, 5, 0 1×10 ) x 10 -2 ;

[0186] The parameter settings of the unscented Kalman filter method for comparison are: α = 0.001, β = 2, κ = 0, Δt = 0.1, the same parameters in the adaptive unscented Kalman filter based online parameter estimation method of the unmanned ship are consistent with the above settings. In addition, L = 50, and the simulation duration is 30 seconds.

[0187] Figures 2 to 5 respectively represent the estimation effects of the inertia parameters, the damping parameters, the thrust distribution parameters and the water flow velocity. The black curve represents the parameter estimation results of the estimation strategy designed in the application, and the gray curve represents the estimation results obtained by using the unscented Kalman filter technology. It can be seen that the adaptive unscented Kalman filter based online parameter estimation strategy of the unmanned ship can achieve better estimation effect.

[0188] The above-mentioned is only an embodiment of the application, and the specific technical solutions and / or common knowledge of the scheme are not described in detail. It should be pointed out that for those skilled in the art, without departing from the technical scheme of the application, a number of modifications and improvements can be made, which should also be regarded as the protection scope of the application, and these will not affect the effect and practicality of the application. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode and the like recorded in the specification can be used to explain the content of the claims.

Claims

1. A method for online parameter identification of unmanned surface vessels based on adaptive unscented Kalman filtering, characterized in that, Includes the following steps: S1. Construct the unmanned vessel model to be identified based on the unmanned vessel dynamics equations; S2. Using the parameters of the unmanned vessel model to be identified as the augmented state, the unmanned vessel model to be identified is extended into an augmented model. S3. Adaptive unscented Kalman filtering is used to estimate the augmented state online and obtain the parameters of the unmanned vessel model; Step S2 includes: S21. Select augmentation state The augmented model of the unmanned vessel is represented as ; in, Indicates the speed of water flow; ； S22, Considering the first Time-of-flight noise and measuring noise The discretized augmented model of the unmanned surface vessel is expressed as: ； ； in: ; It is the sampling period; Step S3 includes: S31. Initialize the adaptive unscented Kalman filter; S32. Obtain the input and output data of the augmented model; S33. Update the covariance matrix of the augmented state and the augmented state error; S34. Adaptive estimation of process noise covariance matrix; S35, the covariance matrix of the predicted augmented state and the error of the augmented state; S36. Determine whether to terminate online identification. If not, return to step S32.

2. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 1, characterized in that, Step S1 includes: S11. Based on the unmanned vessel dynamics equations, the unmanned vessel model can be expressed as: ； ； in, , Indicates the position of the unmanned vessel's center of mass. Indicates the yaw angle of the unmanned vessel. Represents the rotation matrix. This represents the water flow velocity, assumed to be constant. It is the inertia matrix. It is the Coriolis force matrix. It is the damping matrix. It is the thrust distribution matrix. These are the inputs to the unmanned surface vessel system, consisting of forward thrust and rudder angle. Rotation matrix and inertia matrix They are represented as follows: ； ； Coriolis force matrix Damping matrix and thrust distribution matrix They are represented as follows: ； ； ； S12. The unmanned vessel model is further represented as follows: ； in, This represents the parameter to be identified.

3. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 1, characterized in that, Step S31 includes: giving an initial prediction augmentation state. The covariance matrix of the initial augmented state error .

4. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 3, characterized in that, Step S32 includes: the input data being the input signal data of the unmanned vessel system acquired in real time during the periodic motion of the unmanned vessel. The output data is the real-time position of the unmanned surface vessel obtained through real-time differential positioning. and yaw angle The inertial measurement unit obtains the real-time acceleration of the unmanned vessel. and yaw rate As a measured value, that is .

5. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 4, characterized in that, Step S33 includes: (1) Generate the first Moment Sample points : ； ； ； in, This represents the dimension of the state vector. and They are the first The time adopts the first The covariance matrix of the augmented state predicted by the observations at time t and the error of the augmented state. Represents the coefficient. express The lower triangular decomposition of the first List; (2) Calculate each Measurement values ​​of sample points: ； (3) Calculate the first Predicted observations at time : ； ； ； (4) Calculate the measurement prediction covariance matrix : ； ； ； in It is the first The noise covariance matrix of the observed values ​​at time t; (5) Calculation and cross-covariance matrix : ； (6) Calculate the Kalman gain: ； (7) Update augmentation status and the covariance matrix of augmented state error : ； 。 6. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 5, characterized in that, Step S34 includes: ； Among them, weight , Indicates adaptive window size. It is an approximation of the covariance matrix of the measurement residuals: ； in, It is the first The actual measured value at that moment.

7. The online parameter identification method for unmanned surface vessels based on adaptive unscented Kalman filtering according to claim 6, characterized in that, Step S35 includes: a. Regenerate the first Moment Sample points : ； ； ； b. Calculate each Predicted augmentation state of sample points : ； c. Predicting augmentation state : ； ； ； d. Covariance matrix of the predicted augmented state error : ； ； 。

Images