阅读说明：本技术 一种基于Elman神经网络的载体姿态估计方法 (Carrier attitude estimation method based on Elman neural network ) 是由 王守华 黄涛 孙希延 纪元法 江灿辉 陈昱均 付文涛 肖建明 于 2019-09-11 设计创作，主要内容包括：本发明公开了一种基于Elman神经网络的姿态估计方法,为了解决测姿系统中载波相位差分测姿易受干扰、惯性导航系统存在误差积累的问题,通过Elman神经网络,利用惯性导航系统的数据求出俯仰角、航向角和横滚角三个姿态角的校正值,并用此校正值去修正利用载波相位差分技术所求出的姿态角,测得所述修正值与所述未修正的姿态角相加即可得到精确的载体姿态。(The invention discloses an attitude estimation method based on an Elman neural network, which aims to solve the problems that carrier phase difference attitude measurement in an attitude measurement system is easy to interfere and an inertial navigation system has error accumulation.)

1. An attitude estimation method based on an Elman neural network is characterized by comprising the following steps:

initializing weights of all layers of the Elman neural network;

collecting historical sample data, wherein the historical sample data is used as an input value of the Elman neural network;

obtaining output values of each layer of the Elman neural network by the historical sample data through initial connection weights of each layer;

calculating an error square function according to the output value of the output layer and the expected value output by the output layer, and obtaining the final connection weight of each layer of the Elman neural network after the error square function reaches a set threshold value;

taking the measured data as the input of the Elman neural network which completes training to obtain the output correction value of the Elman neural network;

and adding the corrected value and the uncorrected attitude angle measured by the carrier phase differential technology to obtain the accurate attitude of the carrier.

2. An Elman neural network-based pose estimation method as claimed in claim 1,

the Elman neural network comprises an input layer, a hidden layer, a carrying layer and an output layer.

3. An Elman neural network-based pose estimation method as claimed in claim 1,

the Elman neural network takes historical sample data and measurement data as network input, and takes the correction value of the attitude angle as network output.

4. An Elman neural network-based pose estimation method as claimed in claim 1,

the historical sample data comprises an axis acceleration value, an axis angular velocity value and an attitude angle obtained by resolving a baseline vector through a carrier phase differential technology.

5. The method for estimating the attitude based on the Elman neural network as claimed in claim 1, wherein the step of acquiring the values of each layer by the Elman neural network comprises the following steps:

the product of the connection weight from the hidden layer to the output layer and the hidden layer data is a first preset value, the first preset value is subjected to first function processing to obtain a result value, the result value is equal to the output value of the output layer, the first preset value is x, and the first function is g (x) k.x + c;

the product of the connection weight from the input layer to the hidden layer and the input data in the previous iteration is a second preset value, the product of the connection weight from the receiving layer to the hidden layer and the receiving layer data is a third preset value, the sum of the second preset value and the third preset value is processed through a second function to obtain the output value of the hidden layer, and the second function is

6. An Elman neural network-based pose estimation method as claimed in claim 1,

before the square error function reaches a set threshold, the method further comprises the following steps: and the error square function is smaller than a threshold value, the connection weight values of all layers of the Elman neural network are updated, and the historical sample data is substituted into the updated connection weight values of all layers for iterative calculation until the error square function reaches the threshold value.

7. An Elman neural network-based pose estimation method as claimed in claim 1,

the measurement data refers to the uncorrected attitude angle of the carrier calculated by the carrier phase differential technology and the motion parameters of the carrier measured by the inertial navigation system.

8. An Elman neural network-based pose estimation method as claimed in claim 7,

the carrier motion data measured by the inertial navigation system includes the axial acceleration and the axial angular velocity of the carrier.

9. The method as claimed in claim 8, wherein the uncorrected attitude angle measured by the carrier-phase differential technique means that a baseline vector between antennas is calculated by the carrier-phase differential technique, and an uncorrected attitude angle of a carrier is calculated from the baseline vector.

Technical Field

The invention relates to the technical field of attitude measurement, in particular to a carrier attitude estimation method based on an Elman neural network.

Background

The method for measuring the attitude of the carrier is one of important technical applications of the GNSS, can provide attitude information of the carrier in real time, uses the three antennas to measure the attitude of the GNSS, and can measure the course angle, the roll angle and the pitch angle of the carrier. Errors generated by the carrier phase differential attitude measurement system cannot accumulate over time, but are limited by the GNSS system, and the precision of the carrier phase differential attitude measurement system is poor; the inertial navigation system has strong autonomy and is not interfered by the outside world when in work, but the inertial navigation system needs to give initial parameters before running and needs to correct errors in real time during the running process of the system.

In order to solve the problem of poor precision of the existing attitude measurement system, the invention provides an attitude estimation method based on an Elman neural network.

Disclosure of Invention

The invention aims to provide a carrier attitude estimation method, and aims to overcome the defects that the precision of a GNSS attitude measurement system is different from that of a precision inertial navigation system, the GNSS attitude measurement system is not suitable for a high-speed dynamic and satellite signal sheltered scene, initial parameters need to be given before the inertial navigation system runs, and errors need to be corrected in real time in the running process of the system.

The invention provides an attitude estimation method based on an Elman neural network,

initializing weights of all layers of the Elman neural network;

collecting historical sample data, wherein the historical sample data is used as an input value of the Elman neural network;

the historical sample data calculates output values of all layers of the Elman neural network through initial connection weights of all layers;

calculating an error square function according to the output value of the output layer and the expected value output by the output layer, and obtaining the final connection weight of each layer of the Elman neural network after the error square function reaches a set threshold value;

taking the measured data as the input of the Elman neural network which completes training to obtain the output correction value of the Elman neural network;

and adding the corrected value and the uncorrected attitude angle measured by the carrier phase differential technology to obtain the accurate attitude of the carrier.

Further, the Elman neural network comprises an input layer, a hidden layer, a receiving layer and an output layer.

Furthermore, the Elman neural network takes historical sample data as network input and takes the correction value of the attitude angle as network output.

Further, the historical sample data comprises an axis acceleration value, an axis angular velocity value and an attitude angle obtained by resolving a baseline vector through a carrier phase differential technology.

Further, the step of acquiring each layer of numerical values by the Elman neural network is as follows:

the product of the connection weight from the hidden layer to the output layer and the hidden layer data is a first preset value, the first preset value is subjected to first function processing to obtain a result value, the result value is equal to the output value of the output layer, the first preset value is x, and the first function is g (x) k.x + c;

the product of the connection weight from the input layer to the hidden layer and the input data in the previous iteration is a second preset value, the product of the connection weight from the receiving layer to the hidden layer and the receiving layer data is a third preset value, the sum of the second preset value and the third preset value is processed through a second function to obtain the output value of the hidden layer, and the second function is

And the output value of the accepting layer is equal to the output value of the hiding layer processed in the previous iteration.

Further, before the square error function reaches a set threshold, the method further includes: and the error square function is smaller than a threshold value, the connection weight values of all layers of the Elman neural network are updated, and the historical sample data is substituted into the updated connection weight values of all layers for iterative calculation until the error square function reaches the threshold value.

Further, the measurement data refers to the uncorrected attitude angle of the carrier determined by the carrier phase differential technique and the motion parameters of the carrier measured by the inertial navigation system.

Further, the carrier motion data measured by the inertial navigation system includes an axial acceleration and an axial angular velocity of the carrier.

Further, the uncorrected attitude angle measured by using the carrier phase differential technology means that a baseline vector between the antennas is solved by using the carrier phase differential technology, and the uncorrected attitude angle of the carrier is solved from the baseline vector.

The invention provides an Elman neural network-based attitude estimation method, which effectively inhibits error accumulation of an inertial navigation system by using the Elman neural network, assists carrier phase difference attitude measurement by using attitude measurement of the inertial navigation system, does not consider influence of external noise on the system, and enhances the stability of the system. The Elman neural network can obtain the optimal network connection weight value only by training and learning historical data, then can bring the weight value into an input value to directly solve the attitude angle correction value, and complete high-precision attitude calculation with less calculation complexity during calculation.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a flow chart of the present invention based on the posture estimation of the Elman neural network.

Detailed Description

Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.

In the description of the present invention, it is to be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on the orientations or positional relationships illustrated in the drawings, and are used merely for convenience in describing the present invention and for simplicity in description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed in a particular orientation, and be operated, and thus, are not to be construed as limiting the present invention. Further, in the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.

Referring to fig. 1, the present invention provides an attitude estimation method based on Elman neural network, including:

s101, initializing each layer weight w of Elman neural network₁、w₂、w₃；

S102, selecting measurement data collected in advance, such as triaxial acceleration, triaxial angular velocity and an attitude angle calculated by resolving a baseline vector through a carrier phase differential technology, as network input; the corrected values of the pitch angle, the course angle and the roll angle are used as network output;

s103, calculating the output x (t), x of the input layer of the Elman neural network by using the measurement data_c(t)、y(t)；

The input layer of the Elman neural network outputs x (t), x_c(t), y (t) the basic mathematical model is:

y(t)＝g(w₃x(t))

x(t)＝f(w₁x_c(t)+w₂(u(k-1)))

x_c(t)＝x(k-1)

in the formulas, y represents output layer data, x represents hidden layer data, u represents input data, and x_cIndicating the bearer data. w is a₃To be hiddenConnection weight, w, from hidden layer to output layer₂As a connection weight of the input layer to the hidden layer, w₁Is the connection weight from the socket layer to the hidden layer. The g () function computes the result of the output layer as a linear weighting of the hidden layer output data, and the f () function is the transfer function of the hidden layer.

S104, calculating an error square function E (t) by applying the input layer output, if the error square function does not reach the set threshold, calculating an updated value of each layer of connection weight, and calculating the measurement data again by using the updated value of each layer of connection weight until the error square function E (t) output and calculated by the input layer reaches the set threshold; and if the error square function reaches a set threshold value, the Elman neural network completes training to obtain a final network connection weight value.

S105 if the setting of the weight reaches the optimal state, the system will not adjust any more.

The S106Elman neural network parameter learning and adjusting method specifically comprises the following steps:

w(t+1)＝w(t)+Δw(t)

w is the connection weight between each layer of the Elman neural network, and Δ w (t) is the weight adjustment amount, which is calculated as follows:

the learning and adjusting process of each connection weight is deduced, and the method is not difficult to obtain:

connection weight w from hidden layer to output layer₃The adjustment amounts of (a) and (b) are:

Δw_3，ij(t)＝η(y_d，i(t)-y_i(t))·g_i′(·)·x_j(t)

connection weight w from input layer to hidden layer₂The adjustment amounts of (a) and (b) are:

connection weight w from the socket layer to the hidden layer₁The adjustment amounts of (a) and (b) are:

in the above formula, i represents the number of neurons in the output layer, j represents the number of neurons in the hidden layer, q represents the number of neurons in the input layer, and l represents the number of neurons in the carry layer. Eta is the learning rate of the neural network, and f () is taken

g () takes g (x) k · x + c. The optimal weight of each network connection weight can be obtained through training and learning by substituting each adjustment quantity into a formula.

Acquiring an observed value carrier measured value through a GNSS receiver, estimating a common-view satellite space coordinate according to ephemeris and pseudo-range observed values, and forming a double-difference observation equation:

wherein

The geometric distances from the main antenna and the slave antenna to the satellites i and j are double differences, and the geometric distances from the same observation satellite to the main antenna and the slave antenna can be approximately considered to be parallel because the satellites are far away from the ground and the main antenna and the slave antenna are generally close to each other.

Is the dual difference value of the carrier wave,

is double-difference observation noise, lambda is the electromagnetic wave wavelength,

is double difference integer ambiguity.

The relationship between the baseline length and the double-difference carrier observations is as follows:

when the main antenna and the slave antenna observe M satellites, M-1 mutually independent double-difference observed values can be generated in total, and therefore the formula can be formed:

in the formula

The unit vectors representing the main antenna and the auxiliary antenna to the co-view satellite are approximately equal because the distance between the satellite and the ground is far greater than the distance between the antennas,

is the vector of the master and slave antennas. In the formula, the whole-cycle ambiguity is solved by a relatively mature LAMBDA algorithm at present, and after the whole-cycle ambiguity is solved, a baseline vector between antennas can be solved.

At this time, taking the main antenna as the origin and the other two antennas as the slave antennas, two sets of base lines can be formed:

the subscript ANT denotes the baseline vector in the antenna coordinate system.

The calculation formula for converting the subscript ANT into a solution under the standing-center coordinate system is as follows:

the relationship between the center of gravity coordinate system and the carrier coordinate system is as follows:

wherein

To be rotational matrices that rotate around the X, Y, Z axes, respectively, they can be written as a matrix:

and obtaining a course angle and a pitch angle according to the relation between the station center coordinate system and the carrier coordinate system:

after course angle and pitch angle are determined, the coordinates of the secondary antenna center of gravity are determined

Can pass around the Z axis respectively

Length and X-axis theta length, assuming that the length of the base line formed by the main antenna and the auxiliary antenna is L₁₃Then its coordinates in the carrier coordinate system areAfter rotation, the coordinates of the rectangular coordinate system of the secondary antenna are

According to the coordinate rotation transformation relation, the following are provided:

the roll angle can thus be obtained:

therefore, the three attitude angles of the carrier can be obtained by utilizing the baseline vector solved by the carrier phase difference component technology and through the conversion relation among the three coordinate systems.

S107, the uncorrected attitude angle of the carrier obtained by the carrier phase differential technology and the motion parameter of the carrier measured by the inertial navigation system are used as the input of the Elman neural network which completes training, and the correction value output by the Elman neural network is obtained.

And S108, adding the corrected value and the uncorrected attitude angle to obtain the accurate attitude of the carrier.

Further, inputting the three uncorrected attitude angles and inertial navigation system data into a trained Elman neural network to obtain an attitude angle correction value, correcting the attitude angles by the attitude angle correction value, and obtaining a final result. The neural network is used for fusing the carrier phase difference attitude measurement system data and the inertial navigation system data, the specific form of external noise is not considered, the optimal weight value is obtained after training, the correction of the attitude angle can be completed only by bringing the input value into the network for simple calculation, the attitude estimation precision is improved, and the attitude estimation operand is reduced.

While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.