CN108613674A - A kind of attitude error suppressing method based on adaptive differential Evolutionary BP neural network - Google Patents

Abstract

The invention discloses a kind of attitude error suppressing methods based on adaptive differential Evolutionary BP neural network, belong to field of inertia technology.Including initializing inertia/star sensor integrated navigation system；Acquire the data of star sensor and the output of inertia component；Navigation calculation is carried out to inertia component output data；Posture of the naval vessel relative to inertial system is resolved using star sensor output data；Combining current time using ADE BPNN, interior integrated navigation posture predicts relative inertness system of carrier system attitude angle for the previous period；Star sensor measurement error is compensated using prediction posture and resolves naval vessel position using the star sensor output information after compensation；The navigation information of acquisition is combined navigation calculation；It stores and exports integrated navigation information.The present invention predicts star sensor posture by adaptive differential Evolutionary BP neural network, solves the problems, such as star sensor star chart " hangover " caused by ship sway, the applicability of inertia/star sensor integrated navigation system when enhancing ship sway.

Description

A Pose Error Suppression Method Based on Adaptive Differential Evolutionary BP Neural Network

technical field

The invention belongs to a method for reducing navigation information errors in the inertial technical field, and in particular relates to an attitude error suppression method based on an adaptive differential evolution BP neural network.

Background technique

Inertial/star sensor integrated navigation is a navigation system based on the inertial navigation system (Inertial Navigation System, INS), which can provide real-time velocity, attitude and position information for the carrier. INS mainly uses gyroscopes and accelerometers to measure carrier angular velocity and linear motion information. The navigation information is obtained after navigation calculation; the star sensor obtains the attitude information of the installation carrier relative to the inertial space by taking a star map and comparing it with its own star library. The inertial/star sensor integrated navigation system uses the navigation information calculated by the star sensor as a benchmark to periodically recalibrate the divergent positioning error of the INS. It does not require any external information and is a fully autonomous navigation system. The inertial/star sensor integrated navigation system has the advantages of strong real-time performance, high navigation accuracy, and full autonomy, and is widely used in ship navigation. However, due to the horizontal swing of the ship, the star image taken by the star sensor has a "smearing" phenomenon, which affects the calculation accuracy of the star sensor, and then affects the inertial/star sensor integrated navigation accuracy.

In the article "Missile Inertial/Satellite/Starlight High Precision Integrated Navigation" written by Yang Bo et al. in "Journal of Inertial Technology of China" in Volume 18, Issue 4, 2010, an online self-automatic method for gyro error and star sensor installation deviation angle is proposed. The calibration method suppresses the influence of device deviation on navigation accuracy; in the article "Inertial/Astronomical Combination Method for Aerospace Vehicles Directly Sensitive to Horizon" written by Zhang Cheng et al. in the third issue of "China Space Science and Technology" in June 2013, Proposes the use of star sensors and gyroscopes to construct inertial references, and on this basis, performs astronomical positioning calculations based on infrared horizons, and finally performs inertial/astronomical combined positioning, which suppresses the drift error of gyroscopes; the publication number is CN105737858A The patent of "A Method and Device for Calibrating Attitude Parameters of Airborne Inertial Navigation System" published on July 6, 2016, this method uses the attitude data output by the high-precision high-dynamic star sensor as a reference to complete the attitude parameters of the airborne INS In the article "In-flight Calibration of Gyrosand Star Sensor with Observability Analysis for SINS/CNS Integration" written by Yang Yanqiang et al. in "IEEE SENSORSJOURNAL", Volume 17, Issue 21, 2017, by analyzing the error based on singular value State observability, a four-bit decoupling calibration method is proposed, which suppresses the influence of device deviation on navigation accuracy. The above documents have proposed methods that can suppress device deviation, and most of them use missile-borne and airborne applications as backgrounds. , and did not conduct an in-depth study on the inaccurate information calculated by the star sensor caused by the "tailing" of the star map caused by the special application environment of ship swaying in the marine environment.

Contents of the invention

The purpose of the present invention is to provide a method for suppressing attitude errors based on adaptive differential evolution BP neural network that improves navigation accuracy and enhances the applicability of navigation information.

The purpose of the present invention is achieved through the following technical solutions:

Based on ADE-BPNN, the integrated navigation information of the inertial/star sensor integrated navigation system for a period of time is used as input to predict the current carrier attitude at time t, and the Kalman filter is used to predict the carrier attitude at time t and solve the star sensor at time t Data fusion is performed on the attitude of the carrier, and the navigation error of the star sensor is estimated and compensated, and the high-precision navigation information of the star sensor is obtained, which is combined with the INS navigation information for navigation solution.

A posture error suppression method based on adaptive differential evolution BP neural network, comprising the following steps:

Step 1: Power on and initialize the inertial/star sensor integrated navigation system.

Step 2: Collect the output data of the star sensor and inertial components.

Step 3: Carry out navigation calculation on the output data of the inertial component in step 2 to obtain the position of the carrier Carrier three-way speed Carrier three-axis attitude angle in, represent latitude and longitude, respectively; Respectively represent eastward, northward, skyward speed; attitude angle includes pitch angle, roll angle and heading angle, respectively by express.

Step 4: The attitude matrix relative to the inertial space is output by the star sensor in step 2 Get the attitude angle of the carrier relative to the inertial system which is

Among them, b represents the carrier coordinate system, i represents the inertial system, represents the transformation matrix from the carrier system to the inertial system; c _bimn (m,n=1,2,3) represents In the mth row, the nth column matrix element; arcsin is the arcsine function in the trigonometric function, and arctan is the arctangent function in the trigonometric function; Respectively represent the pitch angle, roll angle and heading angle of the carrier relative to the inertial system.

Step 5: Use ADE-BPNN to predict the attitude angle of the carrier relative to the inertial system at time t Among them, t represents the current moment, Respectively represent the predicted values of the carrier's pitch angle, roll angle and heading angle relative to the inertial system.

Step 6: Construct the Kalman filter 1, take the star sensor measurement error δφ _Ci (i=x, y, z) as the state quantity, the difference between the carrier attitude calculated by the star sensor in step 4 and the predicted carrier attitude in step 5 As an observation quantity, calculate δφ _Ci (i=x, y, z), and use the calculation result to compensate the calculation result of step 4 to obtain φ _Ci (i=x, y, z). Among them, δφ _Cx , δφ _Cy , δφ _Cz denote the pitch angle, roll angle and heading angle error of the star sensor respectively; φ _Cx , φ _Cy , φ _Cz denote the pitch angle, roll angle and Heading.

Step 7: Obtain the transformation matrix of the earth coordinate system relative to the quasi-geographical coordinate system according to the mathematical relationship between the transformation matrices The conversion process is Further calculation to obtain the position of the carrier, namely

in, Respectively represent latitude and longitude; e represents the earth system, N represents the quasi-geographical coordinate system, Indicates the conversion matrix from the earth system to the quasi-geographical coordinate system; c _eNmn (m,n=1,2,3) indicates In the mth row, nth column matrix element; Represents the transformation matrix from the carrier system to the quasi-geographical coordinate system; Indicates the transformation matrix from the carrier system to the inertial system; the subscript T indicates the matrix transposition; Represents the transformation matrix from the earth system to the inertial system.

Step 8: Construct a Kalman filter 2, and use the difference between the position calculated in step 7 and the position calculated by INS in step 3 λ _C -λ _S is the observation quantity, INS position error δλ _S , velocity error δv _Si (i=x,y,z), misalignment angle Φ _i (i=x,y,z), three-axis gyroscope constant error ε _i (i=x,y,z), The zero offset error ΔA _i (i=x, y, z) of the three-axis accelerometer is the state quantity, and the solution δλ _S , δv _Si (i=x,y,z), Φ _i (i=x,y,z), ε _i (i=x,y,z), ΔA _i (i=x,y,z) , compensate the solution result of step 3, and obtain the carrier position λ, velocity v _i (i=x, y, z), attitude angle φ _i (i=x, y, z). in, δλ _S represent the latitude and longitude errors respectively; δv _Sx , δv _Sy , δv _Sz represent the eastward, northward and celestial velocity errors respectively; Φ _x , Φ _y , Φ _z represent the pitch misalignment angle and roll misalignment angle respectively , heading misalignment angle.

Step 9: Store and output the position, velocity and attitude angle of the navigation information carrier obtained in step 8.

The beneficial effects of the present invention are:

After obtaining the integrated navigation information of the normal operation of the ship, combined with the ADE-BPNN algorithm, the output inertial attitude of the star sensor is predicted, and the influence of the "tailing" of the star map on the output attitude angle of the star sensor is eliminated, and the integrated navigation solution with INS is carried out. Calculate and output the navigation information. (1) The applicability of the inertial/star sensor integrated navigation system is enhanced when the ship is swaying horizontally, and the problem of insufficient error compensation caused by the ship swaying is reduced; (2) No external auxiliary information is required Improve positioning accuracy; (3) Small amount of calculation, simple and easy to implement.

Description of drawings

Fig. 1 is the combined navigation algorithm flowchart of the present invention;

Fig. 2 is the simulation test result utilizing the present invention to carry out;

Fig. 3 is the actual measurement experiment result that utilizes the present invention to carry out;

Figure 4 is the actual measurement of the experimental ship's trajectories.

Detailed ways

The specific embodiment of the present invention will be further described below in conjunction with accompanying drawing:

As shown in Figure 1, the present invention provides a kind of integrated navigation algorithm for inertial/star sensor, specifically comprises the following steps:

Step 1: Power on and initialize the inertial/star sensor integrated navigation system. To initialize the navigation, the system needs to be initialized:

(1) Initialize the initial value of INS: the carrier position of the hull λ ₀ (units are degrees), carrier three-way velocity v _i0 (i=x, y, z) (units are m/s), carrier three-axis attitude angle φ _i0 (i=x, y, z) ( are in degrees); initialize the transformation matrix Initialize the quaternion q ₀ ;

(2) Initial value of Kalman filter 1 parameter: initial value of state variable _{δφ Ci0} (i=x, y, z), mean square error matrix P _p0 , system noise square matrix Q _p , measurement noise square matrix R _p , Measuring array H _p ; The remaining initial values are set according to the actual situation.

(3) Initial value of Kalman filter 2 parameters: initial value of state variable δv _S0i (i=x,y,z), δλ _S0 , Φ _i0 (i=x,y,z), ε _i0 (i=x,y,z), ΔA _i0 (i=x,y,z), mean square error matrix P ₀ , system noise square matrix Q, measurement noise square matrix R, measurement matrix H;

The remaining initial values are set according to the actual situation.

The initialization transformation matrix is calculated as follows:

Among them, n represents the navigation system, Represents the transition matrix from the initial carrier system to the navigation system;

The initialization quaternion q ₀ is calculated as follows:

q ₀₀ =a, in q ₀ ＝[q ₀₀ q ₀₁ q ₀₂ q ₀₃ ] ^T , c _bnij (i,j=1,2,3) means In row i, column j matrix element;

During the navigation process, the initial information is used for updating to obtain the position, velocity and attitude angle of the carrier at any time.

Step 2: The system collects the star sensor output matrix in real time The three-axis acceleration output by the inertial component accelerometer and the three-axis angular velocity output by the gyroscope in, Respectively represent the components of the specific force measured by the accelerometer on the ox _b -axis, oy _b -axis, and oz _b -axis of the carrier system (the unit is m/s); Respectively represent the components of the angular velocity measured by the gyroscope on the ox _b -axis, oy _b -axis, and oz _b -axis of the carrier system (the unit is rad/s).

Step 3: Carry out navigation calculation on the data collected by the inertial component in step 2:

Quaternion pose matrix update:

Suppose the rotation quaternion of the carrier system relative to the navigation system at any time is q=[q ₀ q ₁ q ₂ q ₃ ] ^T , the timely correction of the quaternion is as follows:

in, respectively represent the rate of change of q _i (i=0,1,2,3);

Replace q _i (i=0,1,2,3) in formula (4) with the rotation quaternion q _i (t)(i=0,1,2,3) of the carrier system relative to the navigation system at time t, and get The rate of change of the rotation quaternion at time t The rotation quaternion of the carrier at time t+1 is

in In the formula, ω _i (i=x, y, z) omits the superscript b; T is the sampling time; When t=1, q(1) is the initial quaternion q ₀ of the carrier in step 1.

Update the strapdown matrix according to the element q _i (t+1) (i=0,1,2,3) in q(t+1)

Among them, q _i (i=0,1,2,3) in the above formula is q _i (t+1) (i=0,1,2,3) in formula (5), update attitude information:

Use conversion relations The projection of the specific force information measured by the accelerometer along the carrier system is converted to the navigation system, and the following differential equation is used to solve the carrier velocity:

in, express rate of change; Respectively represent the components of the specific force measured by the accelerometer on the ox _n- axis, oy _n- axis, and oz _n -axis of the navigation system (the unit is m/s);

Substitute f _i ⁿ (i=x, y, z) in formula (8) by comparing the projection f _i ⁿ (t) (i=x, y, z) in the navigation system at time t to obtain the carrier velocity change Rate Get the speed and position of the carrier at time t+1:

Among them, R represents the radius of the earth; When t=1, is the initial value v _i0 (i=x, y, z) of the carrier velocity in step 1.

So far, the carrier velocity, position and attitude angle calculated by INS are obtained according to formulas (7) and (9).

Step 4: The attitude matrix relative to the inertial space is output by the star sensor in step 2 Get the attitude angle of the carrier relative to the inertial system which is

Step 5: Use ADE-BPNN to predict the attitude angle of the carrier relative to the inertial system at time t The specific process is as follows:

(1) Initialization operation: initialization parameters, including population size N, gene dimension D, variation factor F, crossover rate CR, and the value range of each gene [U _min , U _max ], using the formula x _ij = U _min +rand×(U _max -U _min ) randomly initializes the population; where i=1,2,...,N, j=1,2,...,D, and rand represents a random number subject to uniform distribution.

(2) Judging whether the current minimum fitness value reaches μ or the number of iterations reaches the maximum, if so, the ADE iteration stops and goes to step (6); otherwise, proceed to the next step.

(3) Generate offspring subgroups: for each target vector Perform a mutation operation to generate a mutation vector pair mutation vector Perform crossover operation to generate experimental individuals target individuals Corresponding experiment individual Compete, compare the fitness values of the two, only if The fitness value of It is selected as the offspring when it is better, otherwise it will be directly as offspring. Repeat this step to generate offspring subpopulations. Among them, G represents the number of iterations.

(4) Evaluate the fitness value of the offspring subgroup, the minimum fitness value is the current optimal value, and the corresponding subgroup is the current optimal individual value.

(5) Let G=G+1, go to step (2).

(6) Assign the best individual of ADE as the initial connection weight and threshold of BPNN.

(7) Comparing the integrated navigation horizontal attitude information φ _i (i=x, y) at time t with the threshold θ _i (i=x, y), when the horizontal attitude information is less than the threshold, it is determined that the ship is swaying, using tn～ Combining navigation attitude information at time t-1 to train BPNN to obtain the optimal model.

(8) Using the model in (7), predict the attitude angle φ _pi (i=x, y, z) of the carrier relative to the navigation system at time t, and use the predicted attitude angle to obtain the transformation matrix from the carrier system to the navigation system According to the relationship between the transformation matrices, the transformation matrix of the carrier system relative to the inertial system is obtained The conversion process is Further calculation is performed to obtain the predicted value of the attitude angle of the carrier relative to the inertial system, namely:

in, express The element in row m and column n; Indicates the transformation matrix from the earth system to the inertial system, which is obtained from the earth speed and navigation time; Represents the conversion matrix from the earth system to the navigation system, and the navigation position is combined at time t-1 λ is obtained.

Step 6: Construct the Kalman filter 1, take the star sensor measurement error δφ _Ci (i=x, y, z) as the state quantity, the difference between the carrier attitude calculated by the star sensor in step 4 and the predicted carrier attitude in step 5 The specific filtering process is as follows:

Using the inertial attitude error model of the star sensor, combined with the Kalman filter, the error correction for the inertial attitude is as follows:

Among them, P _p (t) represents the estimated mean square error at time t, P _p (t/t-1) represents the mean square error of one-step prediction from time t-1 to time t, and F _p (t/t-1) represents the mean square error of t- The state transition matrix from time 1 to time t, G _p (t-1) represents the noise driving matrix at time t-1, Q _p (t-1) represents the system noise variance matrix at time t-1, and H _p represents the measurement matrix , R _p represents the measurement noise variance matrix, and K _p (t) represents the filter gain matrix at time t.

The update matrix of state estimation and estimated mean square error at time t is:

Among them, X _p (t) represents the state estimator at time t, X _p (t) = [δφ _Cx δφ _Cy δφ _Cz ] ^T , when t=1, the element in the state quantity X _p (1) is Karl in step 1 The initial value of state quantity _{δφ Ci0} (i=x, y, z) of Mann filter 1. According to the calculation result, the calculation result of step 4 is compensated to obtain the inertial attitude angle φ _Ci (i=x, y, z).

Step 7: Obtain the transformation matrix of the earth coordinate system relative to the quasi-geographical coordinate system according to the mathematical relationship between the transformation matrices The conversion process is Further calculation to obtain the position of the carrier, namely

Transformation matrix from carrier system to quasi-geographical system From the horizontal attitude calculated in step 3 Get, the transformation matrix from the carrier system to the inertial system Obtained from the inertial attitude angle calculated in step 6.

Step 8: Construct a Kalman filter 2, and use the difference between the position calculated in step 7 and the position calculated by INS in step 3 λ _C -λ _S is the observation quantity, INS position error δλ _S , velocity error δv _Si (i=x,y,z), misalignment angle Φ _i (i=x,y,z), three-axis gyroscope constant error ε _i (i=x,y,z), The zero offset error ΔA _i (i=x, y, z) of the three-axis accelerometer is a state quantity, and the specific filtering process is as follows:

Using the speed, position and misalignment angle error model of the carrier, combined with the Kalman filter, the error correction of the carrier state quantity is as follows:

Among them, P(t) represents the estimated mean square error at time t, P(t/t-1) represents the mean square error of one-step prediction from time t-1 to time t, and F(t/t-1) represents the estimated mean square error from time t-1 to time t. The state transition matrix at time t, G(t-1) represents the noise driving array at time t-1, Q(t-1) represents the system noise variance matrix at time t-1, H represents the measurement matrix, and R represents the measurement noise Variance matrix, K(t) represents the filter gain matrix at time t.

The update matrix of state estimation and estimated mean square error at time t is:

Wherein, X (t) represents the state estimation quantity at time t, when t=1, the element in the state quantity X (1) is the state quantity initial value δv _S0i (i=x, y, z), δλ _S0 , Φ _i0 (i=x,y,z), ε _i0 (i=x,y,z), ΔA _i0 (i=x,y,z). According to the calculation result, the INS calculation result in step 3 is compensated to obtain the position of the navigation information carrier λ, velocity v _i (i=x, y, z), attitude angle φ _i (i=x, y, z).

Step 9: Store and output the position, velocity and attitude angle of the navigation information carrier obtained in step 8.

Example:

Beneficial effect of the present invention is verified by guideline and measured experiment:

Simulation:

Vessel sailing route: linear motion

The system initialization parameters are as follows:

Ship motion parameters: ship speed: v _x0 = 0m/s, v _y0 = 0m/s, v _z0 = 0m/s

Ship attitude:

Effect of sway on ship speed:

Uniformly distributed in [0,2π]

Effect of high frequency vibration on ship speed:

Uniformly distributed in [0,2π]

Local latitude and longitude: λ ₀ =126.6705°

Gyro parameters: Gyro constant drift: 0.01°/h

Gyro scale factor error: 1×10 ^-5

Gyro white noise error: 0.001°/h

Accelerometer parameters: Accelerometer bias: 10 ^-4 g

Accelerometer scale factor error: 1×10 ^-5

Accelerometer white noise error: 10 ^-5 g

ADE-BPNN parameters:

The best fit network has 20 input neurons, 10 hidden neurons and 1 output neuron

Maximum number of iterations: 100

Training target error: 0.00004

Learning rate: 0.1

Hidden layer function: logsig

Activation layer function: purelin

Training function: trainlm

Connection weight and threshold range: [-1,1]

Population size: N=50

Maximum number of iterations: GenM=50

Accuracy requirement: μ＝0.15

Cross rate: CR＝0.1

Minimum mutation factor: F _min =0.2

Maximum mutation factor: F _max ＝0.6

Kalman filter 1 value setting:

P _p0 =diag([0.01 0.01 0.01] ² )

Q _p ＝diag([0.01 0.01 0.01] ² )

R _p =diag([0.001 0.001 0.001] ² )

Kalman filter 2 value setting:

P ₀ ＝diag([10 10 0.0005 0.0005 0.05 0.05 0.15 0.01 0.01 0.01 10 ^-5 g 10 ^-5 g10 ^-5 g] ² )

Q＝diag([0.5×10 ^-5 g 0.5×10 ^-5 g O _1×2 0.5×0.01 0.5×0.01 0.5×0.01 O _1×6 ] ² )

R＝diag([100R 100R] ² )

Sampling frequency: 0.1s

Sampling time: 1h

Measured experiment:

Ship navigation route: as shown in Figure 4.

Gyro parameters: Dynamic range: ±100°/s

Zero deviation: ≤0.01°/h

Non-linearity of scale factor: ≤20ppm

Star sensor parameters: Field of view: 24°

Interstellar observation level: not less than level 7

Data update frequency: 20Hz

Attitude accuracy: 5″

By using the method described in the invention, the error curve of the integrated navigation information of the hull inertial/star sensor with and without the ADE-BPNN algorithm is obtained under the condition of wind and waves. Figure 2 shows the comparison curve of the simulation results, and Figure 3 shows the comparison curve of the measured experimental results. The results show that the present invention reduces the influence of the "tailing" of the star map on the navigation information calculated by the star sensor when the ship is swaying horizontally, and further suppresses the combined navigation error of the inertial/star sensor better, which can meet the actual needs.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (8)

1. a posture error suppression method based on adaptive differential evolution BP neural network, is characterized in that, comprises the following steps: (1) Power on, initialize the inertial/star sensor integrated navigation system; (2) The system collects the star sensor output matrix in real time The three-axis acceleration output by the inertial component accelerometer and the three-axis angular velocity output by the gyroscope in, Respectively represent the components of the specific force measured by the accelerometer on the ox _b -axis, oy _b -axis, and oz _b -axis of the carrier system, and the unit is m/s; Respectively represent the components of the angular velocity measured by the gyroscope on the ox _b -axis, oy _b -axis, and oz _b -axis of the carrier system, and the unit is rad/s; (3) Carry out navigation solution to the data collected by inertial components in step (2); (4) The attitude matrix relative to the inertial space is output by the star sensor in step (2) Get the attitude angle of the carrier relative to the inertial system (5) Use ADE-BPNN to predict the attitude angle of the carrier relative to the inertial system at time t Among them, t represents the current moment, respectively represent the predicted values of the carrier’s pitch angle, roll angle and heading angle relative to the inertial system; (6) Construct the Kalman filter 1, take the star sensor measurement error δφ _Ci (i=x, y, z) as the state quantity, the star sensor in step (4) calculates the attitude of the carrier and the prediction in step (5) carrier attitude difference is the observation quantity; according to the solution result, the step (4) solution result is compensated to obtain the inertia attitude angle φ _Ci (i=x, y, z); (7) According to the mathematical relationship between the transformation matrices, the transformation matrix of the earth coordinate system relative to the geographic coordinate system is obtained The conversion process is Further calculation to obtain the position of the carrier and the transformation matrix from the carrier system to the quasi-geographical system From the horizontal attitude calculated in step (3) Get, the transformation matrix from the carrier system to the inertial system Obtained by the solution result inertial attitude angle in the step (6); (8) Construct a Kalman filter 2, and use the difference between the position calculated in step (7) and the position calculated by INS in step (3) λ _C -λ _S is the observation quantity, INS position error Velocity error δv _Si (i=x,y,z), misalignment angle Φ _i (i=x,y,z), three-axis gyro constant error ε _i (i=x,y,z), three-axis acceleration Calculate the zero position offset error ΔA _i (i=x, y, z) as the state quantity, according to the calculation result, compensate the INS calculation result in step (3), and obtain the position of the navigation information carrier λ, velocity v _i (i=x,y,z), attitude angle φ _i (i=x,y,z); (9) Store and output the position, velocity and attitude angle of the navigation information carrier obtained in step (8). 2. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (1) specifically comprises: (1.1) Initialize the initial value of INS: the carrier position of the hull λ ₀ (units are degrees), carrier three-way velocity v _i0 (i=x, y, z) (units are m/s), carrier three-axis attitude angle φ _i0 (i=x, y, z) ( are in degrees); initialize the transformation matrix Initialize the quaternion q ₀ ; (1.2) Initial value of Kalman filter 1 parameter: initial value of state variable _{δφ Ci0} (i=x, y, z), mean square error matrix P _p0 , system noise square matrix Q _p , measurement noise square matrix R _p , Measuring array H _p ; The remaining initial values are set according to the actual situation. (1.3) Initial value of Kalman filter 2 parameters: initial value of state variable δv _S0i (i=x,y,z), δλ _S0 , Φ _i0 (i=x,y,z), ε _i0 (i=x,y,z), ΔA _i0 (i=x,y,z), mean square error matrix P ₀ , system noise square matrix Q, measurement noise square matrix R, measurement matrix H; The remaining initial values are set according to the actual situation. The initialization transformation matrix is calculated as follows: Among them, n represents the navigation system, Represents the transition matrix from the initial carrier system to the navigation system; The initialization quaternion q ₀ is calculated as follows: q ₀₀ =a, in q ₀ ＝[q ₀₀ q ₀₁ q ₀₂ q ₀₃ ] ^T , c _bnij (i,j=1,2,3) means In row i, column j matrix element; During the navigation process, the initial information is used for updating to obtain the position, velocity and attitude angle of the carrier at any time. 3. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (3) specifically comprises: (3.1) Quaternion attitude matrix update: Suppose the rotation quaternion of the carrier system relative to the navigation system at any time is q=[q ₀ q ₁ q ₂ q ₃ ] ^T , the timely correction of the quaternion is as follows: in, respectively represent the rate of change of q _i (i=0,1,2,3); (3.2) Replace q _i (i=0,1,2,3) in formula (4) with the rotation quaternion q _i (t)(i=0,1,2,3) of the carrier system relative to the navigation system at time t ), to get the rate of change of the rotation quaternion at time t The rotation quaternion of the carrier at time t+1 is in In the formula, ω _i (i=x, y, z) omits the superscript b; T is the sampling time; When t=1, q(1) is the initial quaternion q ₀ of the carrier in step 1. (3.3) Update the strapdown matrix according to the element q _i (t+1) (i=0,1,2,3) in q(t+1) Among them, q _i (i=0,1,2,3) in the above formula is q _i (t+1) (i=0,1,2,3) in formula (5), update attitude information: (3.4) Using conversion relation The projection of the specific force information measured by the accelerometer along the carrier system is converted to the navigation system, and the following differential equation is used to solve the carrier velocity: in, express rate of change; Respectively represent the components of the specific force measured by the accelerometer on the ox _n- axis, oy _n- axis, and oz _n -axis of the navigation system (the unit is m/s); (3.5) Replace the f _i ⁿ (i=x, y, z) in formula (8) with the projection f _i ⁿ (t) (i=x, y, z) of the comparison force in the navigation system at time t, and get Carrier velocity change rate Get the speed and position of the carrier at time t+1: Among them, R represents the radius of the earth; When t=1, is the initial value v _i0 of the carrier velocity in step 1 (i=x, y, z); So far, the carrier velocity, position and attitude angle calculated by INS are obtained according to equations (5) and (7). 4. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (4) specifically comprises: The attitude angle of the carrier relative to the inertial system for: 5. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (6) specifically comprises: (6.1) Using the inertial attitude error model of the star sensor, combined with the Kalman filter, the error correction of the inertial attitude is as follows: Among them, P _p (t) represents the estimated mean square error at time t, P _p (t/t-1) represents the mean square error of one-step prediction from time t-1 to time t, and F _p (t/t-1) represents the mean square error of t- The state transition matrix from time 1 to time t, G _p (t-1) represents the noise driving matrix at time t-1, Q _p (t-1) represents the system noise variance matrix at time t-1, and H _p represents the measurement matrix , R _p represents the measurement noise variance matrix, K _p (t) represents the filter gain matrix at time t; (6.2) The update matrix of state estimation and estimated mean square error at time t is: Among them, X _p (t) represents the state estimator at time t, X _p (t) = [δφ _Cx δφ _Cy δφ _Cz ] ^T , when t=1, the element in the state quantity X _p (1) is Karl in step 1 The initial value of the state quantity _{δφ Ci0} (i=x, y, z) of Mann filter 1, according to the calculation result, the calculation result of step (4) is compensated, and the inertial attitude angle φ _Ci (i=x, y, z). 6. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (7) specifically comprises: According to the mathematical relationship between the transformation matrices, the transformation matrix of the earth coordinate system relative to the quasi-geographical coordinate system is obtained The conversion process is Further calculation to obtain the position of the carrier, namely Transformation matrix from carrier system to quasi-geographical system From the horizontal attitude calculated in step (3) Get, the transformation matrix from the carrier system to the inertial system Obtained from the inertial attitude angle of the solution result in step (6). 7. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (8) specifically comprises: Using the speed, position and misalignment angle error model of the carrier, combined with the Kalman filter, the error correction of the carrier state quantity is as follows: Among them, P(t) represents the estimated mean square error at time t, P(t/t-1) represents the mean square error of one-step prediction from time t-1 to time t, and F(t/t-1) represents the estimated mean square error from time t-1 to time t. The state transition matrix at time t, G(t-1) represents the noise driving array at time t-1, Q(t-1) represents the system noise variance matrix at time t-1, H represents the measurement matrix, and R represents the measurement noise Variance matrix, K(t) represents the filter gain matrix at time t; The update matrix of state estimation and estimated mean square error at time t is: Wherein, X (t) represents the state estimation quantity at time t, when t=1, the element in the state quantity X (1) is the state quantity initial value δv _S0i (i=x, y, z), δλ _S0 , Φ _i0 (i=x,y,z), ε _i0 (i=x,y,z), ΔA _i0 (i=x,y,z). According to the calculation result, the INS calculation result in step 3 is compensated to obtain the position of the navigation information carrier λ, velocity v _i (i=x, y, z), attitude angle φ _i (i=x, y, z). 8. a kind of attitude error suppression method based on adaptive differential evolution BP neural network according to claim 1, is characterized in that, described step (5) specifically comprises: (5.1) Initialization operation: initialization parameters, including population size N, gene dimension D, variation factor F, crossover rate CR, and the value range of each gene [U _min , U _max ], using the formula x _ij = U _min +rand×(U _max -U _min ) randomly initialize the population; where i=1,2,...,N, j=1,2,...,D, rand represents a random number that obeys the uniform distribution; (5.2) Judging whether the current minimum fitness value reaches μ or the number of iterations reaches the maximum, if so, the ADE iteration stops and goes to step (5.6); otherwise, proceed to the next step; (5.3) Generate offspring subgroups: for each target vector Perform a mutation operation to generate a mutation vector pair mutation vector Perform crossover operation to generate experimental individuals target individuals Corresponding experiment individual Compete, compare the fitness values of the two, only if The fitness value of It is selected as the offspring when it is better, otherwise it will be directly as offspring. Repeat this step to generate offspring subgroups, where G represents the number of iterations; (5.4) Evaluate the fitness value of the offspring subgroup, the minimum fitness value is the current optimal value, and the corresponding subgroup is the current optimal individual value; (5.5) Make G=G+1, go to step (5.2); (5.6) assign the best individual of ADE as the initial connection weight and threshold of BPNN; (5.7) Comparing the combined navigation horizontal attitude information φ _i (i=x, y) at time t with the threshold θ _i (i=x, y), when the horizontal attitude information is less than the threshold, it is determined that the ship is swaying, using tn～ Combining navigation attitude information at time t-1 to train BPNN to obtain the optimal model; (5.8) Use the model obtained in step (5.7) to predict the attitude angle φ _pi (i=x, y, z) of the carrier relative to the navigation system at time t, and use the predicted attitude angle to obtain the transformation matrix from the carrier system to the navigation system According to the relationship between the transformation matrices, the transformation matrix of the carrier system relative to the inertial system is obtained The conversion process is Further calculation is performed to obtain the predicted value of the attitude angle of the carrier relative to the inertial system, namely: in, express The element in row m and column n; Indicates the transformation matrix from the earth system to the inertial system, which is obtained from the earth speed and navigation time; Represents the conversion matrix from the earth system to the navigation system, and the navigation position is combined at time t-1 λ is obtained.